builtin-apply.con commit git-apply --whitespace=fix: fix whitespace fuzz introduced by previous run (c1beba5)
   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
 164/*
 165 * A line in a file, len-bytes long (includes the terminating LF,
 166 * except for an incomplete line at the end if the file ends with
 167 * one), and its contents hashes to 'hash'.
 168 */
 169struct line {
 170        size_t len;
 171        unsigned hash : 24;
 172        unsigned flag : 8;
 173#define LINE_COMMON     1
 174};
 175
 176/*
 177 * This represents a "file", which is an array of "lines".
 178 */
 179struct image {
 180        char *buf;
 181        size_t len;
 182        size_t nr;
 183        size_t alloc;
 184        struct line *line_allocated;
 185        struct line *line;
 186};
 187
 188static uint32_t hash_line(const char *cp, size_t len)
 189{
 190        size_t i;
 191        uint32_t h;
 192        for (i = 0, h = 0; i < len; i++) {
 193                if (!isspace(cp[i])) {
 194                        h = h * 3 + (cp[i] & 0xff);
 195                }
 196        }
 197        return h;
 198}
 199
 200static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 201{
 202        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 203        img->line_allocated[img->nr].len = len;
 204        img->line_allocated[img->nr].hash = hash_line(bol, len);
 205        img->line_allocated[img->nr].flag = flag;
 206        img->nr++;
 207}
 208
 209static void prepare_image(struct image *image, char *buf, size_t len,
 210                          int prepare_linetable)
 211{
 212        const char *cp, *ep;
 213
 214        memset(image, 0, sizeof(*image));
 215        image->buf = buf;
 216        image->len = len;
 217
 218        if (!prepare_linetable)
 219                return;
 220
 221        ep = image->buf + image->len;
 222        cp = image->buf;
 223        while (cp < ep) {
 224                const char *next;
 225                for (next = cp; next < ep && *next != '\n'; next++)
 226                        ;
 227                if (next < ep)
 228                        next++;
 229                add_line_info(image, cp, next - cp, 0);
 230                cp = next;
 231        }
 232        image->line = image->line_allocated;
 233}
 234
 235static void clear_image(struct image *image)
 236{
 237        free(image->buf);
 238        image->buf = NULL;
 239        image->len = 0;
 240}
 241
 242static void say_patch_name(FILE *output, const char *pre,
 243                           struct patch *patch, const char *post)
 244{
 245        fputs(pre, output);
 246        if (patch->old_name && patch->new_name &&
 247            strcmp(patch->old_name, patch->new_name)) {
 248                quote_c_style(patch->old_name, NULL, output, 0);
 249                fputs(" => ", output);
 250                quote_c_style(patch->new_name, NULL, output, 0);
 251        } else {
 252                const char *n = patch->new_name;
 253                if (!n)
 254                        n = patch->old_name;
 255                quote_c_style(n, NULL, output, 0);
 256        }
 257        fputs(post, output);
 258}
 259
 260#define CHUNKSIZE (8192)
 261#define SLOP (16)
 262
 263static void read_patch_file(struct strbuf *sb, int fd)
 264{
 265        if (strbuf_read(sb, fd, 0) < 0)
 266                die("git-apply: read returned %s", strerror(errno));
 267
 268        /*
 269         * Make sure that we have some slop in the buffer
 270         * so that we can do speculative "memcmp" etc, and
 271         * see to it that it is NUL-filled.
 272         */
 273        strbuf_grow(sb, SLOP);
 274        memset(sb->buf + sb->len, 0, SLOP);
 275}
 276
 277static unsigned long linelen(const char *buffer, unsigned long size)
 278{
 279        unsigned long len = 0;
 280        while (size--) {
 281                len++;
 282                if (*buffer++ == '\n')
 283                        break;
 284        }
 285        return len;
 286}
 287
 288static int is_dev_null(const char *str)
 289{
 290        return !memcmp("/dev/null", str, 9) && isspace(str[9]);
 291}
 292
 293#define TERM_SPACE      1
 294#define TERM_TAB        2
 295
 296static int name_terminate(const char *name, int namelen, int c, int terminate)
 297{
 298        if (c == ' ' && !(terminate & TERM_SPACE))
 299                return 0;
 300        if (c == '\t' && !(terminate & TERM_TAB))
 301                return 0;
 302
 303        return 1;
 304}
 305
 306static char *find_name(const char *line, char *def, int p_value, int terminate)
 307{
 308        int len;
 309        const char *start = line;
 310
 311        if (*line == '"') {
 312                struct strbuf name;
 313
 314                /*
 315                 * Proposed "new-style" GNU patch/diff format; see
 316                 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
 317                 */
 318                strbuf_init(&name, 0);
 319                if (!unquote_c_style(&name, line, NULL)) {
 320                        char *cp;
 321
 322                        for (cp = name.buf; p_value; p_value--) {
 323                                cp = strchr(cp, '/');
 324                                if (!cp)
 325                                        break;
 326                                cp++;
 327                        }
 328                        if (cp) {
 329                                /* name can later be freed, so we need
 330                                 * to memmove, not just return cp
 331                                 */
 332                                strbuf_remove(&name, 0, cp - name.buf);
 333                                free(def);
 334                                return strbuf_detach(&name, NULL);
 335                        }
 336                }
 337                strbuf_release(&name);
 338        }
 339
 340        for (;;) {
 341                char c = *line;
 342
 343                if (isspace(c)) {
 344                        if (c == '\n')
 345                                break;
 346                        if (name_terminate(start, line-start, c, terminate))
 347                                break;
 348                }
 349                line++;
 350                if (c == '/' && !--p_value)
 351                        start = line;
 352        }
 353        if (!start)
 354                return def;
 355        len = line - start;
 356        if (!len)
 357                return def;
 358
 359        /*
 360         * Generally we prefer the shorter name, especially
 361         * if the other one is just a variation of that with
 362         * something else tacked on to the end (ie "file.orig"
 363         * or "file~").
 364         */
 365        if (def) {
 366                int deflen = strlen(def);
 367                if (deflen < len && !strncmp(start, def, deflen))
 368                        return def;
 369                free(def);
 370        }
 371
 372        return xmemdupz(start, len);
 373}
 374
 375static int count_slashes(const char *cp)
 376{
 377        int cnt = 0;
 378        char ch;
 379
 380        while ((ch = *cp++))
 381                if (ch == '/')
 382                        cnt++;
 383        return cnt;
 384}
 385
 386/*
 387 * Given the string after "--- " or "+++ ", guess the appropriate
 388 * p_value for the given patch.
 389 */
 390static int guess_p_value(const char *nameline)
 391{
 392        char *name, *cp;
 393        int val = -1;
 394
 395        if (is_dev_null(nameline))
 396                return -1;
 397        name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
 398        if (!name)
 399                return -1;
 400        cp = strchr(name, '/');
 401        if (!cp)
 402                val = 0;
 403        else if (prefix) {
 404                /*
 405                 * Does it begin with "a/$our-prefix" and such?  Then this is
 406                 * very likely to apply to our directory.
 407                 */
 408                if (!strncmp(name, prefix, prefix_length))
 409                        val = count_slashes(prefix);
 410                else {
 411                        cp++;
 412                        if (!strncmp(cp, prefix, prefix_length))
 413                                val = count_slashes(prefix) + 1;
 414                }
 415        }
 416        free(name);
 417        return val;
 418}
 419
 420/*
 421 * Get the name etc info from the --/+++ lines of a traditional patch header
 422 *
 423 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 424 * files, we can happily check the index for a match, but for creating a
 425 * new file we should try to match whatever "patch" does. I have no idea.
 426 */
 427static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 428{
 429        char *name;
 430
 431        first += 4;     /* skip "--- " */
 432        second += 4;    /* skip "+++ " */
 433        if (!p_value_known) {
 434                int p, q;
 435                p = guess_p_value(first);
 436                q = guess_p_value(second);
 437                if (p < 0) p = q;
 438                if (0 <= p && p == q) {
 439                        p_value = p;
 440                        p_value_known = 1;
 441                }
 442        }
 443        if (is_dev_null(first)) {
 444                patch->is_new = 1;
 445                patch->is_delete = 0;
 446                name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
 447                patch->new_name = name;
 448        } else if (is_dev_null(second)) {
 449                patch->is_new = 0;
 450                patch->is_delete = 1;
 451                name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 452                patch->old_name = name;
 453        } else {
 454                name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 455                name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
 456                patch->old_name = patch->new_name = name;
 457        }
 458        if (!name)
 459                die("unable to find filename in patch at line %d", linenr);
 460}
 461
 462static int gitdiff_hdrend(const char *line, struct patch *patch)
 463{
 464        return -1;
 465}
 466
 467/*
 468 * We're anal about diff header consistency, to make
 469 * sure that we don't end up having strange ambiguous
 470 * patches floating around.
 471 *
 472 * As a result, gitdiff_{old|new}name() will check
 473 * their names against any previous information, just
 474 * to make sure..
 475 */
 476static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
 477{
 478        if (!orig_name && !isnull)
 479                return find_name(line, NULL, p_value, TERM_TAB);
 480
 481        if (orig_name) {
 482                int len;
 483                const char *name;
 484                char *another;
 485                name = orig_name;
 486                len = strlen(name);
 487                if (isnull)
 488                        die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
 489                another = find_name(line, NULL, p_value, TERM_TAB);
 490                if (!another || memcmp(another, name, len))
 491                        die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
 492                free(another);
 493                return orig_name;
 494        }
 495        else {
 496                /* expect "/dev/null" */
 497                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 498                        die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
 499                return NULL;
 500        }
 501}
 502
 503static int gitdiff_oldname(const char *line, struct patch *patch)
 504{
 505        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
 506        return 0;
 507}
 508
 509static int gitdiff_newname(const char *line, struct patch *patch)
 510{
 511        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
 512        return 0;
 513}
 514
 515static int gitdiff_oldmode(const char *line, struct patch *patch)
 516{
 517        patch->old_mode = strtoul(line, NULL, 8);
 518        return 0;
 519}
 520
 521static int gitdiff_newmode(const char *line, struct patch *patch)
 522{
 523        patch->new_mode = strtoul(line, NULL, 8);
 524        return 0;
 525}
 526
 527static int gitdiff_delete(const char *line, struct patch *patch)
 528{
 529        patch->is_delete = 1;
 530        patch->old_name = patch->def_name;
 531        return gitdiff_oldmode(line, patch);
 532}
 533
 534static int gitdiff_newfile(const char *line, struct patch *patch)
 535{
 536        patch->is_new = 1;
 537        patch->new_name = patch->def_name;
 538        return gitdiff_newmode(line, patch);
 539}
 540
 541static int gitdiff_copysrc(const char *line, struct patch *patch)
 542{
 543        patch->is_copy = 1;
 544        patch->old_name = find_name(line, NULL, 0, 0);
 545        return 0;
 546}
 547
 548static int gitdiff_copydst(const char *line, struct patch *patch)
 549{
 550        patch->is_copy = 1;
 551        patch->new_name = find_name(line, NULL, 0, 0);
 552        return 0;
 553}
 554
 555static int gitdiff_renamesrc(const char *line, struct patch *patch)
 556{
 557        patch->is_rename = 1;
 558        patch->old_name = find_name(line, NULL, 0, 0);
 559        return 0;
 560}
 561
 562static int gitdiff_renamedst(const char *line, struct patch *patch)
 563{
 564        patch->is_rename = 1;
 565        patch->new_name = find_name(line, NULL, 0, 0);
 566        return 0;
 567}
 568
 569static int gitdiff_similarity(const char *line, struct patch *patch)
 570{
 571        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 572                patch->score = 0;
 573        return 0;
 574}
 575
 576static int gitdiff_dissimilarity(const char *line, struct patch *patch)
 577{
 578        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 579                patch->score = 0;
 580        return 0;
 581}
 582
 583static int gitdiff_index(const char *line, struct patch *patch)
 584{
 585        /*
 586         * index line is N hexadecimal, "..", N hexadecimal,
 587         * and optional space with octal mode.
