builtin-apply.con commit builtin-apply.c: push match-beginning/end logic down (dc41976)
   1/*
   2 * apply.c
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This applies patches on top of some (arbitrary) version of the SCM.
   7 *
   8 */
   9#include "cache.h"
  10#include "cache-tree.h"
  11#include "quote.h"
  12#include "blob.h"
  13#include "delta.h"
  14#include "builtin.h"
  15
  16/*
  17 *  --check turns on checking that the working tree matches the
  18 *    files that are being modified, but doesn't apply the patch
  19 *  --stat does just a diffstat, and doesn't actually apply
  20 *  --numstat does numeric diffstat, and doesn't actually apply
  21 *  --index-info shows the old and new index info for paths if available.
  22 *  --index updates the cache as well.
  23 *  --cached updates only the cache without ever touching the working tree.
  24 */
  25static const char *prefix;
  26static int prefix_length = -1;
  27static int newfd = -1;
  28
  29static int unidiff_zero;
  30static int p_value = 1;
  31static int p_value_known;
  32static int check_index;
  33static int update_index;
  34static int cached;
  35static int diffstat;
  36static int numstat;
  37static int summary;
  38static int check;
  39static int apply = 1;
  40static int apply_in_reverse;
  41static int apply_with_reject;
  42static int apply_verbosely;
  43static int no_add;
  44static const char *fake_ancestor;
  45static int line_termination = '\n';
  46static unsigned long p_context = ULONG_MAX;
  47static 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|fix|error|error-all>] <patch>...";
  49
  50static enum ws_error_action {
  51        nowarn_ws_error,
  52        warn_on_ws_error,
  53        die_on_ws_error,
  54        correct_ws_error,
  55} ws_error_action = warn_on_ws_error;
  56static int whitespace_error;
  57static int squelch_whitespace_errors = 5;
  58static int applied_after_fixing_ws;
  59static const char *patch_input_file;
  60
  61static void parse_whitespace_option(const char *option)
  62{
  63        if (!option) {
  64                ws_error_action = warn_on_ws_error;
  65                return;
  66        }
  67        if (!strcmp(option, "warn")) {
  68                ws_error_action = warn_on_ws_error;
  69                return;
  70        }
  71        if (!strcmp(option, "nowarn")) {
  72                ws_error_action = nowarn_ws_error;
  73                return;
  74        }
  75        if (!strcmp(option, "error")) {
  76                ws_error_action = die_on_ws_error;
  77                return;
  78        }
  79        if (!strcmp(option, "error-all")) {
  80                ws_error_action = die_on_ws_error;
  81                squelch_whitespace_errors = 0;
  82                return;
  83        }
  84        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
  85                ws_error_action = correct_ws_error;
  86                return;
  87        }
  88        die("unrecognized whitespace option '%s'", option);
  89}
  90
  91static void set_default_whitespace_mode(const char *whitespace_option)
  92{
  93        if (!whitespace_option && !apply_default_whitespace)
  94                ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
  95}
  96
  97/*
  98 * For "diff-stat" like behaviour, we keep track of the biggest change
  99 * we've seen, and the longest filename. That allows us to do simple
 100 * scaling.
 101 */
 102static int max_change, max_len;
 103
 104/*
 105 * Various "current state", notably line numbers and what
 106 * file (and how) we're patching right now.. The "is_xxxx"
 107 * things are flags, where -1 means "don't know yet".
 108 */
 109static int linenr = 1;
 110
 111/*
 112 * This represents one "hunk" from a patch, starting with
 113 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 114 * patch text is pointed at by patch, and its byte length
 115 * is stored in size.  leading and trailing are the number
 116 * of context lines.
 117 */
 118struct fragment {
 119        unsigned long leading, trailing;
 120        unsigned long oldpos, oldlines;
 121        unsigned long newpos, newlines;
 122        const char *patch;
 123        int size;
 124        int rejected;
 125        struct fragment *next;
 126};
 127
 128/*
 129 * When dealing with a binary patch, we reuse "leading" field
 130 * to store the type of the binary hunk, either deflated "delta"
 131 * or deflated "literal".
 132 */
 133#define binary_patch_method leading
 134#define BINARY_DELTA_DEFLATED   1
 135#define BINARY_LITERAL_DEFLATED 2
 136
 137/*
 138 * This represents a "patch" to a file, both metainfo changes
 139 * such as creation/deletion, filemode and content changes represented
 140 * as a series of fragments.
 141 */
 142struct patch {
 143        char *new_name, *old_name, *def_name;
 144        unsigned int old_mode, new_mode;
 145        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 146        int rejected;
 147        unsigned ws_rule;
 148        unsigned long deflate_origlen;
 149        int lines_added, lines_deleted;
 150        int score;
 151        unsigned int is_toplevel_relative:1;
 152        unsigned int inaccurate_eof:1;
 153        unsigned int is_binary:1;
 154        unsigned int is_copy:1;
 155        unsigned int is_rename:1;
 156        struct fragment *fragments;
 157        char *result;
 158        size_t resultsize;
 159        char old_sha1_prefix[41];
 160        char new_sha1_prefix[41];
 161        struct patch *next;
 162};
 163
 164static void say_patch_name(FILE *output, const char *pre,
 165                           struct patch *patch, const char *post)
 166{
 167        fputs(pre, output);
 168        if (patch->old_name && patch->new_name &&
 169            strcmp(patch->old_name, patch->new_name)) {
 170                quote_c_style(patch->old_name, NULL, output, 0);
 171                fputs(" => ", output);
 172                quote_c_style(patch->new_name, NULL, output, 0);
 173        } else {
 174                const char *n = patch->new_name;
 175                if (!n)
 176                        n = patch->old_name;
 177                quote_c_style(n, NULL, output, 0);
 178        }
 179        fputs(post, output);
 180}
 181
 182#define CHUNKSIZE (8192)
 183#define SLOP (16)
 184
 185static void read_patch_file(struct strbuf *sb, int fd)
 186{
 187        if (strbuf_read(sb, fd, 0) < 0)
 188                die("git-apply: read returned %s", strerror(errno));
 189
 190        /*
 191         * Make sure that we have some slop in the buffer
 192         * so that we can do speculative "memcmp" etc, and
 193         * see to it that it is NUL-filled.
 194         */
 195        strbuf_grow(sb, SLOP);
 196        memset(sb->buf + sb->len, 0, SLOP);
 197}
 198
 199static unsigned long linelen(const char *buffer, unsigned long size)
 200{
 201        unsigned long len = 0;
 202        while (size--) {
 203                len++;
 204                if (*buffer++ == '\n')
 205                        break;
 206        }
 207        return len;
 208}
 209
 210static int is_dev_null(const char *str)
 211{
 212        return !memcmp("/dev/null", str, 9) && isspace(str[9]);
 213}
 214
 215#define TERM_SPACE      1
 216#define TERM_TAB        2
 217
 218static int name_terminate(const char *name, int namelen, int c, int terminate)
 219{
 220        if (c == ' ' && !(terminate & TERM_SPACE))
 221                return 0;
 222        if (c == '\t' && !(terminate & TERM_TAB))
 223                return 0;
 224
 225        return 1;
 226}
 227
 228static char *find_name(const char *line, char *def, int p_value, int terminate)
 229{
 230        int len;
 231        const char *start = line;
 232
 233        if (*line == '"') {
 234                struct strbuf name;
 235
 236                /*
 237                 * Proposed "new-style" GNU patch/diff format; see
 238                 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
 239                 */
 240                strbuf_init(&name, 0);
 241                if (!unquote_c_style(&name, line, NULL)) {
 242                        char *cp;
 243
 244                        for (cp = name.buf; p_value; p_value--) {
 245                                cp = strchr(cp, '/');
 246                                if (!cp)
 247                                        break;
 248                                cp++;
 249                        }
 250                        if (cp) {
 251                                /* name can later be freed, so we need
 252                                 * to memmove, not just return cp
 253                                 */
 254                                strbuf_remove(&name, 0, cp - name.buf);
 255                                free(def);
 256                                return strbuf_detach(&name, NULL);
 257                        }
 258                }
 259                strbuf_release(&name);
 260        }
 261
 262        for (;;) {
 263                char c = *line;
 264
 265                if (isspace(c)) {
 266                        if (c == '\n')
 267                                break;
 268                        if (name_terminate(start, line-start, c, terminate))
 269                                break;
 270                }
 271                line++;
 272                if (c == '/' && !--p_value)
 273                        start = line;
 274        }
 275        if (!start)
 276                return def;
 277        len = line - start;
 278        if (!len)
 279                return def;
 280
 281        /*
 282         * Generally we prefer the shorter name, especially
 283         * if the other one is just a variation of that with
 284         * something else tacked on to the end (ie "file.orig"
 285         * or "file~").
 286         */
 287        if (def) {
 288                int deflen = strlen(def);
 289                if (deflen < len && !strncmp(start, def, deflen))
 290                        return def;
 291                free(def);
 292        }
 293
 294        return xmemdupz(start, len);
 295}
 296
 297static int count_slashes(const char *cp)
 298{
 299        int cnt = 0;
 300        char ch;
 301
 302        while ((ch = *cp++))
 303                if (ch == '/')
 304                        cnt++;
 305        return cnt;
 306}
 307
 308/*
 309 * Given the string after "--- " or "+++ ", guess the appropriate
 310 * p_value for the given patch.
 311 */
 312static int guess_p_value(const char *nameline)
 313{
 314        char *name, *cp;
 315        int val = -1;
 316
 317        if (is_dev_null(nameline))
 318                return -1;
 319        name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
 320        if (!name)
 321                return -1;
 322        cp = strchr(name, '/');
 323        if (!cp)
 324                val = 0;
 325        else if (prefix) {
 326                /*
 327                 * Does it begin with "a/$our-prefix" and such?  Then this is
 328                 * very likely to apply to our directory.
 329                 */
 330                if (!strncmp(name, prefix, prefix_length))
 331                        val = count_slashes(prefix);
 332                else {
 333                        cp++;
 334                        if (!strncmp(cp, prefix, prefix_length))
 335                                val = count_slashes(prefix) + 1;
 336                }
 337        }
 338        free(name);
 339        return val;
 340}
 341
 342/*
 343 * Get the name etc info from the --/+++ lines of a traditional patch header
 344 *
 345 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 346 * files, we can happily check the index for a match, but for creating a
 347 * new file we should try to match whatever "patch" does. I have no idea.
 348 */
 349static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 350{
 351        char *name;
 352
 353        first += 4;     /* skip "--- " */
 354        second += 4;    /* skip "+++ " */
 355        if (!p_value_known) {
 356                int p, q;
 357                p = guess_p_value(first);
 358                q = guess_p_value(second);
 359                if (p < 0) p = q;
 360                if (0 <= p && p == q) {
 361                        p_value = p;
 362                        p_value_known = 1;
 363                }
 364        }
 365        if (is_dev_null(first)) {
 366                patch->is_new = 1;
 367                patch->is_delete = 0;
 368                name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
 369                patch->new_name = name;
 370        } else if (is_dev_null(second)) {
 371                patch->is_new = 0;
 372                patch->is_delete = 1;
 373                name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 374                patch->old_name = name;
 375        } else {
 376                name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 377                name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
 378                patch->old_name = patch->new_name = name;
 379        }
 380        if (!name)
 381                die("unable to find filename in patch at line %d", linenr);
 382}
 383
 384static int gitdiff_hdrend(const char *line, struct patch *patch)
 385{
 386        return -1;
 387}
 388
 389/*
 390 * We're anal about diff header consistency, to make
 391 * sure that we don't end up having strange ambiguous
 392 * patches floating around.