 588         */
 589        const char *ptr, *eol;
 590        int len;
 591
 592        ptr = strchr(line, '.');
 593        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
 594                return 0;
 595        len = ptr - line;
 596        memcpy(patch->old_sha1_prefix, line, len);
 597        patch->old_sha1_prefix[len] = 0;
 598
 599        line = ptr + 2;
 600        ptr = strchr(line, ' ');
 601        eol = strchr(line, '\n');
 602
 603        if (!ptr || eol < ptr)
 604                ptr = eol;
 605        len = ptr - line;
 606
 607        if (40 < len)
 608                return 0;
 609        memcpy(patch->new_sha1_prefix, line, len);
 610        patch->new_sha1_prefix[len] = 0;
 611        if (*ptr == ' ')
 612                patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
 613        return 0;
 614}
 615
 616/*
 617 * This is normal for a diff that doesn't change anything: we'll fall through
 618 * into the next diff. Tell the parser to break out.
 619 */
 620static int gitdiff_unrecognized(const char *line, struct patch *patch)
 621{
 622        return -1;
 623}
 624
 625static const char *stop_at_slash(const char *line, int llen)
 626{
 627        int i;
 628
 629        for (i = 0; i < llen; i++) {
 630                int ch = line[i];
 631                if (ch == '/')
 632                        return line + i;
 633        }
 634        return NULL;
 635}
 636
 637/*
 638 * This is to extract the same name that appears on "diff --git"
 639 * line.  We do not find and return anything if it is a rename
 640 * patch, and it is OK because we will find the name elsewhere.
 641 * We need to reliably find name only when it is mode-change only,
 642 * creation or deletion of an empty file.  In any of these cases,
 643 * both sides are the same name under a/ and b/ respectively.
 644 */
 645static char *git_header_name(char *line, int llen)
 646{
 647        const char *name;
 648        const char *second = NULL;
 649        size_t len;
 650
 651        line += strlen("diff --git ");
 652        llen -= strlen("diff --git ");
 653
 654        if (*line == '"') {
 655                const char *cp;
 656                struct strbuf first;
 657                struct strbuf sp;
 658
 659                strbuf_init(&first, 0);
 660                strbuf_init(&sp, 0);
 661
 662                if (unquote_c_style(&first, line, &second))
 663                        goto free_and_fail1;
 664
 665                /* advance to the first slash */
 666                cp = stop_at_slash(first.buf, first.len);
 667                /* we do not accept absolute paths */
 668                if (!cp || cp == first.buf)
 669                        goto free_and_fail1;
 670                strbuf_remove(&first, 0, cp + 1 - first.buf);
 671
 672                /*
 673                 * second points at one past closing dq of name.
 674                 * find the second name.
 675                 */
 676                while ((second < line + llen) && isspace(*second))
 677                        second++;
 678
 679                if (line + llen <= second)
 680                        goto free_and_fail1;
 681                if (*second == '"') {
 682                        if (unquote_c_style(&sp, second, NULL))
 683                                goto free_and_fail1;
 684                        cp = stop_at_slash(sp.buf, sp.len);
 685                        if (!cp || cp == sp.buf)
 686                                goto free_and_fail1;
 687                        /* They must match, otherwise ignore */
 688                        if (strcmp(cp + 1, first.buf))
 689                                goto free_and_fail1;
 690                        strbuf_release(&sp);
 691                        return strbuf_detach(&first, NULL);
 692                }
 693
 694                /* unquoted second */
 695                cp = stop_at_slash(second, line + llen - second);
 696                if (!cp || cp == second)
 697                        goto free_and_fail1;
 698                cp++;
 699                if (line + llen - cp != first.len + 1 ||
 700                    memcmp(first.buf, cp, first.len))
 701                        goto free_and_fail1;
 702                return strbuf_detach(&first, NULL);
 703
 704        free_and_fail1:
 705                strbuf_release(&first);
 706                strbuf_release(&sp);
 707                return NULL;
 708        }
 709
 710        /* unquoted first name */
 711        name = stop_at_slash(line, llen);
 712        if (!name || name == line)
 713                return NULL;
 714        name++;
 715
 716        /*
 717         * since the first name is unquoted, a dq if exists must be
 718         * the beginning of the second name.
 719         */
 720        for (second = name; second < line + llen; second++) {
 721                if (*second == '"') {
 722                        struct strbuf sp;
 723                        const char *np;
 724
 725                        strbuf_init(&sp, 0);
 726                        if (unquote_c_style(&sp, second, NULL))
 727                                goto free_and_fail2;
 728
 729                        np = stop_at_slash(sp.buf, sp.len);
 730                        if (!np || np == sp.buf)
 731                                goto free_and_fail2;
 732                        np++;
 733
 734                        len = sp.buf + sp.len - np;
 735                        if (len < second - name &&
 736                            !strncmp(np, name, len) &&
 737                            isspace(name[len])) {
 738                                /* Good */
 739                                strbuf_remove(&sp, 0, np - sp.buf);
 740                                return strbuf_detach(&sp, NULL);
 741                        }
 742
 743                free_and_fail2:
 744                        strbuf_release(&sp);
 745                        return NULL;
 746                }
 747        }
 748
 749        /*
 750         * Accept a name only if it shows up twice, exactly the same
 751         * form.
 752         */
 753        for (len = 0 ; ; len++) {
 754                switch (name[len]) {
 755                default:
 756                        continue;
 757                case '\n':
 758                        return NULL;
 759                case '\t': case ' ':
 760                        second = name+len;
 761                        for (;;) {
 762                                char c = *second++;
 763                                if (c == '\n')
 764                                        return NULL;
 765                                if (c == '/')
 766                                        break;
 767                        }
 768                        if (second[len] == '\n' && !memcmp(name, second, len)) {
 769                                return xmemdupz(name, len);
 770                        }
 771                }
 772        }
 773}
 774
 775/* Verify that we recognize the lines following a git header */
 776static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
 777{
 778        unsigned long offset;
 779
 780        /* A git diff has explicit new/delete information, so we don't guess */
 781        patch->is_new = 0;
 782        patch->is_delete = 0;
 783
 784        /*
 785         * Some things may not have the old name in the
 786         * rest of the headers anywhere (pure mode changes,
 787         * or removing or adding empty files), so we get
 788         * the default name from the header.
 789         */
 790        patch->def_name = git_header_name(line, len);
 791
 792        line += len;
 793        size -= len;
 794        linenr++;
 795        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
 796                static const struct opentry {
 797                        const char *str;
 798                        int (*fn)(const char *, struct patch *);
 799                } optable[] = {
 800                        { "@@ -", gitdiff_hdrend },
 801                        { "--- ", gitdiff_oldname },
 802                        { "+++ ", gitdiff_newname },
 803                        { "old mode ", gitdiff_oldmode },
 804                        { "new mode ", gitdiff_newmode },
 805                        { "deleted file mode ", gitdiff_delete },
 806                        { "new file mode ", gitdiff_newfile },
 807                        { "copy from ", gitdiff_copysrc },
 808                        { "copy to ", gitdiff_copydst },
 809                        { "rename old ", gitdiff_renamesrc },
 810                        { "rename new ", gitdiff_renamedst },
 811                        { "rename from ", gitdiff_renamesrc },
 812                        { "rename to ", gitdiff_renamedst },
 813                        { "similarity index ", gitdiff_similarity },
 814                        { "dissimilarity index ", gitdiff_dissimilarity },
 815                        { "index ", gitdiff_index },
 816                        { "", gitdiff_unrecognized },
 817                };
 818                int i;
 819
 820                len = linelen(line, size);
 821                if (!len || line[len-1] != '\n')
 822                        break;
 823                for (i = 0; i < ARRAY_SIZE(optable); i++) {
 824                        const struct opentry *p = optable + i;
 825                        int oplen = strlen(p->str);
 826                        if (len < oplen || memcmp(p->str, line, oplen))
 827                                continue;
 828                        if (p->fn(line + oplen, patch) < 0)
 829                                return offset;
 830                        break;
 831                }
 832        }
 833
 834        return offset;
 835}
 836
 837static int parse_num(const char *line, unsigned long *p)
 838{
 839        char *ptr;
 840
 841        if (!isdigit(*line))
 842                return 0;
 843        *p = strtoul(line, &ptr, 10);
 844        return ptr - line;
 845}
 846
 847static int parse_range(const char *line, int len, int offset, const char *expect,
 848                       unsigned long *p1, unsigned long *p2)
 849{
 850        int digits, ex;
 851
 852        if (offset < 0 || offset >= len)
 853                return -1;
 854        line += offset;
 855        len -= offset;
 856
 857        digits = parse_num(line, p1);
 858        if (!digits)
 859                return -1;
 860
 861        offset += digits;
 862        line += digits;
 863        len -= digits;
 864
 865        *p2 = 1;
 866        if (*line == ',') {
 867                digits = parse_num(line+1, p2);
 868                if (!digits)
 869                        return -1;
 870
 871                offset += digits+1;
 872                line += digits+1;
 873                len -= digits+1;
 874        }
 875
 876        ex = strlen(expect);
 877        if (ex > len)
 878                return -1;
 879        if (memcmp(line, expect, ex))
 880                return -1;
 881
 882        return offset + ex;
 883}
 884
 885/*
 886 * Parse a unified diff fragment header of the
 887 * form "@@ -a,b +c,d @@"
 888 */
 889static int parse_fragment_header(char *line, int len, struct fragment *fragment)
 890{
 891        int offset;
 892
 893        if (!len || line[len-1] != '\n')
 894                return -1;
 895
 896        /* Figure out the number of lines in a fragment */
 897        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
 898        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
 899
 900        return offset;
 901}
 902
 903static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
 904{
 905        unsigned long offset, len;
 906
 907        patch->is_toplevel_relative = 0;
 908        patch->is_rename = patch->is_copy = 0;
 909        patch->is_new = patch->is_delete = -1;
 910        patch->old_mode = patch->new_mode = 0;
 911        patch->old_name = patch->new_name = NULL;
 912        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
 913                unsigned long nextlen;
 914
 915                len = linelen(line, size);
 916                if (!len)
 917                        break;
 918
 919                /* Testing this early allows us to take a few shortcuts.. */
 920                if (len < 6)
 921                        continue;
 922
 923                /*
 924                 * Make sure we don't find any unconnected patch fragments.
 925                 * That's a sign that we didn't find a header, and that a
 926                 * patch has become corrupted/broken up.
 927                 */
 928                if (!memcmp("@@ -", line, 4)) {
 929                        struct fragment dummy;
 930                        if (parse_fragment_header(line, len, &dummy) < 0)
 931                                continue;
 932                        die("patch fragment without header at line %d: %.*s",
 933                            linenr, (int)len-1, line);
 934                }
 935
 936                if (size < len + 6)
 937                        break;
 938
 939                /*
 940                 * Git patch? It might not have a real patch, just a rename
 941                 * or mode change, so we handle that specially
 942                 */
 943                if (!memcmp("diff --git ", line, 11)) {
 944                        int git_hdr_len = parse_git_header(line, len, size, patch);
 945                        if (git_hdr_len <= len)
 946                                continue;
 947                        if (!patch->old_name && !patch->new_name) {
 948                                if (!patch->def_name)
 949                                        die("git diff header lacks filename information (line %d)", linenr);
 950                                patch->old_name = patch->new_name = patch->def_name;
 951                        }
 952                        patch->is_toplevel_relative = 1;
 953                        *hdrsize = git_hdr_len;
 954                        return offset;
 955                }
 956
 957                /* --- followed by +++ ? */
 958                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
 959                        continue;
 960
 961                /*
 962                 * We only accept unified patches, so we want it to
 963                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
 964                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
 965                 */
 966                nextlen = linelen(line + len, size - len);
 967                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
 968                        continue;
 969
 970                /* Ok, we'll consider it a patch */
 971                parse_traditional_patch(line, line+len, patch);
 972                *hdrsize = len + nextlen;
 973                linenr += 2;
 974                return offset;
 975        }
 976        return -1;
 977}
 978
 979static void check_whitespace(const char *line, int len, unsigned ws_rule)
 980{
 981        char *err;
 982        unsigned result = check_and_emit_line(line + 1, len - 1, ws_rule,
 983            NULL, NULL, NULL, NULL);
 984        if (!result)
 985                return;
 986
 987        whitespace_error++;
 988        if (squelch_whitespace_errors &&
 989            squelch_whitespace_errors < whitespace_error)
 990                ;
 991        else {
 992                err = whitespace_error_string(result);
 993                fprintf(stderr, "%s:%d: %s.\n%.*s\n",
 994                     patch_input_file, linenr, err, len - 2, line + 1);
 995                free(err);
 996        }
 997}
 998
 999/*
1000 * Parse a unified diff. Note that this really needs to parse each
1001 * fragment separately, since the only way to know the difference
1002 * between a "---" that is part of a patch, and a "---" that starts
1003 * the next patch is to look at the line counts..
1004 */
1005static int parse_fragment(char *line, unsigned long size,
1006                          struct patch *patch, struct fragment *fragment)
1007{
1008        int added, deleted;
1009        int len = linelen(line, size), offset;
1010        unsigned long oldlines, newlines;
1011        unsigned long leading, trailing;
1012
1013        offset = parse_fragment_header(line, len, fragment);
1014        if (offset < 0)
1015                return -1;
1016        oldlines = fragment->oldlines;
1017        newlines = fragment->newlines;
1018        leading = 0;
1019        trailing = 0;
1020
1021        /* Parse the thing.. */
1022        line += len;
1023        size -= len;
1024        linenr++;
1025        added = deleted = 0;
1026        for (offset = len;
1027             0 < size;
1028             offset += len, size -= len, line += len, linenr++) {
1029                if (!oldlines && !newlines)
1030                        break;
1031                len = linelen(line, size);
1032                if (!len || line[len-1] != '\n')
1033                        return -1;
1034                switch (*line) {
1035                default:
1036                        return -1;
1037                case '\n': /* newer GNU diff, an empty context line */
1038                case ' ':
1039                        oldlines--;
1040                        newlines--;
1041                        if (!deleted && !added)
1042                                leading++;
1043                        trailing++;
1044                        break;
1045                case '-':
1046                        if (apply_in_reverse &&
1047                            ws_error_action != nowarn_ws_error)
1048                                check_whitespace(line, len, patch->ws_rule);
1049                        deleted++;
1050                        oldlines--;
1051                        trailing = 0;
1052                        break;
1053                case '+':
1054                        if (!apply_in_reverse &&
1055                            ws_error_action != nowarn_ws_error)
1056                                check_whitespace(line, len, patch->ws_rule);
1057                        added++;
1058                        newlines--;
1059                        trailing = 0;
1060                        break;
1061
1062                /*
1063                 * We allow "\ No newline at end of file". Depending
1064                 * on locale settings when the patch was produced we
1065                 * don't know what this line looks like. The only
1066                 * thing we do know is that it begins with "\ ".
1067                 * Checking for 12 is just for sanity check -- any
1068                 * l10n of "\ No newline..." is at least that long.
1069                 */
1070                case '\\':
1071                        if (len < 12 || memcmp(line, "\\ ", 2))
1072                                return -1;
1073                        break;
1074                }
1075        }
1076        if (oldlines || newlines)
1077                return -1;
1078        fragment->leading = leading;
1079        fragment->trailing = trailing;
1080
1081        /*
1082         * If a fragment ends with an incomplete line, we failed to include
1083         * it in the above loop because we hit oldlines == newlines == 0
1084         * before seeing it.
1085         */
1086        if (12 < size && !memcmp(line, "\\ ", 2))
1087                offset += linelen(line, size);
1088
1089        patch->lines_added += added;
1090        patch->lines_deleted += deleted;
1091
1092        if (0 < patch->is_new && oldlines)
1093                return error("new file depends on old contents");
1094        if (0 < patch->is_delete && newlines)
1095                return error("deleted file still has contents");
1096        return offset;
1097}
1098
1099static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1100{
1101        unsigned long offset = 0;
1102        unsigned long oldlines = 0, newlines = 0, context = 0;
1103        struct fragment **fragp = &patch->fragments;
1104
1105        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1106                struct fragment *fragment;
1107                int len;
1108
1109                fragment = xcalloc(1, sizeof(*fragment));
1110                len = parse_fragment(line, size, patch, fragment);
1111                if (len <= 0)
1112                        die("corrupt patch at line %d", linenr);
1113                fragment->patch = line;
1114                fragment->size = len;
1115                oldlines += fragment->oldlines;
1116                newlines += fragment->newlines;
1117                context += fragment->leading + fragment->trailing;
1118
1119                *fragp = fragment;
1120                fragp = &fragment->next;
1121
1122                offset += len;
1123                line += len;
1124                size -= len;
1125        }
1126
1127        /*
1128         * If something was removed (i.e. we have old-lines) it cannot
1129         * be creation, and if something was added it cannot be
1130         * deletion.  However, the reverse is not true; --unified=0
1131         * patches that only add are not necessarily creation even
1132         * though they do not have any old lines, and ones that only
1133         * delete are not necessarily deletion.
1134         *
1135         * Unfortunately, a real creation/deletion patch do _not_ have
1136         * any context line by definition, so we cannot safely tell it
1137         * apart with --unified=0 insanity.  At least if the patch has
1138         * more than one hunk it is not creation or deletion.
1139         */
1140        if (patch->is_new < 0 &&
1141            (oldlines || (patch->fragments && patch->fragments->next)))
1142                patch->is_new = 0;
1143        if (patch->is_delete < 0 &&
1144            (newlines || (patch->fragments && patch->fragments->next)))
1145                patch->is_delete = 0;
1146        if (!unidiff_zero || context) {
1147                /* If the user says the patch is not generated with
1148                 * --unified=0, or if we have seen context lines,
1149                 * then not having oldlines means the patch is creation,
1150                 * and not having newlines means the patch is deletion.
1151                 */
1152                if (patch->is_new < 0 && !oldlines) {
1153                        patch->is_new = 1;
1154                        patch->old_name = NULL;
1155                }
1156                if (patch->is_delete < 0 && !newlines) {
1157                        patch->is_delete = 1;
1158                        patch->new_name = NULL;
1159                }
1160        }
1161
1162        if (0 < patch->is_new && oldlines)
1163                die("new file %s depends on old contents", patch->new_name);
1164        if (0 < patch->is_delete && newlines)
1165                die("deleted file %s still has contents", patch->old_name);
1166        if (!patch->is_delete && !newlines && context)
1167                fprintf(stderr, "** warning: file %s becomes empty but "
1168                        "is not deleted\n", patch->new_name);
1169
1170        return offset;
1171}
1172
1173static inline int metadata_changes(struct patch *patch)
1174{
1175        return  patch->is_rename > 0 ||
1176                patch->is_copy > 0 ||
1177                patch->is_new > 0 ||
1178                patch->is_delete ||
1179                (patch->old_mode && patch->new_mode &&
1180                 patch->old_mode != patch->new_mode);
1181}
1182
1183static char *inflate_it(const void *data, unsigned long size,
1184                        unsigned long inflated_size)
1185{
1186        z_stream stream;
1187        void *out;
1188        int st;
1189
1190        memset(&stream, 0, sizeof(stream));
1191
1192        stream.next_in = (unsigned char *)data;
1193        stream.avail_in = size;
1194        stream.next_out = out = xmalloc(inflated_size);
1195        stream.avail_out = inflated_size;
1196        inflateInit(&stream);
1197        st = inflate(&stream, Z_FINISH);
1198        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1199                free(out);
1200                return NULL;
1201        }
1202        return out;
1203}
1204
1205static struct fragment *parse_binary_hunk(char **buf_p,
1206                                          unsigned long *sz_p,
1207                                          int *status_p,
1208                                          int *used_p)
1209{
1210        /*
1211         * Expect a line that begins with binary patch method ("literal"
1212         * or "delta"), followed by the length of data before deflating.
1213         * a sequence of 'length-byte' followed by base-85 encoded data
1214         * should follow, terminated by a newline.
1215         *
1216         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1217         * and we would limit the patch line to 66 characters,
1218         * so one line can fit up to 13 groups that would decode
1219         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1220         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1221         */
1222        int llen, used;
1223        unsigned long size = *sz_p;
1224        char *buffer = *buf_p;
1225        int patch_method;
1226        unsigned long origlen;
1227        char *data = NULL;
1228        int hunk_size = 0;
1229        struct fragment *frag;
1230
1231        llen = linelen(buffer, size);
1232        used = llen;
1233
1234        *status_p = 0;
1235
1236        if (!prefixcmp(buffer, "delta ")) {
1237                patch_method = BINARY_DELTA_DEFLATED;
1238                origlen = strtoul(buffer + 6, NULL, 10);
1239        }
1240        else if (!prefixcmp(buffer, "literal ")) {
1241                patch_method = BINARY_LITERAL_DEFLATED;
1242                origlen = strtoul(buffer + 8, NULL, 10);
1243        }
1244        else
1245                return NULL;
1246
1247        linenr++;
1248        buffer += llen;
1249        while (1) {
1250                int byte_length, max_byte_length, newsize;
1251                llen = linelen(buffer, size);
1252                used += llen;
1253                linenr++;
1254                if (llen == 1) {
1255                        /* consume the blank line */
1256                        buffer++;
1257                        size--;
1258                        break;
1259                }
1260                /*
1261                 * Minimum line is "A00000\n" which is 7-byte long,
1262                 * and the line length must be multiple of 5 plus 2.