 393 *
 394 * As a result, gitdiff_{old|new}name() will check
 395 * their names against any previous information, just
 396 * to make sure..
 397 */
 398static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
 399{
 400        if (!orig_name && !isnull)
 401                return find_name(line, NULL, p_value, TERM_TAB);
 402
 403        if (orig_name) {
 404                int len;
 405                const char *name;
 406                char *another;
 407                name = orig_name;
 408                len = strlen(name);
 409                if (isnull)
 410                        die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
 411                another = find_name(line, NULL, p_value, TERM_TAB);
 412                if (!another || memcmp(another, name, len))
 413                        die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
 414                free(another);
 415                return orig_name;
 416        }
 417        else {
 418                /* expect "/dev/null" */
 419                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 420                        die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
 421                return NULL;
 422        }
 423}
 424
 425static int gitdiff_oldname(const char *line, struct patch *patch)
 426{
 427        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
 428        return 0;
 429}
 430
 431static int gitdiff_newname(const char *line, struct patch *patch)
 432{
 433        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
 434        return 0;
 435}
 436
 437static int gitdiff_oldmode(const char *line, struct patch *patch)
 438{
 439        patch->old_mode = strtoul(line, NULL, 8);
 440        return 0;
 441}
 442
 443static int gitdiff_newmode(const char *line, struct patch *patch)
 444{
 445        patch->new_mode = strtoul(line, NULL, 8);
 446        return 0;
 447}
 448
 449static int gitdiff_delete(const char *line, struct patch *patch)
 450{
 451        patch->is_delete = 1;
 452        patch->old_name = patch->def_name;
 453        return gitdiff_oldmode(line, patch);
 454}
 455
 456static int gitdiff_newfile(const char *line, struct patch *patch)
 457{
 458        patch->is_new = 1;
 459        patch->new_name = patch->def_name;
 460        return gitdiff_newmode(line, patch);
 461}
 462
 463static int gitdiff_copysrc(const char *line, struct patch *patch)
 464{
 465        patch->is_copy = 1;
 466        patch->old_name = find_name(line, NULL, 0, 0);
 467        return 0;
 468}
 469
 470static int gitdiff_copydst(const char *line, struct patch *patch)
 471{
 472        patch->is_copy = 1;
 473        patch->new_name = find_name(line, NULL, 0, 0);
 474        return 0;
 475}
 476
 477static int gitdiff_renamesrc(const char *line, struct patch *patch)
 478{
 479        patch->is_rename = 1;
 480        patch->old_name = find_name(line, NULL, 0, 0);
 481        return 0;
 482}
 483
 484static int gitdiff_renamedst(const char *line, struct patch *patch)
 485{
 486        patch->is_rename = 1;
 487        patch->new_name = find_name(line, NULL, 0, 0);
 488        return 0;
 489}
 490
 491static int gitdiff_similarity(const char *line, struct patch *patch)
 492{
 493        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 494                patch->score = 0;
 495        return 0;
 496}
 497
 498static int gitdiff_dissimilarity(const char *line, struct patch *patch)
 499{
 500        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 501                patch->score = 0;
 502        return 0;
 503}
 504
 505static int gitdiff_index(const char *line, struct patch *patch)
 506{
 507        /*
 508         * index line is N hexadecimal, "..", N hexadecimal,
 509         * and optional space with octal mode.
 510         */
 511        const char *ptr, *eol;
 512        int len;
 513
 514        ptr = strchr(line, '.');
 515        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
 516                return 0;
 517        len = ptr - line;
 518        memcpy(patch->old_sha1_prefix, line, len);
 519        patch->old_sha1_prefix[len] = 0;
 520
 521        line = ptr + 2;
 522        ptr = strchr(line, ' ');
 523        eol = strchr(line, '\n');
 524
 525        if (!ptr || eol < ptr)
 526                ptr = eol;
 527        len = ptr - line;
 528
 529        if (40 < len)
 530                return 0;
 531        memcpy(patch->new_sha1_prefix, line, len);
 532        patch->new_sha1_prefix[len] = 0;
 533        if (*ptr == ' ')
 534                patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
 535        return 0;
 536}
 537
 538/*
 539 * This is normal for a diff that doesn't change anything: we'll fall through
 540 * into the next diff. Tell the parser to break out.
 541 */
 542static int gitdiff_unrecognized(const char *line, struct patch *patch)
 543{
 544        return -1;
 545}
 546
 547static const char *stop_at_slash(const char *line, int llen)
 548{
 549        int i;
 550
 551        for (i = 0; i < llen; i++) {
 552                int ch = line[i];
 553                if (ch == '/')
 554                        return line + i;
 555        }
 556        return NULL;
 557}
 558
 559/*
 560 * This is to extract the same name that appears on "diff --git"
 561 * line.  We do not find and return anything if it is a rename
 562 * patch, and it is OK because we will find the name elsewhere.
 563 * We need to reliably find name only when it is mode-change only,
 564 * creation or deletion of an empty file.  In any of these cases,
 565 * both sides are the same name under a/ and b/ respectively.
 566 */
 567static char *git_header_name(char *line, int llen)
 568{
 569        const char *name;
 570        const char *second = NULL;
 571        size_t len;
 572
 573        line += strlen("diff --git ");
 574        llen -= strlen("diff --git ");
 575
 576        if (*line == '"') {
 577                const char *cp;
 578                struct strbuf first;
 579                struct strbuf sp;
 580
 581                strbuf_init(&first, 0);
 582                strbuf_init(&sp, 0);
 583
 584                if (unquote_c_style(&first, line, &second))
 585                        goto free_and_fail1;
 586
 587                /* advance to the first slash */
 588                cp = stop_at_slash(first.buf, first.len);
 589                /* we do not accept absolute paths */
 590                if (!cp || cp == first.buf)
 591                        goto free_and_fail1;
 592                strbuf_remove(&first, 0, cp + 1 - first.buf);
 593
 594                /*
 595                 * second points at one past closing dq of name.
 596                 * find the second name.
 597                 */
 598                while ((second < line + llen) && isspace(*second))
 599                        second++;
 600
 601                if (line + llen <= second)
 602                        goto free_and_fail1;
 603                if (*second == '"') {
 604                        if (unquote_c_style(&sp, second, NULL))
 605                                goto free_and_fail1;
 606                        cp = stop_at_slash(sp.buf, sp.len);
 607                        if (!cp || cp == sp.buf)
 608                                goto free_and_fail1;
 609                        /* They must match, otherwise ignore */
 610                        if (strcmp(cp + 1, first.buf))
 611                                goto free_and_fail1;
 612                        strbuf_release(&sp);
 613                        return strbuf_detach(&first, NULL);
 614                }
 615
 616                /* unquoted second */
 617                cp = stop_at_slash(second, line + llen - second);
 618                if (!cp || cp == second)
 619                        goto free_and_fail1;
 620                cp++;
 621                if (line + llen - cp != first.len + 1 ||
 622                    memcmp(first.buf, cp, first.len))
 623                        goto free_and_fail1;
 624                return strbuf_detach(&first, NULL);
 625
 626        free_and_fail1:
 627                strbuf_release(&first);
 628                strbuf_release(&sp);
 629                return NULL;
 630        }
 631
 632        /* unquoted first name */
 633        name = stop_at_slash(line, llen);
 634        if (!name || name == line)
 635                return NULL;
 636        name++;
 637
 638        /*
 639         * since the first name is unquoted, a dq if exists must be
 640         * the beginning of the second name.
 641         */
 642        for (second = name; second < line + llen; second++) {
 643                if (*second == '"') {
 644                        struct strbuf sp;
 645                        const char *np;
 646
 647                        strbuf_init(&sp, 0);
 648                        if (unquote_c_style(&sp, second, NULL))
 649                                goto free_and_fail2;
 650
 651                        np = stop_at_slash(sp.buf, sp.len);
 652                        if (!np || np == sp.buf)
 653                                goto free_and_fail2;
 654                        np++;
 655
 656                        len = sp.buf + sp.len - np;
 657                        if (len < second - name &&
 658                            !strncmp(np, name, len) &&
 659                            isspace(name[len])) {
 660                                /* Good */
 661                                strbuf_remove(&sp, 0, np - sp.buf);
 662                                return strbuf_detach(&sp, NULL);
 663                        }
 664
 665                free_and_fail2:
 666                        strbuf_release(&sp);
 667                        return NULL;
 668                }
 669        }
 670
 671        /*
 672         * Accept a name only if it shows up twice, exactly the same
 673         * form.
 674         */
 675        for (len = 0 ; ; len++) {
 676                switch (name[len]) {
 677                default:
 678                        continue;
 679                case '\n':
 680                        return NULL;
 681                case '\t': case ' ':
 682                        second = name+len;
 683                        for (;;) {
 684                                char c = *second++;
 685                                if (c == '\n')
 686                                        return NULL;
 687                                if (c == '/')
 688                                        break;
 689                        }
 690                        if (second[len] == '\n' && !memcmp(name, second, len)) {
 691                                return xmemdupz(name, len);
 692                        }
 693                }
 694        }
 695}
 696
 697/* Verify that we recognize the lines following a git header */
 698static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
 699{
 700        unsigned long offset;
 701
 702        /* A git diff has explicit new/delete information, so we don't guess */
 703        patch->is_new = 0;
 704        patch->is_delete = 0;
 705
 706        /*
 707         * Some things may not have the old name in the
 708         * rest of the headers anywhere (pure mode changes,
 709         * or removing or adding empty files), so we get
 710         * the default name from the header.