1263                 */
1264                if ((llen < 7) || (llen-2) % 5)
1265                        goto corrupt;
1266                max_byte_length = (llen - 2) / 5 * 4;
1267                byte_length = *buffer;
1268                if ('A' <= byte_length && byte_length <= 'Z')
1269                        byte_length = byte_length - 'A' + 1;
1270                else if ('a' <= byte_length && byte_length <= 'z')
1271                        byte_length = byte_length - 'a' + 27;
1272                else
1273                        goto corrupt;
1274                /* if the input length was not multiple of 4, we would
1275                 * have filler at the end but the filler should never
1276                 * exceed 3 bytes
1277                 */
1278                if (max_byte_length < byte_length ||
1279                    byte_length <= max_byte_length - 4)
1280                        goto corrupt;
1281                newsize = hunk_size + byte_length;
1282                data = xrealloc(data, newsize);
1283                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1284                        goto corrupt;
1285                hunk_size = newsize;
1286                buffer += llen;
1287                size -= llen;
1288        }
1289
1290        frag = xcalloc(1, sizeof(*frag));
1291        frag->patch = inflate_it(data, hunk_size, origlen);
1292        if (!frag->patch)
1293                goto corrupt;
1294        free(data);
1295        frag->size = origlen;
1296        *buf_p = buffer;
1297        *sz_p = size;
1298        *used_p = used;
1299        frag->binary_patch_method = patch_method;
1300        return frag;
1301
1302 corrupt:
1303        free(data);
1304        *status_p = -1;
1305        error("corrupt binary patch at line %d: %.*s",
1306              linenr-1, llen-1, buffer);
1307        return NULL;
1308}
1309
1310static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1311{
1312        /*
1313         * We have read "GIT binary patch\n"; what follows is a line
1314         * that says the patch method (currently, either "literal" or
1315         * "delta") and the length of data before deflating; a
1316         * sequence of 'length-byte' followed by base-85 encoded data
1317         * follows.
1318         *
1319         * When a binary patch is reversible, there is another binary
1320         * hunk in the same format, starting with patch method (either
1321         * "literal" or "delta") with the length of data, and a sequence
1322         * of length-byte + base-85 encoded data, terminated with another
1323         * empty line.  This data, when applied to the postimage, produces
1324         * the preimage.
1325         */
1326        struct fragment *forward;
1327        struct fragment *reverse;
1328        int status;
1329        int used, used_1;
1330
1331        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1332        if (!forward && !status)
1333                /* there has to be one hunk (forward hunk) */
1334                return error("unrecognized binary patch at line %d", linenr-1);
1335        if (status)
1336                /* otherwise we already gave an error message */
1337                return status;
1338
1339        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1340        if (reverse)
1341                used += used_1;
1342        else if (status) {
1343                /*
1344                 * Not having reverse hunk is not an error, but having
1345                 * a corrupt reverse hunk is.
1346                 */
1347                free((void*) forward->patch);
1348                free(forward);
1349                return status;
1350        }
1351        forward->next = reverse;
1352        patch->fragments = forward;
1353        patch->is_binary = 1;
1354        return used;
1355}
1356
1357static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1358{
1359        int hdrsize, patchsize;
1360        int offset = find_header(buffer, size, &hdrsize, patch);
1361
1362        if (offset < 0)
1363                return offset;
1364
1365        patch->ws_rule = whitespace_rule(patch->new_name
1366                                         ? patch->new_name
1367                                         : patch->old_name);
1368
1369        patchsize = parse_single_patch(buffer + offset + hdrsize,
1370                                       size - offset - hdrsize, patch);
1371
1372        if (!patchsize) {
1373                static const char *binhdr[] = {
1374                        "Binary files ",
1375                        "Files ",
1376                        NULL,
1377                };
1378                static const char git_binary[] = "GIT binary patch\n";
1379                int i;
1380                int hd = hdrsize + offset;
1381                unsigned long llen = linelen(buffer + hd, size - hd);
1382
1383                if (llen == sizeof(git_binary) - 1 &&
1384                    !memcmp(git_binary, buffer + hd, llen)) {
1385                        int used;
1386                        linenr++;
1387                        used = parse_binary(buffer + hd + llen,
1388                                            size - hd - llen, patch);
1389                        if (used)
1390                                patchsize = used + llen;
1391                        else
1392                                patchsize = 0;
1393                }
1394                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1395                        for (i = 0; binhdr[i]; i++) {
1396                                int len = strlen(binhdr[i]);
1397                                if (len < size - hd &&
1398                                    !memcmp(binhdr[i], buffer + hd, len)) {
1399                                        linenr++;
1400                                        patch->is_binary = 1;
1401                                        patchsize = llen;
1402                                        break;
1403                                }
1404                        }
1405                }
1406
1407                /* Empty patch cannot be applied if it is a text patch
1408                 * without metadata change.  A binary patch appears
1409                 * empty to us here.
1410                 */
1411                if ((apply || check) &&
1412                    (!patch->is_binary && !metadata_changes(patch)))
1413                        die("patch with only garbage at line %d", linenr);
1414        }
1415
1416        return offset + hdrsize + patchsize;
1417}
1418
1419#define swap(a,b) myswap((a),(b),sizeof(a))
1420
1421#define myswap(a, b, size) do {         \
1422        unsigned char mytmp[size];      \
1423        memcpy(mytmp, &a, size);                \
1424        memcpy(&a, &b, size);           \
1425        memcpy(&b, mytmp, size);                \
1426} while (0)
1427
1428static void reverse_patches(struct patch *p)
1429{
1430        for (; p; p = p->next) {
1431                struct fragment *frag = p->fragments;
1432
1433                swap(p->new_name, p->old_name);
1434                swap(p->new_mode, p->old_mode);
1435                swap(p->is_new, p->is_delete);
1436                swap(p->lines_added, p->lines_deleted);
1437                swap(p->old_sha1_prefix, p->new_sha1_prefix);
1438
1439                for (; frag; frag = frag->next) {
1440                        swap(frag->newpos, frag->oldpos);
1441                        swap(frag->newlines, frag->oldlines);
1442                }
1443        }
1444}
1445
1446static const char pluses[] =
1447"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1448static const char minuses[]=
1449"----------------------------------------------------------------------";
1450
1451static void show_stats(struct patch *patch)
1452{
1453        struct strbuf qname;
1454        char *cp = patch->new_name ? patch->new_name : patch->old_name;
1455        int max, add, del;
1456
1457        strbuf_init(&qname, 0);
1458        quote_c_style(cp, &qname, NULL, 0);
1459
1460        /*
1461         * "scale" the filename
1462         */
1463        max = max_len;
1464        if (max > 50)
1465                max = 50;
1466
1467        if (qname.len > max) {
1468                cp = strchr(qname.buf + qname.len + 3 - max, '/');
1469                if (!cp)
1470                        cp = qname.buf + qname.len + 3 - max;
1471                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1472        }
1473
1474        if (patch->is_binary) {
1475                printf(" %-*s |  Bin\n", max, qname.buf);
1476                strbuf_release(&qname);
1477                return;
1478        }
1479
1480        printf(" %-*s |", max, qname.buf);
1481        strbuf_release(&qname);
1482
1483        /*
1484         * scale the add/delete
1485         */
1486        max = max + max_change > 70 ? 70 - max : max_change;
1487        add = patch->lines_added;
1488        del = patch->lines_deleted;
1489
1490        if (max_change > 0) {
1491                int total = ((add + del) * max + max_change / 2) / max_change;
1492                add = (add * max + max_change / 2) / max_change;
1493                del = total - add;
1494        }
1495        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1496                add, pluses, del, minuses);
1497}
1498
1499static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1500{
1501        switch (st->st_mode & S_IFMT) {
1502        case S_IFLNK:
1503                strbuf_grow(buf, st->st_size);
1504                if (readlink(path, buf->buf, st->st_size) != st->st_size)
1505                        return -1;
1506                strbuf_setlen(buf, st->st_size);
1507                return 0;
1508        case S_IFREG:
1509                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1510                        return error("unable to open or read %s", path);
1511                convert_to_git(path, buf->buf, buf->len, buf);
1512                return 0;
1513        default:
1514                return -1;
1515        }
1516}
1517
1518static int copy_wsfix(char *output, const char *patch, int plen,
1519                      unsigned ws_rule, int count_error)
1520{
1521        /*
1522         * plen is number of bytes to be copied from patch, starting
1523         * at patch.  Typically patch[plen-1] is '\n', unless this is
1524         * the incomplete last line.
1525         */
1526        int i;
1527        int add_nl_to_tail = 0;
1528        int fixed = 0;
1529        int last_tab_in_indent = -1;
1530        int last_space_in_indent = -1;
1531        int need_fix_leading_space = 0;
1532        char *buf;
1533
1534        /*
1535         * Strip trailing whitespace
1536         */
1537        if ((ws_rule & WS_TRAILING_SPACE) &&
1538            (2 < plen && isspace(patch[plen-2]))) {
1539                if (patch[plen-1] == '\n')
1540                        add_nl_to_tail = 1;
1541                plen--;
1542                while (0 < plen && isspace(patch[plen-1]))
1543                        plen--;
1544                fixed = 1;
1545        }
1546
1547        /*
1548         * Check leading whitespaces (indent)
1549         */
1550        for (i = 0; i < plen; i++) {
1551                char ch = patch[i];
1552                if (ch == '\t') {
1553                        last_tab_in_indent = i;
1554                        if ((ws_rule & WS_SPACE_BEFORE_TAB) &&
1555                            0 <= last_space_in_indent)
1556                            need_fix_leading_space = 1;
1557                } else if (ch == ' ') {
1558                        last_space_in_indent = i;
1559                        if ((ws_rule & WS_INDENT_WITH_NON_TAB) &&
1560                            8 <= i - last_tab_in_indent)
1561                                need_fix_leading_space = 1;
1562                } else
1563                        break;
1564        }
1565
1566        buf = output;
1567        if (need_fix_leading_space) {
1568                /* Process indent ourselves */
1569                int consecutive_spaces = 0;
1570                int last = last_tab_in_indent + 1;
1571
1572                if (ws_rule & WS_INDENT_WITH_NON_TAB) {
1573                        /* have "last" point at one past the indent */
1574                        if (last_tab_in_indent < last_space_in_indent)
1575                                last = last_space_in_indent + 1;
1576                        else
1577                                last = last_tab_in_indent + 1;
1578                }
1579
1580                /*
1581                 * between patch[0..last-1], strip the funny spaces,
1582                 * updating them to tab as needed.
1583                 */
1584                for (i = 0; i < last; i++) {
1585                        char ch = patch[i];
1586                        if (ch != ' ') {
1587                                consecutive_spaces = 0;
1588                                *output++ = ch;
1589                        } else {
1590                                consecutive_spaces++;
1591                                if (consecutive_spaces == 8) {
1592                                        *output++ = '\t';
1593                                        consecutive_spaces = 0;
1594                                }
1595                        }
1596                }
1597                while (0 < consecutive_spaces--)
1598                        *output++ = ' ';
1599                plen -= last;
1600                patch += last;
1601                fixed = 1;
1602        }
1603
1604        memcpy(output, patch, plen);
1605        if (add_nl_to_tail)
1606                output[plen++] = '\n';
1607        if (fixed && count_error)
1608                applied_after_fixing_ws++;
1609        return output + plen - buf;
1610}
1611
1612static void update_pre_post_images(struct image *preimage,
1613                                   struct image *postimage,
1614                                   char *buf,
1615                                   size_t len)
1616{
1617        int i, ctx;
1618        char *new, *old, *fixed;
1619        struct image fixed_preimage;
1620
1621        /*
1622         * Update the preimage with whitespace fixes.  Note that we
1623         * are not losing preimage->buf -- apply_one_fragment() will
1624         * free "oldlines".
1625         */
1626        prepare_image(&fixed_preimage, buf, len, 1);
1627        assert(fixed_preimage.nr == preimage->nr);
1628        for (i = 0; i < preimage->nr; i++)
1629                fixed_preimage.line[i].flag = preimage->line[i].flag;
1630        free(preimage->line_allocated);
1631        *preimage = fixed_preimage;
1632
1633        /*
1634         * Adjust the common context lines in postimage, in place.
1635         * This is possible because whitespace fixing does not make
1636         * the string grow.
1637         */
1638        new = old = postimage->buf;
1639        fixed = preimage->buf;
1640        for (i = ctx = 0; i < postimage->nr; i++) {
1641                size_t len = postimage->line[i].len;
1642                if (!(postimage->line[i].flag & LINE_COMMON)) {
1643                        /* an added line -- no counterparts in preimage */
1644                        memmove(new, old, len);
1645                        old += len;
1646                        new += len;
1647                        continue;
1648                }
1649
1650                /* a common context -- skip it in the original postimage */
1651                old += len;
1652
1653                /* and find the corresponding one in the fixed preimage */
1654                while (ctx < preimage->nr &&
1655                       !(preimage->line[ctx].flag & LINE_COMMON)) {
1656                        fixed += preimage->line[ctx].len;
1657                        ctx++;
1658                }
1659                if (preimage->nr <= ctx)
1660                        die("oops");
1661
1662                /* and copy it in, while fixing the line length */
1663                len = preimage->line[ctx].len;
1664                memcpy(new, fixed, len);
1665                new += len;
1666                fixed += len;
1667                postimage->line[i].len = len;
1668                ctx++;
1669        }
1670
1671        /* Fix the length of the whole thing */
1672        postimage->len = new - postimage->buf;
1673}
1674
1675static int match_fragment(struct image *img,
1676                          struct image *preimage,
1677                          struct image *postimage,
1678                          unsigned long try,
1679                          int try_lno,
1680                          unsigned ws_rule,
1681                          int match_beginning, int match_end)
1682{
1683        int i;
1684        char *fixed_buf, *buf, *orig, *target;
1685
1686        if (preimage->nr + try_lno > img->nr)
1687                return 0;
1688
1689        if (match_beginning && try_lno)
1690                return 0;
1691
1692        if (match_end && preimage->nr + try_lno != img->nr)
1693                return 0;
1694
1695        /* Quick hash check */
1696        for (i = 0; i < preimage->nr; i++)
1697                if (preimage->line[i].hash != img->line[try_lno + i].hash)
1698                        return 0;
1699
1700        /*
1701         * Do we have an exact match?  If we were told to match
1702         * at the end, size must be exactly at try+fragsize,
1703         * otherwise try+fragsize must be still within the preimage,
1704         * and either case, the old piece should match the preimage
1705         * exactly.
1706         */
1707        if ((match_end
1708             ? (try + preimage->len == img->len)
1709             : (try + preimage->len <= img->len)) &&
1710            !memcmp(img->buf + try, preimage->buf, preimage->len))
1711                return 1;
1712
1713        if (ws_error_action != correct_ws_error)
1714                return 0;
1715
1716        /*
1717         * The hunk does not apply byte-by-byte, but the hash says
1718         * it might with whitespace fuzz.
1719         */
1720        fixed_buf = xmalloc(preimage->len + 1);
1721        buf = fixed_buf;
1722        orig = preimage->buf;
1723        target = img->buf + try;
1724        for (i = 0; i < preimage->nr; i++) {
1725                size_t fixlen; /* length after fixing the preimage */
1726                size_t oldlen = preimage->line[i].len;
1727                size_t tgtlen = img->line[try_lno + i].len;
1728                size_t tgtfixlen; /* length after fixing the target line */
1729                char tgtfixbuf[1024], *tgtfix;
1730                int match;
1731
1732                /* Try fixing the line in the preimage */
1733                fixlen = copy_wsfix(buf, orig, oldlen, ws_rule, 0);
1734
1735                /* Try fixing the line in the target */
1736                if (sizeof(tgtfixbuf) < tgtlen)
1737                        tgtfix = tgtfixbuf;
1738                else
1739                        tgtfix = xmalloc(tgtlen);
1740                tgtfixlen = copy_wsfix(tgtfix, target, tgtlen, ws_rule, 0);
1741
1742                /*
1743                 * If they match, either the preimage was based on
1744                 * a version before our tree fixed whitespace breakage,
1745                 * or we are lacking a whitespace-fix patch the tree
1746                 * the preimage was based on already had (i.e. target
1747                 * has whitespace breakage, the preimage doesn't).
1748                 * In either case, we are fixing the whitespace breakages
1749                 * so we might as well take the fix together with their
1750                 * real change.
1751                 */
1752                match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1753
1754                if (tgtfix != tgtfixbuf)
1755                        free(tgtfix);
1756                if (!match)
1757                        goto unmatch_exit;
1758
1759                orig += oldlen;
1760                buf += fixlen;
1761                target += tgtlen;
1762        }
1763
1764        /*
1765         * Yes, the preimage is based on an older version that still
1766         * has whitespace breakages unfixed, and fixing them makes the
1767         * hunk match.  Update the context lines in the postimage.
1768         */
1769        update_pre_post_images(preimage, postimage,
1770                               fixed_buf, buf - fixed_buf);
1771        return 1;
1772
1773 unmatch_exit:
1774        free(fixed_buf);
1775        return 0;
1776}
1777
1778static int find_pos(struct image *img,
1779                    struct image *preimage,
1780                    struct image *postimage,
1781                    int line,
1782                    unsigned ws_rule,
1783                    int match_beginning, int match_end)
1784{
1785        int i;
1786        unsigned long backwards, forwards, try;
1787        int backwards_lno, forwards_lno, try_lno;
1788
1789        if (preimage->nr > img->nr)
1790                return -1;
1791
1792        /*
1793         * If match_begining or match_end is specified, there is no
1794         * point starting from a wrong line that will never match and
1795         * wander around and wait for a match at the specified end.
1796         */
1797        if (match_beginning)
1798                line = 0;
1799        else if (match_end)
1800                line = img->nr - preimage->nr;
1801
1802        try = 0;
1803        for (i = 0; i < line; i++)
1804                try += img->line[i].len;
1805
1806        /*
1807         * There's probably some smart way to do this, but I'll leave
1808         * that to the smart and beautiful people. I'm simple and stupid.
1809         */
1810        backwards = try;
1811        backwards_lno = line;
1812        forwards = try;
1813        forwards_lno = line;
1814        try_lno = line;
1815
1816        for (i = 0; ; i++) {
1817                if (match_fragment(img, preimage, postimage,
1818                                   try, try_lno, ws_rule,
1819                                   match_beginning, match_end))
1820                        return try_lno;
1821
1822        again:
1823                if (backwards_lno == 0 && forwards_lno == img->nr)
1824                        break;
1825
1826                if (i & 1) {
1827                        if (backwards_lno == 0) {
1828                                i++;
1829                                goto again;
1830                        }
1831                        backwards_lno--;
1832                        backwards -= img->line[backwards_lno].len;
1833                        try = backwards;
1834                        try_lno = backwards_lno;
1835                } else {
1836                        if (forwards_lno == img->nr) {
1837                                i++;
1838                                goto again;
1839                        }
1840                        forwards += img->line[forwards_lno].len;
1841                        forwards_lno++;
1842                        try = forwards;
1843                        try_lno = forwards_lno;
1844                }
1845
1846        }
1847        return -1;
1848}
1849
1850static void remove_first_line(struct image *img)
1851{
1852        img->buf += img->line[0].len;
1853        img->len -= img->line[0].len;
1854        img->line++;
1855        img->nr--;
1856}
1857
1858static void remove_last_line(struct image *img)
1859{
1860        img->len -= img->line[--img->nr].len;
1861}
1862
1863static void update_image(struct image *img,
1864                         int applied_pos,
1865                         struct image *preimage,
1866                         struct image *postimage)
1867{
1868        /*
1869         * remove the copy of preimage at offset in img
1870         * and replace it with postimage
1871         */
1872        int i, nr;
1873        size_t remove_count, insert_count, applied_at = 0;
1874        char *result;
1875
1876        for (i = 0; i < applied_pos; i++)
1877                applied_at += img->line[i].len;
1878
1879        remove_count = 0;
1880        for (i = 0; i < preimage->nr; i++)
1881                remove_count += img->line[applied_pos + i].len;
1882        insert_count = postimage->len;
1883
1884        /* Adjust the contents */
1885        result = xmalloc(img->len + insert_count - remove_count + 1);
1886        memcpy(result, img->buf, applied_at);
1887        memcpy(result + applied_at, postimage->buf, postimage->len);
1888        memcpy(result + applied_at + postimage->len,
1889               img->buf + (applied_at + remove_count),
1890               img->len - (applied_at + remove_count));
1891        free(img->buf);
1892        img->buf = result;
1893        img->len += insert_count - remove_count;
1894        result[img->len] = '\0';
1895
1896        /* Adjust the line table */
1897        nr = img->nr + postimage->nr - preimage->nr;
1898        if (preimage->nr < postimage->nr) {
1899                /*
1900                 * NOTE: this knows that we never call remove_first_line()
1901                 * on anything other than pre/post image.
1902                 */
1903                img->line = xrealloc(img->line, nr * sizeof(*img->line));
1904                img->line_allocated = img->line;
1905        }
1906        if (preimage->nr != postimage->nr)
1907                memmove(img->line + applied_pos + postimage->nr,
1908                        img->line + applied_pos + preimage->nr,
1909                        (img->nr - (applied_pos + preimage->nr)) *
1910                        sizeof(*img->line));
1911        memcpy(img->line + applied_pos,
1912               postimage->line,
1913               postimage->nr * sizeof(*img->line));
1914        img->nr = nr;
1915}
1916
1917static int apply_one_fragment(struct image *img, struct fragment *frag,
1918                              int inaccurate_eof, unsigned ws_rule)
1919{
1920        int match_beginning, match_end;
1921        const char *patch = frag->patch;
1922        int size = frag->size;
1923        char *old, *new, *oldlines, *newlines;
1924        int new_blank_lines_at_end = 0;
1925        unsigned long leading, trailing;
1926        int pos, applied_pos;
1927        struct image preimage;
1928        struct image postimage;
1929
1930        memset(&preimage, 0, sizeof(preimage));
1931        memset(&postimage, 0, sizeof(postimage));
1932        oldlines = xmalloc(size);
1933        newlines = xmalloc(size);
1934
1935        old = oldlines;
1936        new = newlines;
1937        while (size > 0) {
1938                char first;
1939                int len = linelen(patch, size);
1940                int plen, added;
1941                int added_blank_line = 0;
1942
1943                if (!len)
1944                        break;
1945
1946                /*
1947                 * "plen" is how much of the line we should use for
1948                 * the actual patch data. Normally we just remove the
1949                 * first character on the line, but if the line is
1950                 * followed by "\ No newline", then we also remove the
1951                 * last one (which is the newline, of course).