 711         */
 712        patch->def_name = git_header_name(line, len);
 713
 714        line += len;
 715        size -= len;
 716        linenr++;
 717        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
 718                static const struct opentry {
 719                        const char *str;
 720                        int (*fn)(const char *, struct patch *);
 721                } optable[] = {
 722                        { "@@ -", gitdiff_hdrend },
 723                        { "--- ", gitdiff_oldname },
 724                        { "+++ ", gitdiff_newname },
 725                        { "old mode ", gitdiff_oldmode },
 726                        { "new mode ", gitdiff_newmode },
 727                        { "deleted file mode ", gitdiff_delete },
 728                        { "new file mode ", gitdiff_newfile },
 729                        { "copy from ", gitdiff_copysrc },
 730                        { "copy to ", gitdiff_copydst },
 731                        { "rename old ", gitdiff_renamesrc },
 732                        { "rename new ", gitdiff_renamedst },
 733                        { "rename from ", gitdiff_renamesrc },
 734                        { "rename to ", gitdiff_renamedst },
 735                        { "similarity index ", gitdiff_similarity },
 736                        { "dissimilarity index ", gitdiff_dissimilarity },
 737                        { "index ", gitdiff_index },
 738                        { "", gitdiff_unrecognized },
 739                };
 740                int i;
 741
 742                len = linelen(line, size);
 743                if (!len || line[len-1] != '\n')
 744                        break;
 745                for (i = 0; i < ARRAY_SIZE(optable); i++) {
 746                        const struct opentry *p = optable + i;
 747                        int oplen = strlen(p->str);
 748                        if (len < oplen || memcmp(p->str, line, oplen))
 749                                continue;
 750                        if (p->fn(line + oplen, patch) < 0)
 751                                return offset;
 752                        break;
 753                }
 754        }
 755
 756        return offset;
 757}
 758
 759static int parse_num(const char *line, unsigned long *p)
 760{
 761        char *ptr;
 762
 763        if (!isdigit(*line))
 764                return 0;
 765        *p = strtoul(line, &ptr, 10);
 766        return ptr - line;
 767}
 768
 769static int parse_range(const char *line, int len, int offset, const char *expect,
 770                       unsigned long *p1, unsigned long *p2)
 771{
 772        int digits, ex;
 773
 774        if (offset < 0 || offset >= len)
 775                return -1;
 776        line += offset;
 777        len -= offset;
 778
 779        digits = parse_num(line, p1);
 780        if (!digits)
 781                return -1;
 782
 783        offset += digits;
 784        line += digits;
 785        len -= digits;
 786
 787        *p2 = 1;
 788        if (*line == ',') {
 789                digits = parse_num(line+1, p2);
 790                if (!digits)
 791                        return -1;
 792
 793                offset += digits+1;
 794                line += digits+1;
 795                len -= digits+1;
 796        }
 797
 798        ex = strlen(expect);
 799        if (ex > len)
 800                return -1;
 801        if (memcmp(line, expect, ex))
 802                return -1;
 803
 804        return offset + ex;
 805}
 806
 807/*
 808 * Parse a unified diff fragment header of the
 809 * form "@@ -a,b +c,d @@"
 810 */
 811static int parse_fragment_header(char *line, int len, struct fragment *fragment)
 812{
 813        int offset;
 814
 815        if (!len || line[len-1] != '\n')
 816                return -1;
 817
 818        /* Figure out the number of lines in a fragment */
 819        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
 820        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
 821
 822        return offset;
 823}
 824
 825static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
 826{
 827        unsigned long offset, len;
 828
 829        patch->is_toplevel_relative = 0;
 830        patch->is_rename = patch->is_copy = 0;
 831        patch->is_new = patch->is_delete = -1;
 832        patch->old_mode = patch->new_mode = 0;
 833        patch->old_name = patch->new_name = NULL;
 834        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
 835                unsigned long nextlen;
 836
 837                len = linelen(line, size);
 838                if (!len)
 839                        break;
 840
 841                /* Testing this early allows us to take a few shortcuts.. */
 842                if (len < 6)
 843                        continue;
 844
 845                /*
 846                 * Make sure we don't find any unconnected patch fragments.
 847                 * That's a sign that we didn't find a header, and that a
 848                 * patch has become corrupted/broken up.
 849                 */
 850                if (!memcmp("@@ -", line, 4)) {
 851                        struct fragment dummy;
 852                        if (parse_fragment_header(line, len, &dummy) < 0)
 853                                continue;
 854                        die("patch fragment without header at line %d: %.*s",
 855                            linenr, (int)len-1, line);
 856                }
 857
 858                if (size < len + 6)
 859                        break;
 860
 861                /*
 862                 * Git patch? It might not have a real patch, just a rename
 863                 * or mode change, so we handle that specially
 864                 */
 865                if (!memcmp("diff --git ", line, 11)) {
 866                        int git_hdr_len = parse_git_header(line, len, size, patch);
 867                        if (git_hdr_len <= len)
 868                                continue;
 869                        if (!patch->old_name && !patch->new_name) {
 870                                if (!patch->def_name)
 871                                        die("git diff header lacks filename information (line %d)", linenr);
 872                                patch->old_name = patch->new_name = patch->def_name;
 873                        }
 874                        patch->is_toplevel_relative = 1;
 875                        *hdrsize = git_hdr_len;
 876                        return offset;
 877                }
 878
 879                /* --- followed by +++ ? */
 880                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
 881                        continue;
 882
 883                /*
 884                 * We only accept unified patches, so we want it to
 885                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
 886                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
 887                 */
 888                nextlen = linelen(line + len, size - len);
 889                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
 890                        continue;
 891
 892                /* Ok, we'll consider it a patch */
 893                parse_traditional_patch(line, line+len, patch);
 894                *hdrsize = len + nextlen;
 895                linenr += 2;
 896                return offset;
 897        }
 898        return -1;
 899}
 900
 901static void check_whitespace(const char *line, int len, unsigned ws_rule)
 902{
 903        char *err;
 904        unsigned result = check_and_emit_line(line + 1, len - 1, ws_rule,
 905            NULL, NULL, NULL, NULL);
 906        if (!result)
 907                return;
 908
 909        whitespace_error++;
 910        if (squelch_whitespace_errors &&
 911            squelch_whitespace_errors < whitespace_error)
 912                ;
 913        else {
 914                err = whitespace_error_string(result);
 915                fprintf(stderr, "%s:%d: %s.\n%.*s\n",
 916                     patch_input_file, linenr, err, len - 2, line + 1);
 917                free(err);
 918        }
 919}
 920
 921/*
 922 * Parse a unified diff. Note that this really needs to parse each
 923 * fragment separately, since the only way to know the difference
 924 * between a "---" that is part of a patch, and a "---" that starts
 925 * the next patch is to look at the line counts..
 926 */
 927static int parse_fragment(char *line, unsigned long size,
 928                          struct patch *patch, struct fragment *fragment)
 929{
 930        int added, deleted;
 931        int len = linelen(line, size), offset;
 932        unsigned long oldlines, newlines;
 933        unsigned long leading, trailing;
 934
 935        offset = parse_fragment_header(line, len, fragment);
 936        if (offset < 0)
 937                return -1;
 938        oldlines = fragment->oldlines;
 939        newlines = fragment->newlines;
 940        leading = 0;
 941        trailing = 0;
 942
 943        /* Parse the thing.. */
 944        line += len;
 945        size -= len;
 946        linenr++;
 947        added = deleted = 0;
 948        for (offset = len;
 949             0 < size;
 950             offset += len, size -= len, line += len, linenr++) {
 951                if (!oldlines && !newlines)
 952                        break;
 953                len = linelen(line, size);
 954                if (!len || line[len-1] != '\n')
 955                        return -1;
 956                switch (*line) {
 957                default:
 958                        return -1;
 959                case '\n': /* newer GNU diff, an empty context line */
 960                case ' ':
 961                        oldlines--;
 962                        newlines--;
 963                        if (!deleted && !added)
 964                                leading++;
 965                        trailing++;
 966                        break;
 967                case '-':
 968                        if (apply_in_reverse &&
 969                            ws_error_action != nowarn_ws_error)
 970                                check_whitespace(line, len, patch->ws_rule);
 971                        deleted++;
 972                        oldlines--;
 973                        trailing = 0;
 974                        break;
 975                case '+':
 976                        if (!apply_in_reverse &&
 977                            ws_error_action != nowarn_ws_error)
 978                                check_whitespace(line, len, patch->ws_rule);
 979                        added++;
 980                        newlines--;
 981                        trailing = 0;
 982                        break;
 983
 984                /*
 985                 * We allow "\ No newline at end of file". Depending
 986                 * on locale settings when the patch was produced we
 987                 * don't know what this line looks like. The only
 988                 * thing we do know is that it begins with "\ ".
 989                 * Checking for 12 is just for sanity check -- any
 990                 * l10n of "\ No newline..." is at least that long.
 991                 */
 992                case '\\':
 993                        if (len < 12 || memcmp(line, "\\ ", 2))
 994                                return -1;
 995                        break;
 996                }
 997        }
 998        if (oldlines || newlines)
 999                return -1;
1000        fragment->leading = leading;
1001        fragment->trailing = trailing;
1002
1003        /*
1004         * If a fragment ends with an incomplete line, we failed to include
1005         * it in the above loop because we hit oldlines == newlines == 0
1006         * before seeing it.
1007         */
1008        if (12 < size && !memcmp(line, "\\ ", 2))
1009                offset += linelen(line, size);
1010
1011        patch->lines_added += added;
1012        patch->lines_deleted += deleted;
1013
1014        if (0 < patch->is_new && oldlines)
1015                return error("new file depends on old contents");
1016        if (0 < patch->is_delete && newlines)
1017                return error("deleted file still has contents");
1018        return offset;
1019}
1020
1021static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1022{
1023        unsigned long offset = 0;
1024        unsigned long oldlines = 0, newlines = 0, context = 0;
1025        struct fragment **fragp = &patch->fragments;
1026
1027        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1028                struct fragment *fragment;
1029                int len;
1030
1031                fragment = xcalloc(1, sizeof(*fragment));
1032                len = parse_fragment(line, size, patch, fragment);
1033                if (len <= 0)
1034                        die("corrupt patch at line %d", linenr);
1035                fragment->patch = line;
1036                fragment->size = len;
1037                oldlines += fragment->oldlines;
1038                newlines += fragment->newlines;
1039                context += fragment->leading + fragment->trailing;
1040
1041                *fragp = fragment;
1042                fragp = &fragment->next;
1043
1044                offset += len;
1045                line += len;
1046                size -= len;
1047        }
1048
1049        /*
1050         * If something was removed (i.e. we have old-lines) it cannot
1051         * be creation, and if something was added it cannot be
1052         * deletion.  However, the reverse is not true; --unified=0
1053         * patches that only add are not necessarily creation even
1054         * though they do not have any old lines, and ones that only
1055         * delete are not necessarily deletion.
1056         *
1057         * Unfortunately, a real creation/deletion patch do _not_ have
1058         * any context line by definition, so we cannot safely tell it
1059         * apart with --unified=0 insanity.  At least if the patch has
1060         * more than one hunk it is not creation or deletion.
1061         */
1062        if (patch->is_new < 0 &&
1063            (oldlines || (patch->fragments && patch->fragments->next)))
1064                patch->is_new = 0;
1065        if (patch->is_delete < 0 &&
1066            (newlines || (patch->fragments && patch->fragments->next)))
1067                patch->is_delete = 0;
1068        if (!unidiff_zero || context) {
1069                /* If the user says the patch is not generated with
1070                 * --unified=0, or if we have seen context lines,
1071                 * then not having oldlines means the patch is creation,
1072                 * and not having newlines means the patch is deletion.
1073                 */
1074                if (patch->is_new < 0 && !oldlines) {
1075                        patch->is_new = 1;
1076                        patch->old_name = NULL;
1077                }
1078                if (patch->is_delete < 0 && !newlines) {
1079                        patch->is_delete = 1;
1080                        patch->new_name = NULL;
1081                }
1082        }
1083
1084        if (0 < patch->is_new && oldlines)
1085                die("new file %s depends on old contents", patch->new_name);
1086        if (0 < patch->is_delete && newlines)
1087                die("deleted file %s still has contents", patch->old_name);
1088        if (!patch->is_delete && !newlines && context)
1089                fprintf(stderr, "** warning: file %s becomes empty but "
1090                        "is not deleted\n", patch->new_name);
1091
1092        return offset;
1093}
1094
1095static inline int metadata_changes(struct patch *patch)
1096{
1097        return  patch->is_rename > 0 ||
1098                patch->is_copy > 0 ||
1099                patch->is_new > 0 ||
1100                patch->is_delete ||
1101                (patch->old_mode && patch->new_mode &&
1102                 patch->old_mode != patch->new_mode);
1103}
1104
1105static char *inflate_it(const void *data, unsigned long size,
1106                        unsigned long inflated_size)
1107{
1108        z_stream stream;
1109        void *out;
1110        int st;
1111
1112        memset(&stream, 0, sizeof(stream));
1113
1114        stream.next_in = (unsigned char *)data;
1115        stream.avail_in = size;
1116        stream.next_out = out = xmalloc(inflated_size);
1117        stream.avail_out = inflated_size;
1118        inflateInit(&stream);
1119        st = inflate(&stream, Z_FINISH);
1120        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1121                free(out);
1122                return NULL;
1123        }
1124        return out;
1125}
1126
1127static struct fragment *parse_binary_hunk(char **buf_p,
1128                                          unsigned long *sz_p,
1129                                          int *status_p,
1130                                          int *used_p)
1131{
1132        /*
1133         * Expect a line that begins with binary patch method ("literal"
1134         * or "delta"), followed by the length of data before deflating.