1952                 */
1953                plen = len - 1;
1954                if (len < size && patch[len] == '\\')
1955                        plen--;
1956                first = *patch;
1957                if (apply_in_reverse) {
1958                        if (first == '-')
1959                                first = '+';
1960                        else if (first == '+')
1961                                first = '-';
1962                }
1963
1964                switch (first) {
1965                case '\n':
1966                        /* Newer GNU diff, empty context line */
1967                        if (plen < 0)
1968                                /* ... followed by '\No newline'; nothing */
1969                                break;
1970                        *old++ = '\n';
1971                        *new++ = '\n';
1972                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
1973                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
1974                        break;
1975                case ' ':
1976                case '-':
1977                        memcpy(old, patch + 1, plen);
1978                        add_line_info(&preimage, old, plen,
1979                                      (first == ' ' ? LINE_COMMON : 0));
1980                        old += plen;
1981                        if (first == '-')
1982                                break;
1983                /* Fall-through for ' ' */
1984                case '+':
1985                        /* --no-add does not add new lines */
1986                        if (first == '+' && no_add)
1987                                break;
1988
1989                        if (first != '+' ||
1990                            !whitespace_error ||
1991                            ws_error_action != correct_ws_error) {
1992                                memcpy(new, patch + 1, plen);
1993                                added = plen;
1994                        }
1995                        else {
1996                                added = copy_wsfix(new, patch + 1, plen,
1997                                                   ws_rule, 1);
1998                        }
1999                        add_line_info(&postimage, new, added,
2000                                      (first == '+' ? 0 : LINE_COMMON));
2001                        new += added;
2002                        if (first == '+' &&
2003                            added == 1 && new[-1] == '\n')
2004                                added_blank_line = 1;
2005                        break;
2006                case '@': case '\\':
2007                        /* Ignore it, we already handled it */
2008                        break;
2009                default:
2010                        if (apply_verbosely)
2011                                error("invalid start of line: '%c'", first);
2012                        return -1;
2013                }
2014                if (added_blank_line)
2015                        new_blank_lines_at_end++;
2016                else
2017                        new_blank_lines_at_end = 0;
2018                patch += len;
2019                size -= len;
2020        }
2021        if (inaccurate_eof &&
2022            old > oldlines && old[-1] == '\n' &&
2023            new > newlines && new[-1] == '\n') {
2024                old--;
2025                new--;
2026        }
2027
2028        leading = frag->leading;
2029        trailing = frag->trailing;
2030
2031        /*
2032         * If we don't have any leading/trailing data in the patch,
2033         * we want it to match at the beginning/end of the file.
2034         *
2035         * But that would break if the patch is generated with
2036         * --unified=0; sane people wouldn't do that to cause us
2037         * trouble, but we try to please not so sane ones as well.
2038         */
2039        if (unidiff_zero) {
2040                match_beginning = (!leading && !frag->oldpos);
2041                match_end = 0;
2042        }
2043        else {
2044                match_beginning = !leading && (frag->oldpos == 1);
2045                match_end = !trailing;
2046        }
2047
2048        pos = frag->newpos ? (frag->newpos - 1) : 0;
2049        preimage.buf = oldlines;
2050        preimage.len = old - oldlines;
2051        postimage.buf = newlines;
2052        postimage.len = new - newlines;
2053        preimage.line = preimage.line_allocated;
2054        postimage.line = postimage.line_allocated;
2055
2056        for (;;) {
2057
2058                applied_pos = find_pos(img, &preimage, &postimage, pos,
2059                                       ws_rule, match_beginning, match_end);
2060
2061                if (applied_pos >= 0)
2062                        break;
2063
2064                /* Am I at my context limits? */
2065                if ((leading <= p_context) && (trailing <= p_context))
2066                        break;
2067                if (match_beginning || match_end) {
2068                        match_beginning = match_end = 0;
2069                        continue;
2070                }
2071
2072                /*
2073                 * Reduce the number of context lines; reduce both
2074                 * leading and trailing if they are equal otherwise
2075                 * just reduce the larger context.
2076                 */
2077                if (leading >= trailing) {
2078                        remove_first_line(&preimage);
2079                        remove_first_line(&postimage);
2080                        pos--;
2081                        leading--;
2082                }
2083                if (trailing > leading) {
2084                        remove_last_line(&preimage);
2085                        remove_last_line(&postimage);
2086                        trailing--;
2087                }
2088        }
2089
2090        if (applied_pos >= 0) {
2091                if (ws_error_action == correct_ws_error &&
2092                    new_blank_lines_at_end &&
2093                    postimage.nr + applied_pos == img->nr) {
2094                        /*
2095                         * If the patch application adds blank lines
2096                         * at the end, and if the patch applies at the
2097                         * end of the image, remove those added blank
2098                         * lines.
2099                         */
2100                        while (new_blank_lines_at_end--)
2101                                remove_last_line(&postimage);
2102                }
2103
2104                /*
2105                 * Warn if it was necessary to reduce the number
2106                 * of context lines.
2107                 */
2108                if ((leading != frag->leading) ||
2109                    (trailing != frag->trailing))
2110                        fprintf(stderr, "Context reduced to (%ld/%ld)"
2111                                " to apply fragment at %d\n",
2112                                leading, trailing, applied_pos+1);
2113                update_image(img, applied_pos, &preimage, &postimage);
2114        } else {
2115                if (apply_verbosely)
2116                        error("while searching for:\n%.*s",
2117                              (int)(old - oldlines), oldlines);
2118        }
2119
2120        free(oldlines);
2121        free(newlines);
2122        free(preimage.line_allocated);
2123        free(postimage.line_allocated);
2124
2125        return (applied_pos < 0);
2126}
2127
2128static int apply_binary_fragment(struct image *img, struct patch *patch)
2129{
2130        struct fragment *fragment = patch->fragments;
2131        unsigned long len;
2132        void *dst;
2133
2134        /* Binary patch is irreversible without the optional second hunk */
2135        if (apply_in_reverse) {
2136                if (!fragment->next)
2137                        return error("cannot reverse-apply a binary patch "
2138                                     "without the reverse hunk to '%s'",
2139                                     patch->new_name
2140                                     ? patch->new_name : patch->old_name);
2141                fragment = fragment->next;
2142        }
2143        switch (fragment->binary_patch_method) {
2144        case BINARY_DELTA_DEFLATED:
2145                dst = patch_delta(img->buf, img->len, fragment->patch,
2146                                  fragment->size, &len);
2147                if (!dst)
2148                        return -1;
2149                clear_image(img);
2150                img->buf = dst;
2151                img->len = len;
2152                return 0;
2153        case BINARY_LITERAL_DEFLATED:
2154                clear_image(img);
2155                img->len = fragment->size;
2156                img->buf = xmalloc(img->len+1);
2157                memcpy(img->buf, fragment->patch, img->len);
2158                img->buf[img->len] = '\0';
2159                return 0;
2160        }
2161        return -1;
2162}
2163
2164static int apply_binary(struct image *img, struct patch *patch)
2165{
2166        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2167        unsigned char sha1[20];
2168
2169        /*
2170         * For safety, we require patch index line to contain
2171         * full 40-byte textual SHA1 for old and new, at least for now.
2172         */
2173        if (strlen(patch->old_sha1_prefix) != 40 ||
2174            strlen(patch->new_sha1_prefix) != 40 ||
2175            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2176            get_sha1_hex(patch->new_sha1_prefix, sha1))
2177                return error("cannot apply binary patch to '%s' "
2178                             "without full index line", name);
2179
2180        if (patch->old_name) {
2181                /*
2182                 * See if the old one matches what the patch
2183                 * applies to.
2184                 */
2185                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2186                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2187                        return error("the patch applies to '%s' (%s), "
2188                                     "which does not match the "
2189                                     "current contents.",
2190                                     name, sha1_to_hex(sha1));
2191        }
2192        else {
2193                /* Otherwise, the old one must be empty. */
2194                if (img->len)
2195                        return error("the patch applies to an empty "
2196                                     "'%s' but it is not empty", name);
2197        }
2198
2199        get_sha1_hex(patch->new_sha1_prefix, sha1);
2200        if (is_null_sha1(sha1)) {
2201                clear_image(img);
2202                return 0; /* deletion patch */
2203        }
2204
2205        if (has_sha1_file(sha1)) {
2206                /* We already have the postimage */
2207                enum object_type type;
2208                unsigned long size;
2209                char *result;
2210
2211                result = read_sha1_file(sha1, &type, &size);
2212                if (!result)
2213                        return error("the necessary postimage %s for "
2214                                     "'%s' cannot be read",
2215                                     patch->new_sha1_prefix, name);
2216                clear_image(img);
2217                img->buf = result;
2218                img->len = size;
2219        } else {
2220                /*
2221                 * We have verified buf matches the preimage;
2222                 * apply the patch data to it, which is stored
2223                 * in the patch->fragments->{patch,size}.
2224                 */
2225                if (apply_binary_fragment(img, patch))
2226                        return error("binary patch does not apply to '%s'",
2227                                     name);
2228
2229                /* verify that the result matches */
2230                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2231                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2232                        return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2233                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2234        }
2235
2236        return 0;
2237}
2238
2239static int apply_fragments(struct image *img, struct patch *patch)
2240{
2241        struct fragment *frag = patch->fragments;
2242        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2243        unsigned ws_rule = patch->ws_rule;
2244        unsigned inaccurate_eof = patch->inaccurate_eof;
2245
2246        if (patch->is_binary)
2247                return apply_binary(img, patch);
2248
2249        while (frag) {
2250                if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2251                        error("patch failed: %s:%ld", name, frag->oldpos);
2252                        if (!apply_with_reject)
2253                                return -1;
2254                        frag->rejected = 1;
2255                }
2256                frag = frag->next;
2257        }
2258        return 0;
2259}
2260
2261static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2262{
2263        if (!ce)
2264                return 0;
2265
2266        if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2267                strbuf_grow(buf, 100);
2268                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2269        } else {
2270                enum object_type type;
2271                unsigned long sz;
2272                char *result;
2273
2274                result = read_sha1_file(ce->sha1, &type, &sz);
2275                if (!result)
2276                        return -1;
2277                /* XXX read_sha1_file NUL-terminates */
2278                strbuf_attach(buf, result, sz, sz + 1);
2279        }
2280        return 0;
2281}
2282
2283static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2284{
2285        struct strbuf buf;
2286        struct image image;
2287        size_t len;
2288        char *img;
2289
2290        strbuf_init(&buf, 0);
2291        if (cached) {
2292                if (read_file_or_gitlink(ce, &buf))
2293                        return error("read of %s failed", patch->old_name);
2294        } else if (patch->old_name) {
2295                if (S_ISGITLINK(patch->old_mode)) {
2296                        if (ce) {
2297                                read_file_or_gitlink(ce, &buf);
2298                        } else {
2299                                /*
2300                                 * There is no way to apply subproject
2301                                 * patch without looking at the index.