1135         * a sequence of 'length-byte' followed by base-85 encoded data
1136         * should follow, terminated by a newline.
1137         *
1138         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1139         * and we would limit the patch line to 66 characters,
1140         * so one line can fit up to 13 groups that would decode
1141         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1142         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1143         */
1144        int llen, used;
1145        unsigned long size = *sz_p;
1146        char *buffer = *buf_p;
1147        int patch_method;
1148        unsigned long origlen;
1149        char *data = NULL;
1150        int hunk_size = 0;
1151        struct fragment *frag;
1152
1153        llen = linelen(buffer, size);
1154        used = llen;
1155
1156        *status_p = 0;
1157
1158        if (!prefixcmp(buffer, "delta ")) {
1159                patch_method = BINARY_DELTA_DEFLATED;
1160                origlen = strtoul(buffer + 6, NULL, 10);
1161        }
1162        else if (!prefixcmp(buffer, "literal ")) {
1163                patch_method = BINARY_LITERAL_DEFLATED;
1164                origlen = strtoul(buffer + 8, NULL, 10);
1165        }
1166        else
1167                return NULL;
1168
1169        linenr++;
1170        buffer += llen;
1171        while (1) {
1172                int byte_length, max_byte_length, newsize;
1173                llen = linelen(buffer, size);
1174                used += llen;
1175                linenr++;
1176                if (llen == 1) {
1177                        /* consume the blank line */
1178                        buffer++;
1179                        size--;
1180                        break;
1181                }
1182                /*
1183                 * Minimum line is "A00000\n" which is 7-byte long,
1184                 * and the line length must be multiple of 5 plus 2.
1185                 */
1186                if ((llen < 7) || (llen-2) % 5)
1187                        goto corrupt;
1188                max_byte_length = (llen - 2) / 5 * 4;
1189                byte_length = *buffer;
1190                if ('A' <= byte_length && byte_length <= 'Z')
1191                        byte_length = byte_length - 'A' + 1;
1192                else if ('a' <= byte_length && byte_length <= 'z')
1193                        byte_length = byte_length - 'a' + 27;
1194                else
1195                        goto corrupt;
1196                /* if the input length was not multiple of 4, we would
1197                 * have filler at the end but the filler should never
1198                 * exceed 3 bytes
1199                 */
1200                if (max_byte_length < byte_length ||
1201                    byte_length <= max_byte_length - 4)
1202                        goto corrupt;
1203                newsize = hunk_size + byte_length;
1204                data = xrealloc(data, newsize);
1205                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1206                        goto corrupt;
1207                hunk_size = newsize;
1208                buffer += llen;
1209                size -= llen;
1210        }
1211
1212        frag = xcalloc(1, sizeof(*frag));
1213        frag->patch = inflate_it(data, hunk_size, origlen);
1214        if (!frag->patch)
1215                goto corrupt;
1216        free(data);
1217        frag->size = origlen;
1218        *buf_p = buffer;
1219        *sz_p = size;
1220        *used_p = used;
1221        frag->binary_patch_method = patch_method;
1222        return frag;
1223
1224 corrupt:
1225        free(data);
1226        *status_p = -1;
1227        error("corrupt binary patch at line %d: %.*s",
1228              linenr-1, llen-1, buffer);
1229        return NULL;
1230}
1231
1232static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1233{
1234        /*
1235         * We have read "GIT binary patch\n"; what follows is a line
1236         * that says the patch method (currently, either "literal" or
1237         * "delta") and the length of data before deflating; a
1238         * sequence of 'length-byte' followed by base-85 encoded data
1239         * follows.
1240         *
1241         * When a binary patch is reversible, there is another binary
1242         * hunk in the same format, starting with patch method (either
1243         * "literal" or "delta") with the length of data, and a sequence
1244         * of length-byte + base-85 encoded data, terminated with another
1245         * empty line.  This data, when applied to the postimage, produces
1246         * the preimage.
1247         */
1248        struct fragment *forward;
1249        struct fragment *reverse;
1250        int status;
1251        int used, used_1;
1252
1253        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1254        if (!forward && !status)
1255                /* there has to be one hunk (forward hunk) */
1256                return error("unrecognized binary patch at line %d", linenr-1);
1257        if (status)
1258                /* otherwise we already gave an error message */
1259                return status;
1260
1261        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1262        if (reverse)
1263                used += used_1;
1264        else if (status) {
1265                /*
1266                 * Not having reverse hunk is not an error, but having
1267                 * a corrupt reverse hunk is.
1268                 */
1269                free((void*) forward->patch);
1270                free(forward);
1271                return status;
1272        }
1273        forward->next = reverse;
1274        patch->fragments = forward;
1275        patch->is_binary = 1;
1276        return used;
1277}
1278
1279static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1280{
1281        int hdrsize, patchsize;
1282        int offset = find_header(buffer, size, &hdrsize, patch);
1283
1284        if (offset < 0)
1285                return offset;
1286
1287        patch->ws_rule = whitespace_rule(patch->new_name
1288                                         ? patch->new_name
1289                                         : patch->old_name);
1290
1291        patchsize = parse_single_patch(buffer + offset + hdrsize,
1292                                       size - offset - hdrsize, patch);
1293
1294        if (!patchsize) {
1295                static const char *binhdr[] = {
1296                        "Binary files ",
1297                        "Files ",
1298                        NULL,
1299                };
1300                static const char git_binary[] = "GIT binary patch\n";
1301                int i;
1302                int hd = hdrsize + offset;
1303                unsigned long llen = linelen(buffer + hd, size - hd);
1304
1305                if (llen == sizeof(git_binary) - 1 &&
1306                    !memcmp(git_binary, buffer + hd, llen)) {
1307                        int used;
1308                        linenr++;
1309                        used = parse_binary(buffer + hd + llen,
1310                                            size - hd - llen, patch);
1311                        if (used)
1312                                patchsize = used + llen;
1313                        else
1314                                patchsize = 0;
1315                }
1316                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1317                        for (i = 0; binhdr[i]; i++) {
1318                                int len = strlen(binhdr[i]);
1319                                if (len < size - hd &&
1320                                    !memcmp(binhdr[i], buffer + hd, len)) {
1321                                        linenr++;
1322                                        patch->is_binary = 1;
1323                                        patchsize = llen;
1324                                        break;
1325                                }
1326                        }
1327                }
1328
1329                /* Empty patch cannot be applied if it is a text patch
1330                 * without metadata change.  A binary patch appears
1331                 * empty to us here.
1332                 */
1333                if ((apply || check) &&
1334                    (!patch->is_binary && !metadata_changes(patch)))
1335                        die("patch with only garbage at line %d", linenr);
1336        }
1337
1338        return offset + hdrsize + patchsize;
1339}
1340
1341#define swap(a,b) myswap((a),(b),sizeof(a))
1342
1343#define myswap(a, b, size) do {         \
1344        unsigned char mytmp[size];      \
1345        memcpy(mytmp, &a, size);                \
1346        memcpy(&a, &b, size);           \
1347        memcpy(&b, mytmp, size);                \
1348} while (0)
1349
1350static void reverse_patches(struct patch *p)
1351{
1352        for (; p; p = p->next) {
1353                struct fragment *frag = p->fragments;
1354
1355                swap(p->new_name, p->old_name);
1356                swap(p->new_mode, p->old_mode);
1357                swap(p->is_new, p->is_delete);
1358                swap(p->lines_added, p->lines_deleted);
1359                swap(p->old_sha1_prefix, p->new_sha1_prefix);
1360
1361                for (; frag; frag = frag->next) {
1362                        swap(frag->newpos, frag->oldpos);
1363                        swap(frag->newlines, frag->oldlines);
1364                }
1365        }
1366}
1367
1368static const char pluses[] =
1369"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1370static const char minuses[]=
1371"----------------------------------------------------------------------";
1372
1373static void show_stats(struct patch *patch)
1374{
1375        struct strbuf qname;
1376        char *cp = patch->new_name ? patch->new_name : patch->old_name;
1377        int max, add, del;
1378
1379        strbuf_init(&qname, 0);
1380        quote_c_style(cp, &qname, NULL, 0);
1381
1382        /*
1383         * "scale" the filename
1384         */
1385        max = max_len;
1386        if (max > 50)
1387                max = 50;
1388
1389        if (qname.len > max) {
1390                cp = strchr(qname.buf + qname.len + 3 - max, '/');
1391                if (!cp)
1392                        cp = qname.buf + qname.len + 3 - max;
1393                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1394        }
1395
1396        if (patch->is_binary) {
1397                printf(" %-*s |  Bin\n", max, qname.buf);
1398                strbuf_release(&qname);
1399                return;
1400        }
1401
1402        printf(" %-*s |", max, qname.buf);
1403        strbuf_release(&qname);
1404
1405        /*
1406         * scale the add/delete
1407         */
1408        max = max + max_change > 70 ? 70 - max : max_change;
1409        add = patch->lines_added;
1410        del = patch->lines_deleted;
1411
1412        if (max_change > 0) {
1413                int total = ((add + del) * max + max_change / 2) / max_change;
1414                add = (add * max + max_change / 2) / max_change;
1415                del = total - add;
1416        }
1417        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1418                add, pluses, del, minuses);
1419}
1420
1421static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1422{
1423        switch (st->st_mode & S_IFMT) {
1424        case S_IFLNK:
1425                strbuf_grow(buf, st->st_size);
1426                if (readlink(path, buf->buf, st->st_size) != st->st_size)
1427                        return -1;
1428                strbuf_setlen(buf, st->st_size);
1429                return 0;
1430        case S_IFREG:
1431                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1432                        return error("unable to open or read %s", path);
1433                convert_to_git(path, buf->buf, buf->len, buf);
1434                return 0;
1435        default:
1436                return -1;
1437        }
1438}
1439
1440static int match_fragment(const char *buf, unsigned long size,
1441                          unsigned long try,
1442                          const char *fragment, unsigned long fragsize,
1443                          int match_beginning, int match_end)
1444{
1445        if (match_beginning && try)
1446                return 0;
1447
1448        /*
1449         * Do we have an exact match?  If we were told to match
1450         * at the end, size must be exactly at try+fragsize,
1451         * otherwise try+fragsize must be still within the preimage,
1452         * and either case, the old piece should match the preimage
1453         * exactly.
1454         */
1455        if ((match_end
1456             ? (try + fragsize == size)
1457             : (try + fragsize <= size)) &&
1458            !memcmp(buf + try, fragment, fragsize))
1459                return 1;
1460
1461        /*
1462         * NEEDSWORK: We can optionally match fuzzily here, but
1463         * that is for a later round.
1464         */
1465        return 0;
1466}
1467
1468static int find_offset(const char *buf, unsigned long size,
1469                       const char *fragment, unsigned long fragsize,
1470                       int line, int *lines,
1471                       int match_beginning, int match_end)
1472{
1473        int i, no_more_backwards, no_more_forwards;
1474        unsigned long start, backwards, forwards, try;
1475
1476        if (fragsize > size)
1477                return -1;
1478
1479        start = 0;
1480        if (line > 1) {
1481                unsigned long offset = 0;
1482                i = line-1;
1483                while (offset + fragsize <= size) {
1484                        if (buf[offset++] == '\n') {
1485                                start = offset;
1486                                if (!--i)
1487                                        break;
1488                        }
1489                }
1490        }
1491
1492        /*
1493         * There's probably some smart way to do this, but I'll leave
1494         * that to the smart and beautiful people. I'm simple and stupid.
1495         */
1496        backwards = start;
1497        forwards = start;
1498        try = start;
1499
1500        for (i = 0; ; i++) {
1501                no_more_backwards = !backwards;
1502                no_more_forwards = (forwards + fragsize > size);
1503
1504                if (match_fragment(buf, size, try, fragment, fragsize,
1505                                   match_beginning, match_end)) {
1506                        int shift = ((i+1) >> 1);
1507                        if (i & 1)
1508                                shift = -shift;
1509                        *lines = shift;
1510                        return try;
1511                }
1512
1513        again:
1514                if (no_more_backwards && no_more_forwards)
1515                        break;
1516
1517                if (i & 1) {
1518                        if (no_more_backwards) {
1519                                i++;
1520                                goto again;
1521                        }
1522                        do {
1523                                --backwards;
1524                        } while (backwards && buf[backwards-1] != '\n');
1525                        try = backwards;
1526                } else {
1527                        if (no_more_forwards) {
1528                                i++;
1529                                goto again;
1530                        }
1531                        while (forwards + fragsize <= size) {
1532                                if (buf[forwards++] == '\n')
1533                                        break;
1534                        }
1535                        try = forwards;
1536                }
1537
1538        }
1539        return -1;
1540}
1541
1542static void remove_first_line(const char **rbuf, int *rsize)
1543{
1544        const char *buf = *rbuf;
1545        int size = *rsize;
1546        unsigned long offset;
1547        offset = 0;
1548        while (offset <= size) {
1549                if (buf[offset++] == '\n')
1550                        break;
1551        }
1552        *rsize = size - offset;
1553        *rbuf = buf + offset;
1554}
1555
1556static void remove_last_line(const char **rbuf, int *rsize)
1557{
1558        const char *buf = *rbuf;
1559        int size = *rsize;
1560        unsigned long offset;
1561        offset = size - 1;
1562        while (offset > 0) {
1563                if (buf[--offset] == '\n')
1564                        break;
1565        }
1566        *rsize = offset + 1;
1567}
1568
1569static int apply_line(char *output, const char *patch, int plen,
1570                      unsigned ws_rule)
1571{
1572        /*
1573         * plen is number of bytes to be copied from patch,
1574         * starting at patch+1 (patch[0] is '+').  Typically
1575         * patch[plen] is '\n', unless this is the incomplete
1576         * last line.
1577         */
1578        int i;
1579        int add_nl_to_tail = 0;
1580        int fixed = 0;
1581        int last_tab_in_indent = 0;
1582        int last_space_in_indent = 0;
1583        int need_fix_leading_space = 0;
1584        char *buf;
1585
1586        if ((ws_error_action != correct_ws_error) || !whitespace_error ||
1587            *patch != '+') {
1588                memcpy(output, patch + 1, plen);
1589                return plen;
1590        }
1591
1592        /*
1593         * Strip trailing whitespace
1594         */
1595        if ((ws_rule & WS_TRAILING_SPACE) &&
1596            (1 < plen && isspace(patch[plen-1]))) {
1597                if (patch[plen] == '\n')
1598                        add_nl_to_tail = 1;
1599                plen--;
1600                while (0 < plen && isspace(patch[plen]))
1601                        plen--;
1602                fixed = 1;
1603        }
1604
1605        /*
1606         * Check leading whitespaces (indent)
1607         */
1608        for (i = 1; i < plen; i++) {
1609                char ch = patch[i];
1610                if (ch == '\t') {
1611                        last_tab_in_indent = i;
1612                        if ((ws_rule & WS_SPACE_BEFORE_TAB) &&
1613                            0 < last_space_in_indent)
1614                            need_fix_leading_space = 1;
1615                } else if (ch == ' ') {
1616                        last_space_in_indent = i;
1617                        if ((ws_rule & WS_INDENT_WITH_NON_TAB) &&
1618                            8 <= i - last_tab_in_indent)
1619                                need_fix_leading_space = 1;
1620                }
1621                else
1622                        break;
1623        }
1624
1625        buf = output;
1626        if (need_fix_leading_space) {
1627                int consecutive_spaces = 0;
1628                int last = last_tab_in_indent + 1;
1629
1630                if (ws_rule & WS_INDENT_WITH_NON_TAB) {
1631                        /* have "last" point at one past the indent */
1632                        if (last_tab_in_indent < last_space_in_indent)
1633                                last = last_space_in_indent + 1;
1634                        else
1635                                last = last_tab_in_indent + 1;
1636                }
1637
1638                /*
1639                 * between patch[1..last], strip the funny spaces,
1640                 * updating them to tab as needed.
1641                 */
1642                for (i = 1; i < last; i++, plen--) {
1643                        char ch = patch[i];
1644                        if (ch != ' ') {
1645                                consecutive_spaces = 0;
1646                                *output++ = ch;
1647                        } else {
1648                                consecutive_spaces++;
1649                                if (consecutive_spaces == 8) {
1650                                        *output++ = '\t';
1651                                        consecutive_spaces = 0;
1652                                }
1653                        }
1654                }
1655                while (0 < consecutive_spaces--)
1656                        *output++ = ' ';
1657                fixed = 1;
1658                i = last;
1659        }
1660        else
1661                i = 1;
1662
1663        memcpy(output, patch + i, plen);
1664        if (add_nl_to_tail)
1665                output[plen++] = '\n';
1666        if (fixed)
1667                applied_after_fixing_ws++;
1668        return output + plen - buf;
1669}
1670
1671static int apply_one_fragment(struct strbuf *buf, struct fragment *frag,
1672                              int inaccurate_eof, unsigned ws_rule)
1673{
1674        int match_beginning, match_end;
1675        const char *patch = frag->patch;
1676        int offset, size = frag->size;
1677        char *old = xmalloc(size);
1678        char *new = xmalloc(size);
1679        const char *oldlines, *newlines;
1680        int oldsize = 0, newsize = 0;
1681        int new_blank_lines_at_end = 0;
1682        unsigned long leading, trailing;
1683        int pos, lines;
1684
1685        while (size > 0) {
1686                char first;
1687                int len = linelen(patch, size);
1688                int plen;
1689                int added_blank_line = 0;
1690
1691                if (!len)
1692                        break;
1693
1694                /*
1695                 * "plen" is how much of the line we should use for
1696                 * the actual patch data. Normally we just remove the
1697                 * first character on the line, but if the line is
1698                 * followed by "\ No newline", then we also remove the
1699                 * last one (which is the newline, of course).
1700                 */
1701                plen = len-1;
1702                if (len < size && patch[len] == '\\')
1703                        plen--;
1704                first = *patch;
1705                if (apply_in_reverse) {
1706                        if (first == '-')
1707                                first = '+';
1708                        else if (first == '+')
1709                                first = '-';
1710                }
1711
1712                switch (first) {
1713                case '\n':
1714                        /* Newer GNU diff, empty context line */
1715                        if (plen < 0)
1716                                /* ... followed by '\No newline'; nothing */
1717                                break;
1718                        old[oldsize++] = '\n';
1719                        new[newsize++] = '\n';
1720                        break;
1721                case ' ':
1722                case '-':
1723                        memcpy(old + oldsize, patch + 1, plen);
1724                        oldsize += plen;
1725                        if (first == '-')
1726                                break;
1727                /* Fall-through for ' ' */
1728                case '+':
1729                        if (first != '+' || !no_add) {
1730                                int added = apply_line(new + newsize, patch,
1731                                                       plen, ws_rule);
1732                                newsize += added;
1733                                if (first == '+' &&
1734                                    added == 1 && new[newsize-1] == '\n')
1735                                        added_blank_line = 1;
1736                        }
1737                        break;
1738                case '@': case '\\':
1739                        /* Ignore it, we already handled it */
1740                        break;
1741                default:
1742                        if (apply_verbosely)
1743                                error("invalid start of line: '%c'", first);
1744                        return -1;
1745                }
1746                if (added_blank_line)
1747                        new_blank_lines_at_end++;
1748                else
1749                        new_blank_lines_at_end = 0;
1750                patch += len;
1751                size -= len;
1752        }
1753
1754        if (inaccurate_eof &&
1755            oldsize > 0 && old[oldsize - 1] == '\n' &&
1756            newsize > 0 && new[newsize - 1] == '\n') {
1757                oldsize--;
1758                newsize--;
1759        }
1760
1761        oldlines = old;
1762        newlines = new;
1763        leading = frag->leading;
1764        trailing = frag->trailing;
1765
1766        /*
1767         * If we don't have any leading/trailing data in the patch,
1768         * we want it to match at the beginning/end of the file.
1769         *
1770         * But that would break if the patch is generated with
1771         * --unified=0; sane people wouldn't do that to cause us
1772         * trouble, but we try to please not so sane ones as well.
1773         */
1774        if (unidiff_zero) {
1775                match_beginning = (!leading && !frag->oldpos);
1776                match_end = 0;
1777        }
1778        else {
1779                match_beginning = !leading && (frag->oldpos == 1);
1780                match_end = !trailing;
1781        }
1782
1783        lines = 0;
1784        pos = frag->newpos;
1785        for (;;) {
1786                offset = find_offset(buf->buf, buf->len,
1787                                     oldlines, oldsize, pos, &lines,
1788                                     match_beginning, match_end);
1789                if (offset >= 0) {
1790                        if (ws_error_action == correct_ws_error &&
1791                            (buf->len - oldsize - offset == 0))
1792                                /* end of file? */
1793                                newsize -= new_blank_lines_at_end;
1794
1795                        /*
1796                         * Warn if it was necessary to reduce the number
1797                         * of context lines.
1798                         */
1799                        if ((leading != frag->leading) ||
1800                            (trailing != frag->trailing))
1801                                fprintf(stderr, "Context reduced to (%ld/%ld)"
1802                                        " to apply fragment at %d\n",
1803                                        leading, trailing, pos + lines);
1804
1805                        strbuf_splice(buf, offset, oldsize, newlines, newsize);
1806                        offset = 0;
1807                        break;
1808                }
1809
1810                /* Am I at my context limits? */
1811                if ((leading <= p_context) && (trailing <= p_context))
1812                        break;
1813                if (match_beginning || match_end) {
1814                        match_beginning = match_end = 0;
1815                        continue;
1816                }
1817                /*
1818                 * Reduce the number of context lines; reduce both
1819                 * leading and trailing if they are equal otherwise
1820                 * just reduce the larger context.
1821                 */
1822                if (leading >= trailing) {
1823                        remove_first_line(&oldlines, &oldsize);
1824                        remove_first_line(&newlines, &newsize);
1825                        pos--;
1826                        leading--;
1827                }
1828                if (trailing > leading) {
1829                        remove_last_line(&oldlines, &oldsize);
1830                        remove_last_line(&newlines, &newsize);
1831                        trailing--;
1832                }
1833        }
1834
1835        if (offset && apply_verbosely)
1836                error("while searching for:\n%.*s", oldsize, oldlines);
1837
1838        free(old);
1839        free(new);
1840        return offset;
1841}
1842
1843static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1844{
1845        struct fragment *fragment = patch->fragments;
1846        unsigned long len;
1847        void *dst;
1848
1849        /* Binary patch is irreversible without the optional second hunk */
1850        if (apply_in_reverse) {
1851                if (!fragment->next)
1852                        return error("cannot reverse-apply a binary patch "
1853                                     "without the reverse hunk to '%s'",
1854                                     patch->new_name
1855                                     ? patch->new_name : patch->old_name);
1856                fragment = fragment->next;
1857        }
1858        switch (fragment->binary_patch_method) {
1859        case BINARY_DELTA_DEFLATED:
1860                dst = patch_delta(buf->buf, buf->len, fragment->patch,
1861                                  fragment->size, &len);
1862                if (!dst)
1863                        return -1;
1864                /* XXX patch_delta NUL-terminates */
1865                strbuf_attach(buf, dst, len, len + 1);
1866                return 0;
1867        case BINARY_LITERAL_DEFLATED:
1868                strbuf_reset(buf);
1869                strbuf_add(buf, fragment->patch, fragment->size);
1870                return 0;
1871        }
1872        return -1;
1873}
1874
1875static int apply_binary(struct strbuf *buf, struct patch *patch)
1876{
1877        const char *name = patch->old_name ? patch->old_name : patch->new_name;
1878        unsigned char sha1[20];
1879
1880        /*
1881         * For safety, we require patch index line to contain
1882         * full 40-byte textual SHA1 for old and new, at least for now.