2302                                 */
2303                                patch->fragments = NULL;
2304                        }
2305                } else {
2306                        if (read_old_data(st, patch->old_name, &buf))
2307                                return error("read of %s failed", patch->old_name);
2308                }
2309        }
2310
2311        img = strbuf_detach(&buf, &len);
2312        prepare_image(&image, img, len, !patch->is_binary);
2313
2314        if (apply_fragments(&image, patch) < 0)
2315                return -1; /* note with --reject this succeeds. */
2316        patch->result = image.buf;
2317        patch->resultsize = image.len;
2318        free(image.line_allocated);
2319
2320        if (0 < patch->is_delete && patch->resultsize)
2321                return error("removal patch leaves file contents");
2322
2323        return 0;
2324}
2325
2326static int check_to_create_blob(const char *new_name, int ok_if_exists)
2327{
2328        struct stat nst;
2329        if (!lstat(new_name, &nst)) {
2330                if (S_ISDIR(nst.st_mode) || ok_if_exists)
2331                        return 0;
2332                /*
2333                 * A leading component of new_name might be a symlink
2334                 * that is going to be removed with this patch, but
2335                 * still pointing at somewhere that has the path.
2336                 * In such a case, path "new_name" does not exist as
2337                 * far as git is concerned.
2338                 */
2339                if (has_symlink_leading_path(new_name, NULL))
2340                        return 0;
2341
2342                return error("%s: already exists in working directory", new_name);
2343        }
2344        else if ((errno != ENOENT) && (errno != ENOTDIR))
2345                return error("%s: %s", new_name, strerror(errno));
2346        return 0;
2347}
2348
2349static int verify_index_match(struct cache_entry *ce, struct stat *st)
2350{
2351        if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2352                if (!S_ISDIR(st->st_mode))
2353                        return -1;
2354                return 0;
2355        }
2356        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2357}
2358
2359static int check_patch(struct patch *patch, struct patch *prev_patch)
2360{
2361        struct stat st;
2362        const char *old_name = patch->old_name;
2363        const char *new_name = patch->new_name;
2364        const char *name = old_name ? old_name : new_name;
2365        struct cache_entry *ce = NULL;
2366        int ok_if_exists;
2367
2368        patch->rejected = 1; /* we will drop this after we succeed */
2369
2370        /*
2371         * Make sure that we do not have local modifications from the
2372         * index when we are looking at the index.  Also make sure
2373         * we have the preimage file to be patched in the work tree,
2374         * unless --cached, which tells git to apply only in the index.
2375         */
2376        if (old_name) {
2377                int stat_ret = 0;
2378                unsigned st_mode = 0;
2379
2380                if (!cached)
2381                        stat_ret = lstat(old_name, &st);
2382                if (check_index) {
2383                        int pos = cache_name_pos(old_name, strlen(old_name));
2384                        if (pos < 0)
2385                                return error("%s: does not exist in index",
2386                                             old_name);
2387                        ce = active_cache[pos];
2388                        if (stat_ret < 0) {
2389                                struct checkout costate;
2390                                if (errno != ENOENT)
2391                                        return error("%s: %s", old_name,
2392                                                     strerror(errno));
2393                                /* checkout */
2394                                costate.base_dir = "";
2395                                costate.base_dir_len = 0;
2396                                costate.force = 0;
2397                                costate.quiet = 0;
2398                                costate.not_new = 0;
2399                                costate.refresh_cache = 1;
2400                                if (checkout_entry(ce,
2401                                                   &costate,
2402                                                   NULL) ||
2403                                    lstat(old_name, &st))
2404                                        return -1;
2405                        }
2406                        if (!cached && verify_index_match(ce, &st))
2407                                return error("%s: does not match index",
2408                                             old_name);
2409                        if (cached)
2410                                st_mode = ntohl(ce->ce_mode);
2411                } else if (stat_ret < 0)
2412                        return error("%s: %s", old_name, strerror(errno));
2413
2414                if (!cached)
2415                        st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2416
2417                if (patch->is_new < 0)
2418                        patch->is_new = 0;
2419                if (!patch->old_mode)
2420                        patch->old_mode = st_mode;
2421                if ((st_mode ^ patch->old_mode) & S_IFMT)
2422                        return error("%s: wrong type", old_name);
2423                if (st_mode != patch->old_mode)
2424                        fprintf(stderr, "warning: %s has type %o, expected %o\n",
2425                                old_name, st_mode, patch->old_mode);
2426        }
2427
2428        if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2429            !strcmp(prev_patch->old_name, new_name))
2430                /*
2431                 * A type-change diff is always split into a patch to
2432                 * delete old, immediately followed by a patch to
2433                 * create new (see diff.c::run_diff()); in such a case
2434                 * it is Ok that the entry to be deleted by the
2435                 * previous patch is still in the working tree and in
2436                 * the index.
2437                 */
2438                ok_if_exists = 1;
2439        else
2440                ok_if_exists = 0;
2441
2442        if (new_name &&
2443            ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2444                if (check_index &&
2445                    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2446                    !ok_if_exists)
2447                        return error("%s: already exists in index", new_name);
2448                if (!cached) {
2449                        int err = check_to_create_blob(new_name, ok_if_exists);
2450                        if (err)
2451                                return err;
2452                }
2453                if (!patch->new_mode) {
2454                        if (0 < patch->is_new)
2455                                patch->new_mode = S_IFREG | 0644;
2456                        else
2457                                patch->new_mode = patch->old_mode;
2458                }
2459        }
2460
2461        if (new_name && old_name) {
2462                int same = !strcmp(old_name, new_name);
2463                if (!patch->new_mode)
2464                        patch->new_mode = patch->old_mode;
2465                if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2466                        return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2467                                patch->new_mode, new_name, patch->old_mode,
2468                                same ? "" : " of ", same ? "" : old_name);
2469        }
2470
2471        if (apply_data(patch, &st, ce) < 0)
2472                return error("%s: patch does not apply", name);
2473        patch->rejected = 0;
2474        return 0;
2475}
2476
2477static int check_patch_list(struct patch *patch)
2478{
2479        struct patch *prev_patch = NULL;
2480        int err = 0;
2481
2482        for (prev_patch = NULL; patch ; patch = patch->next) {
2483                if (apply_verbosely)
2484                        say_patch_name(stderr,
2485                                       "Checking patch ", patch, "...\n");
2486                err |= check_patch(patch, prev_patch);
2487                prev_patch = patch;
2488        }
2489        return err;
2490}
2491
2492/* This function tries to read the sha1 from the current index */
2493static int get_current_sha1(const char *path, unsigned char *sha1)
2494{
2495        int pos;
2496
2497        if (read_cache() < 0)
2498                return -1;
2499        pos = cache_name_pos(path, strlen(path));
2500        if (pos < 0)
2501                return -1;
2502        hashcpy(sha1, active_cache[pos]->sha1);
2503        return 0;
2504}
2505
2506/* Build an index that contains the just the files needed for a 3way merge */
2507static void build_fake_ancestor(struct patch *list, const char *filename)
2508{
2509        struct patch *patch;
2510        struct index_state result = { 0 };
2511        int fd;
2512
2513        /* Once we start supporting the reverse patch, it may be
2514         * worth showing the new sha1 prefix, but until then...
2515         */
2516        for (patch = list; patch; patch = patch->next) {
2517                const unsigned char *sha1_ptr;
2518                unsigned char sha1[20];
2519                struct cache_entry *ce;
2520                const char *name;
2521
2522                name = patch->old_name ? patch->old_name : patch->new_name;
2523                if (0 < patch->is_new)
2524                        continue;
2525                else if (get_sha1(patch->old_sha1_prefix, sha1))
2526                        /* git diff has no index line for mode/type changes */
2527                        if (!patch->lines_added && !patch->lines_deleted) {
2528                                if (get_current_sha1(patch->new_name, sha1) ||
2529                                    get_current_sha1(patch->old_name, sha1))
2530                                        die("mode change for %s, which is not "
2531                                                "in current HEAD", name);
2532                                sha1_ptr = sha1;
2533                        } else
2534                                die("sha1 information is lacking or useless "
2535                                        "(%s).", name);
2536                else
2537                        sha1_ptr = sha1;
2538
2539                ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2540                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2541                        die ("Could not add %s to temporary index", name);
2542        }
2543
2544        fd = open(filename, O_WRONLY | O_CREAT, 0666);
2545        if (fd < 0 || write_index(&result, fd) || close(fd))
2546                die ("Could not write temporary index to %s", filename);
2547
2548        discard_index(&result);
2549}
2550
2551static void stat_patch_list(struct patch *patch)
2552{
2553        int files, adds, dels;
2554
2555        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2556                files++;
2557                adds += patch->lines_added;
2558                dels += patch->lines_deleted;
2559                show_stats(patch);
2560        }
2561
2562        printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2563}
2564
2565static void numstat_patch_list(struct patch *patch)
2566{
2567        for ( ; patch; patch = patch->next) {
2568                const char *name;
2569                name = patch->new_name ? patch->new_name : patch->old_name;
2570                if (patch->is_binary)
2571                        printf("-\t-\t");
2572                else
2573                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2574                write_name_quoted(name, stdout, line_termination);
2575        }
2576}
2577
2578static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2579{
2580        if (mode)
2581                printf(" %s mode %06o %s\n", newdelete, mode, name);
2582        else
2583                printf(" %s %s\n", newdelete, name);
2584}
2585
2586static void show_mode_change(struct patch *p, int show_name)
2587{
2588        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2589                if (show_name)
2590                        printf(" mode change %06o => %06o %s\n",
2591                               p->old_mode, p->new_mode, p->new_name);
2592                else
2593                        printf(" mode change %06o => %06o\n",
2594                               p->old_mode, p->new_mode);
2595        }
2596}
2597
2598static void show_rename_copy(struct patch *p)
2599{
2600        const char *renamecopy = p->is_rename ? "rename" : "copy";
2601        const char *old, *new;
2602
2603        /* Find common prefix */
2604        old = p->old_name;
2605        new = p->new_name;
2606        while (1) {
2607                const char *slash_old, *slash_new;
2608                slash_old = strchr(old, '/');
2609                slash_new = strchr(new, '/');
2610                if (!slash_old ||
2611                    !slash_new ||
2612                    slash_old - old != slash_new - new ||
2613                    memcmp(old, new, slash_new - new))
2614                        break;
2615                old = slash_old + 1;
2616                new = slash_new + 1;
2617        }
2618        /* p->old_name thru old is the common prefix, and old and new
2619         * through the end of names are renames
2620         */
2621        if (old != p->old_name)
2622                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2623                       (int)(old - p->old_name), p->old_name,
2624                       old, new, p->score);
2625        else
2626                printf(" %s %s => %s (%d%%)\n", renamecopy,
2627                       p->old_name, p->new_name, p->score);
2628        show_mode_change(p, 0);
2629}
2630
2631static void summary_patch_list(struct patch *patch)
2632{
2633        struct patch *p;
2634
2635        for (p = patch; p; p = p->next) {
2636                if (p->is_new)
2637                        show_file_mode_name("create", p->new_mode, p->new_name);
2638                else if (p->is_delete)
2639                        show_file_mode_name("delete", p->old_mode, p->old_name);
2640                else {
2641                        if (p->is_rename || p->is_copy)
2642                                show_rename_copy(p);
2643                        else {
2644                                if (p->score) {
2645                                        printf(" rewrite %s (%d%%)\n",
2646                                               p->new_name, p->score);
2647                                        show_mode_change(p, 0);
2648                                }
2649                                else
2650                                        show_mode_change(p, 1);
2651                        }
2652                }
2653        }
2654}
2655
2656static void patch_stats(struct patch *patch)
2657{
2658        int lines = patch->lines_added + patch->lines_deleted;
2659
2660        if (lines > max_change)
2661                max_change = lines;
2662        if (patch->old_name) {
2663                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2664                if (!len)
2665                        len = strlen(patch->old_name);
2666                if (len > max_len)
2667                        max_len = len;
2668        }
2669        if (patch->new_name) {
2670                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2671                if (!len)
2672                        len = strlen(patch->new_name);
2673                if (len > max_len)
2674                        max_len = len;
2675        }
2676}
2677
2678static void remove_file(struct patch *patch, int rmdir_empty)
2679{
2680        if (update_index) {
2681                if (remove_file_from_cache(patch->old_name) < 0)
2682                        die("unable to remove %s from index", patch->old_name);
2683        }
2684        if (!cached) {
2685                if (S_ISGITLINK(patch->old_mode)) {
2686                        if (rmdir(patch->old_name))
2687                                warning("unable to remove submodule %s",
2688                                        patch->old_name);
2689                } else if (!unlink(patch->old_name) && rmdir_empty) {
2690                        char *name = xstrdup(patch->old_name);
2691                        char *end = strrchr(name, '/');
2692                        while (end) {
2693                                *end = 0;
2694                                if (rmdir(name))
2695                                        break;
2696                                end = strrchr(name, '/');
2697                        }
2698                        free(name);
2699                }
2700        }
2701}
2702
2703static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2704{
2705        struct stat st;
2706        struct cache_entry *ce;
2707        int namelen = strlen(path);
2708        unsigned ce_size = cache_entry_size(namelen);
2709
2710        if (!update_index)
2711                return;
2712
2713        ce = xcalloc(1, ce_size);
2714        memcpy(ce->name, path, namelen);
2715        ce->ce_mode = create_ce_mode(mode);
2716        ce->ce_flags = htons(namelen);
2717        if (S_ISGITLINK(mode)) {
2718                const char *s = buf;
2719
2720                if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2721                        die("corrupt patch for subproject %s", path);
2722        } else {
2723                if (!cached) {
2724                        if (lstat(path, &st) < 0)
2725                                die("unable to stat newly created file %s",
2726                                    path);
2727                        fill_stat_cache_info(ce, &st);
2728                }
2729                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2730                        die("unable to create backing store for newly created file %s", path);
2731        }
2732        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2733                die("unable to add cache entry for %s", path);
2734}
2735
2736static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2737{
2738        int fd;
2739        struct strbuf nbuf;
2740
2741        if (S_ISGITLINK(mode)) {
2742                struct stat st;
2743                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2744                        return 0;
2745                return mkdir(path, 0777);
2746        }
2747
2748        if (has_symlinks && S_ISLNK(mode))
2749                /* Although buf:size is counted string, it also is NUL
2750                 * terminated.