1883         */
1884        if (strlen(patch->old_sha1_prefix) != 40 ||
1885            strlen(patch->new_sha1_prefix) != 40 ||
1886            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1887            get_sha1_hex(patch->new_sha1_prefix, sha1))
1888                return error("cannot apply binary patch to '%s' "
1889                             "without full index line", name);
1890
1891        if (patch->old_name) {
1892                /*
1893                 * See if the old one matches what the patch
1894                 * applies to.
1895                 */
1896                hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1897                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1898                        return error("the patch applies to '%s' (%s), "
1899                                     "which does not match the "
1900                                     "current contents.",
1901                                     name, sha1_to_hex(sha1));
1902        }
1903        else {
1904                /* Otherwise, the old one must be empty. */
1905                if (buf->len)
1906                        return error("the patch applies to an empty "
1907                                     "'%s' but it is not empty", name);
1908        }
1909
1910        get_sha1_hex(patch->new_sha1_prefix, sha1);
1911        if (is_null_sha1(sha1)) {
1912                strbuf_release(buf);
1913                return 0; /* deletion patch */
1914        }
1915
1916        if (has_sha1_file(sha1)) {
1917                /* We already have the postimage */
1918                enum object_type type;
1919                unsigned long size;
1920                char *result;
1921
1922                result = read_sha1_file(sha1, &type, &size);
1923                if (!result)
1924                        return error("the necessary postimage %s for "
1925                                     "'%s' cannot be read",
1926                                     patch->new_sha1_prefix, name);
1927                /* XXX read_sha1_file NUL-terminates */
1928                strbuf_attach(buf, result, size, size + 1);
1929        } else {
1930                /*
1931                 * We have verified buf matches the preimage;
1932                 * apply the patch data to it, which is stored
1933                 * in the patch->fragments->{patch,size}.
1934                 */
1935                if (apply_binary_fragment(buf, patch))
1936                        return error("binary patch does not apply to '%s'",
1937                                     name);
1938
1939                /* verify that the result matches */
1940                hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1941                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1942                        return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1943                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1944        }
1945
1946        return 0;
1947}
1948
1949static int apply_fragments(struct strbuf *buf, struct patch *patch)
1950{
1951        struct fragment *frag = patch->fragments;
1952        const char *name = patch->old_name ? patch->old_name : patch->new_name;
1953        unsigned ws_rule = patch->ws_rule;
1954        unsigned inaccurate_eof = patch->inaccurate_eof;
1955
1956        if (patch->is_binary)
1957                return apply_binary(buf, patch);
1958
1959        while (frag) {
1960                if (apply_one_fragment(buf, frag, inaccurate_eof, ws_rule)) {
1961                        error("patch failed: %s:%ld", name, frag->oldpos);
1962                        if (!apply_with_reject)
1963                                return -1;
1964                        frag->rejected = 1;
1965                }
1966                frag = frag->next;
1967        }
1968        return 0;
1969}
1970
1971static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1972{
1973        if (!ce)
1974                return 0;
1975
1976        if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1977                strbuf_grow(buf, 100);
1978                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1979        } else {
1980                enum object_type type;
1981                unsigned long sz;
1982                char *result;
1983
1984                result = read_sha1_file(ce->sha1, &type, &sz);
1985                if (!result)
1986                        return -1;
1987                /* XXX read_sha1_file NUL-terminates */
1988                strbuf_attach(buf, result, sz, sz + 1);
1989        }
1990        return 0;
1991}
1992
1993static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1994{
1995        struct strbuf buf;
1996
1997        strbuf_init(&buf, 0);
1998        if (cached) {
1999                if (read_file_or_gitlink(ce, &buf))
2000                        return error("read of %s failed", patch->old_name);
2001        } else if (patch->old_name) {
2002                if (S_ISGITLINK(patch->old_mode)) {
2003                        if (ce) {
2004                                read_file_or_gitlink(ce, &buf);
2005                        } else {
2006                                /*
2007                                 * There is no way to apply subproject
2008                                 * patch without looking at the index.
2009                                 */
2010                                patch->fragments = NULL;
2011                        }
2012                } else {
2013                        if (read_old_data(st, patch->old_name, &buf))
2014                                return error("read of %s failed", patch->old_name);
2015                }
2016        }
2017
2018        if (apply_fragments(&buf, patch) < 0)
2019                return -1; /* note with --reject this succeeds. */
2020        patch->result = strbuf_detach(&buf, &patch->resultsize);
2021
2022        if (0 < patch->is_delete && patch->resultsize)
2023                return error("removal patch leaves file contents");
2024
2025        return 0;
2026}
2027
2028static int check_to_create_blob(const char *new_name, int ok_if_exists)
2029{
2030        struct stat nst;
2031        if (!lstat(new_name, &nst)) {
2032                if (S_ISDIR(nst.st_mode) || ok_if_exists)
2033                        return 0;
2034                /*
2035                 * A leading component of new_name might be a symlink
2036                 * that is going to be removed with this patch, but
2037                 * still pointing at somewhere that has the path.
2038                 * In such a case, path "new_name" does not exist as
2039                 * far as git is concerned.
2040                 */
2041                if (has_symlink_leading_path(new_name, NULL))
2042                        return 0;
2043
2044                return error("%s: already exists in working directory", new_name);
2045        }
2046        else if ((errno != ENOENT) && (errno != ENOTDIR))
2047                return error("%s: %s", new_name, strerror(errno));
2048        return 0;
2049}
2050
2051static int verify_index_match(struct cache_entry *ce, struct stat *st)
2052{
2053        if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2054                if (!S_ISDIR(st->st_mode))
2055                        return -1;
2056                return 0;
2057        }
2058        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2059}
2060
2061static int check_patch(struct patch *patch, struct patch *prev_patch)
2062{
2063        struct stat st;
2064        const char *old_name = patch->old_name;
2065        const char *new_name = patch->new_name;
2066        const char *name = old_name ? old_name : new_name;
2067        struct cache_entry *ce = NULL;
2068        int ok_if_exists;
2069
2070        patch->rejected = 1; /* we will drop this after we succeed */
2071
2072        /*
2073         * Make sure that we do not have local modifications from the
2074         * index when we are looking at the index.  Also make sure
2075         * we have the preimage file to be patched in the work tree,
2076         * unless --cached, which tells git to apply only in the index.
2077         */
2078        if (old_name) {
2079                int stat_ret = 0;
2080                unsigned st_mode = 0;
2081
2082                if (!cached)
2083                        stat_ret = lstat(old_name, &st);
2084                if (check_index) {
2085                        int pos = cache_name_pos(old_name, strlen(old_name));
2086                        if (pos < 0)
2087                                return error("%s: does not exist in index",
2088                                             old_name);
2089                        ce = active_cache[pos];
2090                        if (stat_ret < 0) {
2091                                struct checkout costate;
2092                                if (errno != ENOENT)
2093                                        return error("%s: %s", old_name,
2094                                                     strerror(errno));
2095                                /* checkout */
2096                                costate.base_dir = "";
2097                                costate.base_dir_len = 0;
2098                                costate.force = 0;
2099                                costate.quiet = 0;
2100                                costate.not_new = 0;
2101                                costate.refresh_cache = 1;
2102                                if (checkout_entry(ce,
2103                                                   &costate,
2104                                                   NULL) ||
2105                                    lstat(old_name, &st))
2106                                        return -1;
2107                        }
2108                        if (!cached && verify_index_match(ce, &st))
2109                                return error("%s: does not match index",
2110                                             old_name);
2111                        if (cached)
2112                                st_mode = ntohl(ce->ce_mode);
2113                } else if (stat_ret < 0)
2114                        return error("%s: %s", old_name, strerror(errno));
2115
2116                if (!cached)
2117                        st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2118
2119                if (patch->is_new < 0)
2120                        patch->is_new = 0;
2121                if (!patch->old_mode)
2122                        patch->old_mode = st_mode;
2123                if ((st_mode ^ patch->old_mode) & S_IFMT)
2124                        return error("%s: wrong type", old_name);
2125                if (st_mode != patch->old_mode)
2126                        fprintf(stderr, "warning: %s has type %o, expected %o\n",
2127                                old_name, st_mode, patch->old_mode);
2128        }
2129
2130        if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2131            !strcmp(prev_patch->old_name, new_name))
2132                /*
2133                 * A type-change diff is always split into a patch to
2134                 * delete old, immediately followed by a patch to
2135                 * create new (see diff.c::run_diff()); in such a case
2136                 * it is Ok that the entry to be deleted by the
2137                 * previous patch is still in the working tree and in
2138                 * the index.
2139                 */
2140                ok_if_exists = 1;
2141        else
2142                ok_if_exists = 0;
2143
2144        if (new_name &&
2145            ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2146                if (check_index &&
2147                    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2148                    !ok_if_exists)
2149                        return error("%s: already exists in index", new_name);
2150                if (!cached) {
2151                        int err = check_to_create_blob(new_name, ok_if_exists);
2152                        if (err)
2153                                return err;
2154                }
2155                if (!patch->new_mode) {
2156                        if (0 < patch->is_new)
2157                                patch->new_mode = S_IFREG | 0644;
2158                        else
2159                                patch->new_mode = patch->old_mode;
2160                }
2161        }
2162
2163        if (new_name && old_name) {
2164                int same = !strcmp(old_name, new_name);
2165                if (!patch->new_mode)
2166                        patch->new_mode = patch->old_mode;
2167                if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2168                        return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2169                                patch->new_mode, new_name, patch->old_mode,
2170                                same ? "" : " of ", same ? "" : old_name);
2171        }
2172
2173        if (apply_data(patch, &st, ce) < 0)
2174                return error("%s: patch does not apply", name);
2175        patch->rejected = 0;
2176        return 0;
2177}
2178
2179static int check_patch_list(struct patch *patch)
2180{
2181        struct patch *prev_patch = NULL;
2182        int err = 0;
2183
2184        for (prev_patch = NULL; patch ; patch = patch->next) {
2185                if (apply_verbosely)
2186                        say_patch_name(stderr,
2187                                       "Checking patch ", patch, "...\n");
2188                err |= check_patch(patch, prev_patch);
2189                prev_patch = patch;
2190        }
2191        return err;
2192}
2193
2194/* This function tries to read the sha1 from the current index */
2195static int get_current_sha1(const char *path, unsigned char *sha1)
2196{
2197        int pos;
2198
2199        if (read_cache() < 0)
2200                return -1;
2201        pos = cache_name_pos(path, strlen(path));
2202        if (pos < 0)
2203                return -1;
2204        hashcpy(sha1, active_cache[pos]->sha1);
2205        return 0;
2206}
2207
2208/* Build an index that contains the just the files needed for a 3way merge */
2209static void build_fake_ancestor(struct patch *list, const char *filename)
2210{
2211        struct patch *patch;
2212        struct index_state result = { 0 };
2213        int fd;
2214
2215        /* Once we start supporting the reverse patch, it may be
2216         * worth showing the new sha1 prefix, but until then...
2217         */
2218        for (patch = list; patch; patch = patch->next) {
2219                const unsigned char *sha1_ptr;
2220                unsigned char sha1[20];
2221                struct cache_entry *ce;
2222                const char *name;
2223
2224                name = patch->old_name ? patch->old_name : patch->new_name;
2225                if (0 < patch->is_new)
2226                        continue;
2227                else if (get_sha1(patch->old_sha1_prefix, sha1))
2228                        /* git diff has no index line for mode/type changes */
2229                        if (!patch->lines_added && !patch->lines_deleted) {
2230                                if (get_current_sha1(patch->new_name, sha1) ||
2231                                    get_current_sha1(patch->old_name, sha1))
2232                                        die("mode change for %s, which is not "
2233                                                "in current HEAD", name);
2234                                sha1_ptr = sha1;
2235                        } else
2236                                die("sha1 information is lacking or useless "
2237                                        "(%s).", name);
2238                else
2239                        sha1_ptr = sha1;
2240
2241                ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2242                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2243                        die ("Could not add %s to temporary index", name);
2244        }
2245
2246        fd = open(filename, O_WRONLY | O_CREAT, 0666);
2247        if (fd < 0 || write_index(&result, fd) || close(fd))
2248                die ("Could not write temporary index to %s", filename);
2249
2250        discard_index(&result);
2251}
2252
2253static void stat_patch_list(struct patch *patch)
2254{
2255        int files, adds, dels;
2256
2257        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2258                files++;
2259                adds += patch->lines_added;
2260                dels += patch->lines_deleted;
2261                show_stats(patch);
2262        }
2263
2264        printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2265}
2266
2267static void numstat_patch_list(struct patch *patch)
2268{
2269        for ( ; patch; patch = patch->next) {
2270                const char *name;
2271                name = patch->new_name ? patch->new_name : patch->old_name;
2272                if (patch->is_binary)
2273                        printf("-\t-\t");
2274                else
2275                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2276                write_name_quoted(name, stdout, line_termination);
2277        }
2278}
2279
2280static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2281{
2282        if (mode)
2283                printf(" %s mode %06o %s\n", newdelete, mode, name);
2284        else
2285                printf(" %s %s\n", newdelete, name);
2286}
2287
2288static void show_mode_change(struct patch *p, int show_name)
2289{
2290        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2291                if (show_name)
2292                        printf(" mode change %06o => %06o %s\n",
2293                               p->old_mode, p->new_mode, p->new_name);
2294                else
2295                        printf(" mode change %06o => %06o\n",
2296                               p->old_mode, p->new_mode);
2297        }
2298}
2299
2300static void show_rename_copy(struct patch *p)
2301{
2302        const char *renamecopy = p->is_rename ? "rename" : "copy";
2303        const char *old, *new;
2304
2305        /* Find common prefix */
2306        old = p->old_name;
2307        new = p->new_name;
2308        while (1) {
2309                const char *slash_old, *slash_new;
2310                slash_old = strchr(old, '/');
2311                slash_new = strchr(new, '/');
2312                if (!slash_old ||
2313                    !slash_new ||
2314                    slash_old - old != slash_new - new ||
2315                    memcmp(old, new, slash_new - new))
2316                        break;
2317                old = slash_old + 1;
2318                new = slash_new + 1;
2319        }
2320        /* p->old_name thru old is the common prefix, and old and new
2321         * through the end of names are renames
2322         */
2323        if (old != p->old_name)
2324                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2325                       (int)(old - p->old_name), p->old_name,
2326                       old, new, p->score);
2327        else
2328                printf(" %s %s => %s (%d%%)\n", renamecopy,
2329                       p->old_name, p->new_name, p->score);
2330        show_mode_change(p, 0);
2331}
2332
2333static void summary_patch_list(struct patch *patch)
2334{
2335        struct patch *p;
2336
2337        for (p = patch; p; p = p->next) {
2338                if (p->is_new)
2339                        show_file_mode_name("create", p->new_mode, p->new_name);
2340                else if (p->is_delete)
2341                        show_file_mode_name("delete", p->old_mode, p->old_name);
2342                else {
2343                        if (p->is_rename || p->is_copy)
2344                                show_rename_copy(p);
2345                        else {
2346                                if (p->score) {
2347                                        printf(" rewrite %s (%d%%)\n",
2348                                               p->new_name, p->score);
2349                                        show_mode_change(p, 0);
2350                                }
2351                                else
2352                                        show_mode_change(p, 1);
2353                        }
2354                }
2355        }
2356}
2357
2358static void patch_stats(struct patch *patch)
2359{
2360        int lines = patch->lines_added + patch->lines_deleted;
2361
2362        if (lines > max_change)
2363                max_change = lines;
2364        if (patch->old_name) {
2365                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2366                if (!len)
2367                        len = strlen(patch->old_name);
2368                if (len > max_len)
2369                        max_len = len;
2370        }
2371        if (patch->new_name) {
2372                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2373                if (!len)
2374                        len = strlen(patch->new_name);
2375                if (len > max_len)
2376                        max_len = len;
2377        }
2378}
2379
2380static void remove_file(struct patch *patch, int rmdir_empty)
2381{
2382        if (update_index) {
2383                if (remove_file_from_cache(patch->old_name) < 0)
2384                        die("unable to remove %s from index", patch->old_name);
2385        }
2386        if (!cached) {
2387                if (S_ISGITLINK(patch->old_mode)) {
2388                        if (rmdir(patch->old_name))
2389                                warning("unable to remove submodule %s",
2390                                        patch->old_name);
2391                } else if (!unlink(patch->old_name) && rmdir_empty) {
2392                        char *name = xstrdup(patch->old_name);
2393                        char *end = strrchr(name, '/');
2394                        while (end) {
2395                                *end = 0;
2396                                if (rmdir(name))
2397                                        break;
2398                                end = strrchr(name, '/');
2399                        }
2400                        free(name);
2401                }
2402        }
2403}
2404
2405static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2406{
2407        struct stat st;
2408        struct cache_entry *ce;
2409        int namelen = strlen(path);
2410        unsigned ce_size = cache_entry_size(namelen);
2411
2412        if (!update_index)
2413                return;
2414
2415        ce = xcalloc(1, ce_size);
2416        memcpy(ce->name, path, namelen);
2417        ce->ce_mode = create_ce_mode(mode);
2418        ce->ce_flags = htons(namelen);
2419        if (S_ISGITLINK(mode)) {
2420                const char *s = buf;
2421
2422                if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2423                        die("corrupt patch for subproject %s", path);
2424        } else {
2425                if (!cached) {
2426                        if (lstat(path, &st) < 0)
2427                                die("unable to stat newly created file %s",
2428                                    path);
2429                        fill_stat_cache_info(ce, &st);
2430                }
2431                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2432                        die("unable to create backing store for newly created file %s", path);
2433        }
2434        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2435                die("unable to add cache entry for %s", path);
2436}
2437
2438static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2439{
2440        int fd;
2441        struct strbuf nbuf;
2442
2443        if (S_ISGITLINK(mode)) {
2444                struct stat st;
2445                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2446                        return 0;
2447                return mkdir(path, 0777);
2448        }
2449
2450        if (has_symlinks && S_ISLNK(mode))
2451                /* Although buf:size is counted string, it also is NUL
2452                 * terminated.
2453                 */
2454                return symlink(buf, path);
2455
2456        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2457        if (fd < 0)
2458                return -1;
2459
2460        strbuf_init(&nbuf, 0);
2461        if (convert_to_working_tree(path, buf, size, &nbuf)) {
2462                size = nbuf.len;
2463                buf  = nbuf.buf;
2464        }
2465        write_or_die(fd, buf, size);
2466        strbuf_release(&nbuf);
2467
2468        if (close(fd) < 0)
2469                die("closing file %s: %s", path, strerror(errno));
2470        return 0;
2471}
2472
2473/*
2474 * We optimistically assume that the directories exist,
2475 * which is true 99% of the time anyway. If they don't,
2476 * we create them and try again.
2477 */
2478static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2479{
2480        if (cached)
2481                return;
2482        if (!try_create_file(path, mode, buf, size))
2483                return;
2484
2485        if (errno == ENOENT) {
2486                if (safe_create_leading_directories(path))
2487                        return;
2488                if (!try_create_file(path, mode, buf, size))
2489                        return;
2490        }
2491
2492        if (errno == EEXIST || errno == EACCES) {
2493                /* We may be trying to create a file where a directory
2494                 * used to be.
2495                 */
2496                struct stat st;
2497                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2498                        errno = EEXIST;
2499        }
2500
2501        if (errno == EEXIST) {
2502                unsigned int nr = getpid();
2503
2504                for (;;) {
2505                        const char *newpath;
2506                        newpath = mkpath("%s~%u", path, nr);
2507                        if (!try_create_file(newpath, mode, buf, size)) {
2508                                if (!rename(newpath, path))
2509                                        return;
2510                                unlink(newpath);
2511                                break;
2512                        }
2513                        if (errno != EEXIST)
2514                                break;
2515                        ++nr;
2516                }
2517        }
2518        die("unable to write file %s mode %o", path, mode);
2519}
2520
2521static void create_file(struct patch *patch)
2522{
2523        char *path = patch->new_name;
2524        unsigned mode = patch->new_mode;
2525        unsigned long size = patch->resultsize;
2526        char *buf = patch->result;
2527
2528        if (!mode)
2529                mode = S_IFREG | 0644;
2530        create_one_file(path, mode, buf, size);
2531        add_index_file(path, mode, buf, size);
2532}
2533
2534/* phase zero is to remove, phase one is to create */
2535static void write_out_one_result(struct patch *patch, int phase)
2536{
2537        if (patch->is_delete > 0) {
2538                if (phase == 0)
2539                        remove_file(patch, 1);
2540                return;
2541        }
2542        if (patch->is_new > 0 || patch->is_copy) {
2543                if (phase == 1)
2544                        create_file(patch);
2545                return;
2546        }
2547        /*
2548         * Rename or modification boils down to the same
2549         * thing: remove the old, write the new
2550         */
2551        if (phase == 0)
2552                remove_file(patch, patch->is_rename);
2553        if (phase == 1)
2554                create_file(patch);
2555}
2556
2557static int write_out_one_reject(struct patch *patch)
2558{
2559        FILE *rej;
2560        char namebuf[PATH_MAX];
2561        struct fragment *frag;
2562        int cnt = 0;
2563
2564        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2565                if (!frag->rejected)
2566                        continue;
2567                cnt++;
2568        }
2569
2570        if (!cnt) {
2571                if (apply_verbosely)
2572                        say_patch_name(stderr,
2573                                       "Applied patch ", patch, " cleanly.\n");
2574                return 0;
2575        }
2576
2577        /* This should not happen, because a removal patch that leaves
2578         * contents are marked "rejected" at the patch level.
2579         */
2580        if (!patch->new_name)
2581                die("internal error");
2582
2583        /* Say this even without --verbose */
2584        say_patch_name(stderr, "Applying patch ", patch, " with");
2585        fprintf(stderr, " %d rejects...\n", cnt);
2586
2587        cnt = strlen(patch->new_name);
2588        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2589                cnt = ARRAY_SIZE(namebuf) - 5;
2590                fprintf(stderr,
2591                        "warning: truncating .rej filename to %.*s.rej",
2592                        cnt - 1, patch->new_name);
2593        }
2594        memcpy(namebuf, patch->new_name, cnt);
2595        memcpy(namebuf + cnt, ".rej", 5);
2596
2597        rej = fopen(namebuf, "w");
2598        if (!rej)
2599                return error("cannot open %s: %s", namebuf, strerror(errno));
2600
2601        /* Normal git tools never deal with .rej, so do not pretend
2602         * this is a git patch by saying --git nor give extended
2603         * headers.  While at it, maybe please "kompare" that wants
2604         * the trailing TAB and some garbage at the end of line ;-).
2605         */
2606        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2607                patch->new_name, patch->new_name);
2608        for (cnt = 1, frag = patch->fragments;
2609             frag;
2610             cnt++, frag = frag->next) {
2611                if (!frag->rejected) {
2612                        fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2613                        continue;
2614                }
2615                fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2616                fprintf(rej, "%.*s", frag->size, frag->patch);
2617                if (frag->patch[frag->size-1] != '\n')
2618                        fputc('\n', rej);
2619        }
2620        fclose(rej);
2621        return -1;
2622}
2623
2624static int write_out_results(struct patch *list, int skipped_patch)
2625{
2626        int phase;
2627        int errs = 0;
2628        struct patch *l;
2629
2630        if (!list && !skipped_patch)
2631                return error("No changes");
2632
2633        for (phase = 0; phase < 2; phase++) {
2634                l = list;
2635                while (l) {
2636                        if (l->rejected)
2637                                errs = 1;
2638                        else {
2639                                write_out_one_result(l, phase);
2640                                if (phase == 1 && write_out_one_reject(l))
2641                                        errs = 1;
2642                        }
2643                        l = l->next;
2644                }
2645        }
2646        return errs;
2647}
2648
2649static struct lock_file lock_file;
2650
2651static struct excludes {
2652        struct excludes *next;
2653        const char *path;
2654} *excludes;
2655
2656static int use_patch(struct patch *p)
2657{
2658        const char *pathname = p->new_name ? p->new_name : p->old_name;
2659        struct excludes *x = excludes;
2660        while (x) {
2661                if (fnmatch(x->path, pathname, 0) == 0)
2662                        return 0;
2663                x = x->next;
2664        }
2665        if (0 < prefix_length) {
2666                int pathlen = strlen(pathname);
2667                if (pathlen <= prefix_length ||
2668                    memcmp(prefix, pathname, prefix_length))
2669                        return 0;
2670        }
2671        return 1;
2672}
2673
2674static void prefix_one(char **name)
2675{
2676        char *old_name = *name;
2677        if (!old_name)
2678                return;
2679        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2680        free(old_name);
2681}
2682
2683static void prefix_patches(struct patch *p)
2684{
2685        if (!prefix || p->is_toplevel_relative)
2686                return;
2687        for ( ; p; p = p->next) {
2688                if (p->new_name == p->old_name) {
2689                        char *prefixed = p->new_name;
2690                        prefix_one(&prefixed);
2691                        p->new_name = p->old_name = prefixed;
2692                }
2693                else {
2694                        prefix_one(&p->new_name);
2695                        prefix_one(&p->old_name);
2696                }
2697        }
2698}
2699
2700static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2701{
2702        size_t offset;
2703        struct strbuf buf;
2704        struct patch *list = NULL, **listp = &list;
2705        int skipped_patch = 0;
2706
2707        strbuf_init(&buf, 0);
2708        patch_input_file = filename;
2709        read_patch_file(&buf, fd);
2710        offset = 0;
2711        while (offset < buf.len) {
2712                struct patch *patch;
2713                int nr;
2714
2715                patch = xcalloc(1, sizeof(*patch));
2716                patch->inaccurate_eof = inaccurate_eof;
2717                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2718                if (nr < 0)
2719                        break;
2720                if (apply_in_reverse)
2721                        reverse_patches(patch);
2722                if (prefix)
2723                        prefix_patches(patch);
2724                if (use_patch(patch)) {
2725                        patch_stats(patch);
2726                        *listp = patch;
2727                        listp = &patch->next;
2728                }
2729                else {
2730                        /* perhaps free it a bit better? */
2731                        free(patch);
2732                        skipped_patch++;
2733                }
2734                offset += nr;
2735        }
2736
2737        if (whitespace_error && (ws_error_action == die_on_ws_error))
2738                apply = 0;
2739
2740        update_index = check_index && apply;
2741        if (update_index && newfd < 0)
2742                newfd = hold_locked_index(&lock_file, 1);
2743
2744        if (check_index) {
2745                if (read_cache() < 0)
2746                        die("unable to read index file");
2747        }
2748
2749        if ((check || apply) &&
2750            check_patch_list(list) < 0 &&
2751            !apply_with_reject)
2752                exit(1);
2753
2754        if (apply && write_out_results(list, skipped_patch))
2755                exit(1);
2756
2757        if (fake_ancestor)
2758                build_fake_ancestor(list, fake_ancestor);
2759
2760        if (diffstat)
2761                stat_patch_list(list);
2762
2763        if (numstat)
2764                numstat_patch_list(list);
2765
2766        if (summary)
2767                summary_patch_list(list);
2768
2769        strbuf_release(&buf);
2770        return 0;
2771}
2772
2773static int git_apply_config(const char *var, const char *value)
2774{
2775        if (!strcmp(var, "apply.whitespace")) {
2776                apply_default_whitespace = xstrdup(value);
2777                return 0;
2778        }
2779        return git_default_config(var, value);
2780}
2781
2782
2783int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2784{
2785        int i;
2786        int read_stdin = 1;
2787        int inaccurate_eof = 0;
2788        int errs = 0;
2789        int is_not_gitdir = 0;
2790
2791        const char *whitespace_option = NULL;
2792
2793        prefix = setup_git_directory_gently(&is_not_gitdir);
2794        prefix_length = prefix ? strlen(prefix) : 0;
2795        git_config(git_apply_config);
2796        if (apply_default_whitespace)
2797                parse_whitespace_option(apply_default_whitespace);
2798
2799        for (i = 1; i < argc; i++) {
2800                const char *arg = argv[i];
2801                char *end;
2802                int fd;
2803
2804                if (!strcmp(arg, "-")) {
2805                        errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2806                        read_stdin = 0;
2807                        continue;
2808                }
2809                if (!prefixcmp(arg, "--exclude=")) {
2810                        struct excludes *x = xmalloc(sizeof(*x));
2811                        x->path = arg + 10;
2812                        x->next = excludes;
2813                        excludes = x;
2814                        continue;
2815                }
2816                if (!prefixcmp(arg, "-p")) {
2817                        p_value = atoi(arg + 2);
2818                        p_value_known = 1;
2819                        continue;
2820                }
2821                if (!strcmp(arg, "--no-add")) {
2822                        no_add = 1;
2823                        continue;
2824                }
2825                if (!strcmp(arg, "--stat")) {
2826                        apply = 0;
2827                        diffstat = 1;
2828                        continue;
2829                }
2830                if (!strcmp(arg, "--allow-binary-replacement") ||
2831                    !strcmp(arg, "--binary")) {
2832                        continue; /* now no-op */
2833                }
2834                if (!strcmp(arg, "--numstat")) {
2835                        apply = 0;
2836                        numstat = 1;
2837                        continue;
2838                }
2839                if (!strcmp(arg, "--summary")) {
2840                        apply = 0;
2841                        summary = 1;
2842                        continue;
2843                }
2844                if (!strcmp(arg, "--check")) {
2845                        apply = 0;
2846                        check = 1;
2847                        continue;
2848                }
2849                if (!strcmp(arg, "--index")) {
2850                        if (is_not_gitdir)
2851                                die("--index outside a repository");
2852                        check_index = 1;
2853                        continue;
2854                }
2855                if (!strcmp(arg, "--cached")) {
2856                        if (is_not_gitdir)
2857                                die("--cached outside a repository");
2858                        check_index = 1;
2859                        cached = 1;
2860                        continue;
2861                }
2862                if (!strcmp(arg, "--apply")) {
2863                        apply = 1;
2864                        continue;
2865                }
2866                if (!strcmp(arg, "--build-fake-ancestor")) {
2867                        apply = 0;
2868                        if (++i >= argc)
2869                                die ("need a filename");
2870                        fake_ancestor = argv[i];
2871                        continue;
2872                }
2873                if (!strcmp(arg, "-z")) {
2874                        line_termination = 0;
2875                        continue;
2876                }
2877                if (!prefixcmp(arg, "-C")) {
2878                        p_context = strtoul(arg + 2, &end, 0);
2879                        if (*end != '\0')
2880                                die("unrecognized context count '%s'", arg + 2);
2881                        continue;
2882                }
2883                if (!prefixcmp(arg, "--whitespace=")) {
2884                        whitespace_option = arg + 13;
2885                        parse_whitespace_option(arg + 13);
2886                        continue;
2887                }
2888                if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2889                        apply_in_reverse = 1;
2890                        continue;
2891                }
2892                if (!strcmp(arg, "--unidiff-zero")) {
2893                        unidiff_zero = 1;
2894                        continue;
2895                }
2896                if (!strcmp(arg, "--reject")) {
2897                        apply = apply_with_reject = apply_verbosely = 1;
2898                        continue;
2899                }
2900                if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2901                        apply_verbosely = 1;
2902                        continue;
2903                }
2904                if (!strcmp(arg, "--inaccurate-eof")) {
2905                        inaccurate_eof = 1;
2906                        continue;
2907                }
2908                if (0 < prefix_length)
2909                        arg = prefix_filename(prefix, prefix_length, arg);
2910
2911                fd = open(arg, O_RDONLY);
2912                if (fd < 0)
2913                        usage(apply_usage);
2914                read_stdin = 0;
2915                set_default_whitespace_mode(whitespace_option);
2916                errs |= apply_patch(fd, arg, inaccurate_eof);
2917                close(fd);
2918        }
2919        set_default_whitespace_mode(whitespace_option);
2920        if (read_stdin)
2921                errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2922        if (whitespace_error) {
2923                if (squelch_whitespace_errors &&
2924                    squelch_whitespace_errors < whitespace_error) {
2925                        int squelched =
2926                                whitespace_error - squelch_whitespace_errors;
2927                        fprintf(stderr, "warning: squelched %d "
2928                                "whitespace error%s\n",
2929                                squelched,
2930                                squelched == 1 ? "" : "s");
2931                }
2932                if (ws_error_action == die_on_ws_error)
2933                        die("%d line%s add%s whitespace errors.",
2934                            whitespace_error,
2935                            whitespace_error == 1 ? "" : "s",
2936                            whitespace_error == 1 ? "s" : "");
2937                if (applied_after_fixing_ws && apply)
2938                        fprintf(stderr, "warning: %d line%s applied after"
2939                                " fixing whitespace errors.\n",
2940                                applied_after_fixing_ws,
2941                                applied_after_fixing_ws == 1 ? "" : "s");
2942                else if (whitespace_error)
2943                        fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2944                                whitespace_error,
2945                                whitespace_error == 1 ? "" : "s",
2946                                whitespace_error == 1 ? "s" : "");
2947        }
2948
2949        if (update_index) {
2950                if (write_cache(newfd, active_cache, active_nr) ||
2951                    commit_locked_index(&lock_file))
2952                        die("Unable to write new index file");
2953        }
2954
2955        return !!errs;
2956}