2751                 */
2752                return symlink(buf, path);
2753
2754        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2755        if (fd < 0)
2756                return -1;
2757
2758        strbuf_init(&nbuf, 0);
2759        if (convert_to_working_tree(path, buf, size, &nbuf)) {
2760                size = nbuf.len;
2761                buf  = nbuf.buf;
2762        }
2763        write_or_die(fd, buf, size);
2764        strbuf_release(&nbuf);
2765
2766        if (close(fd) < 0)
2767                die("closing file %s: %s", path, strerror(errno));
2768        return 0;
2769}
2770
2771/*
2772 * We optimistically assume that the directories exist,
2773 * which is true 99% of the time anyway. If they don't,
2774 * we create them and try again.
2775 */
2776static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2777{
2778        if (cached)
2779                return;
2780        if (!try_create_file(path, mode, buf, size))
2781                return;
2782
2783        if (errno == ENOENT) {
2784                if (safe_create_leading_directories(path))
2785                        return;
2786                if (!try_create_file(path, mode, buf, size))
2787                        return;
2788        }
2789
2790        if (errno == EEXIST || errno == EACCES) {
2791                /* We may be trying to create a file where a directory
2792                 * used to be.
2793                 */
2794                struct stat st;
2795                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2796                        errno = EEXIST;
2797        }
2798
2799        if (errno == EEXIST) {
2800                unsigned int nr = getpid();
2801
2802                for (;;) {
2803                        const char *newpath;
2804                        newpath = mkpath("%s~%u", path, nr);
2805                        if (!try_create_file(newpath, mode, buf, size)) {
2806                                if (!rename(newpath, path))
2807                                        return;
2808                                unlink(newpath);
2809                                break;
2810                        }
2811                        if (errno != EEXIST)
2812                                break;
2813                        ++nr;
2814                }
2815        }
2816        die("unable to write file %s mode %o", path, mode);
2817}
2818
2819static void create_file(struct patch *patch)
2820{
2821        char *path = patch->new_name;
2822        unsigned mode = patch->new_mode;
2823        unsigned long size = patch->resultsize;
2824        char *buf = patch->result;
2825
2826        if (!mode)
2827                mode = S_IFREG | 0644;
2828        create_one_file(path, mode, buf, size);
2829        add_index_file(path, mode, buf, size);
2830}
2831
2832/* phase zero is to remove, phase one is to create */
2833static void write_out_one_result(struct patch *patch, int phase)
2834{
2835        if (patch->is_delete > 0) {
2836                if (phase == 0)
2837                        remove_file(patch, 1);
2838                return;
2839        }
2840        if (patch->is_new > 0 || patch->is_copy) {
2841                if (phase == 1)
2842                        create_file(patch);
2843                return;
2844        }
2845        /*
2846         * Rename or modification boils down to the same
2847         * thing: remove the old, write the new
2848         */
2849        if (phase == 0)
2850                remove_file(patch, patch->is_rename);
2851        if (phase == 1)
2852                create_file(patch);
2853}
2854
2855static int write_out_one_reject(struct patch *patch)
2856{
2857        FILE *rej;
2858        char namebuf[PATH_MAX];
2859        struct fragment *frag;
2860        int cnt = 0;
2861
2862        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2863                if (!frag->rejected)
2864                        continue;
2865                cnt++;
2866        }
2867
2868        if (!cnt) {
2869                if (apply_verbosely)
2870                        say_patch_name(stderr,
2871                                       "Applied patch ", patch, " cleanly.\n");
2872                return 0;
2873        }
2874
2875        /* This should not happen, because a removal patch that leaves
2876         * contents are marked "rejected" at the patch level.
2877         */
2878        if (!patch->new_name)
2879                die("internal error");
2880
2881        /* Say this even without --verbose */
2882        say_patch_name(stderr, "Applying patch ", patch, " with");
2883        fprintf(stderr, " %d rejects...\n", cnt);
2884
2885        cnt = strlen(patch->new_name);
2886        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2887                cnt = ARRAY_SIZE(namebuf) - 5;
2888                fprintf(stderr,
2889                        "warning: truncating .rej filename to %.*s.rej",
2890                        cnt - 1, patch->new_name);
2891        }
2892        memcpy(namebuf, patch->new_name, cnt);
2893        memcpy(namebuf + cnt, ".rej", 5);
2894
2895        rej = fopen(namebuf, "w");
2896        if (!rej)
2897                return error("cannot open %s: %s", namebuf, strerror(errno));
2898
2899        /* Normal git tools never deal with .rej, so do not pretend
2900         * this is a git patch by saying --git nor give extended
2901         * headers.  While at it, maybe please "kompare" that wants
2902         * the trailing TAB and some garbage at the end of line ;-).
2903         */
2904        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2905                patch->new_name, patch->new_name);
2906        for (cnt = 1, frag = patch->fragments;
2907             frag;
2908             cnt++, frag = frag->next) {
2909                if (!frag->rejected) {
2910                        fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2911                        continue;
2912                }
2913                fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2914                fprintf(rej, "%.*s", frag->size, frag->patch);
2915                if (frag->patch[frag->size-1] != '\n')
2916                        fputc('\n', rej);
2917        }
2918        fclose(rej);
2919        return -1;
2920}
2921
2922static int write_out_results(struct patch *list, int skipped_patch)
2923{
2924        int phase;
2925        int errs = 0;
2926        struct patch *l;
2927
2928        if (!list && !skipped_patch)
2929                return error("No changes");
2930
2931        for (phase = 0; phase < 2; phase++) {
2932                l = list;
2933                while (l) {
2934                        if (l->rejected)
2935                                errs = 1;
2936                        else {
2937                                write_out_one_result(l, phase);
2938                                if (phase == 1 && write_out_one_reject(l))
2939                                        errs = 1;
2940                        }
2941                        l = l->next;
2942                }
2943        }
2944        return errs;
2945}
2946
2947static struct lock_file lock_file;
2948
2949static struct excludes {
2950        struct excludes *next;
2951        const char *path;
2952} *excludes;
2953
2954static int use_patch(struct patch *p)
2955{
2956        const char *pathname = p->new_name ? p->new_name : p->old_name;
2957        struct excludes *x = excludes;
2958        while (x) {
2959                if (fnmatch(x->path, pathname, 0) == 0)
2960                        return 0;
2961                x = x->next;
2962        }
2963        if (0 < prefix_length) {
2964                int pathlen = strlen(pathname);
2965                if (pathlen <= prefix_length ||
2966                    memcmp(prefix, pathname, prefix_length))
2967                        return 0;
2968        }
2969        return 1;
2970}
2971
2972static void prefix_one(char **name)
2973{
2974        char *old_name = *name;
2975        if (!old_name)
2976                return;
2977        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2978        free(old_name);
2979}
2980
2981static void prefix_patches(struct patch *p)
2982{
2983        if (!prefix || p->is_toplevel_relative)
2984                return;
2985        for ( ; p; p = p->next) {
2986                if (p->new_name == p->old_name) {
2987                        char *prefixed = p->new_name;
2988                        prefix_one(&prefixed);
2989                        p->new_name = p->old_name = prefixed;
2990                }
2991                else {
2992                        prefix_one(&p->new_name);
2993                        prefix_one(&p->old_name);
2994                }
2995        }
2996}
2997
2998static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2999{
3000        size_t offset;
3001        struct strbuf buf;
3002        struct patch *list = NULL, **listp = &list;
3003        int skipped_patch = 0;
3004
3005        strbuf_init(&buf, 0);
3006        patch_input_file = filename;
3007        read_patch_file(&buf, fd);
3008        offset = 0;
3009        while (offset < buf.len) {
3010                struct patch *patch;
3011                int nr;
3012
3013                patch = xcalloc(1, sizeof(*patch));
3014                patch->inaccurate_eof = inaccurate_eof;
3015                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3016                if (nr < 0)
3017                        break;
3018                if (apply_in_reverse)
3019                        reverse_patches(patch);
3020                if (prefix)
3021                        prefix_patches(patch);
3022                if (use_patch(patch)) {
3023                        patch_stats(patch);
3024                        *listp = patch;
3025                        listp = &patch->next;
3026                }
3027                else {
3028                        /* perhaps free it a bit better? */
3029                        free(patch);
3030                        skipped_patch++;
3031                }
3032                offset += nr;
3033        }
3034
3035        if (whitespace_error && (ws_error_action == die_on_ws_error))
3036                apply = 0;
3037
3038        update_index = check_index && apply;
3039        if (update_index && newfd < 0)
3040                newfd = hold_locked_index(&lock_file, 1);
3041
3042        if (check_index) {
3043                if (read_cache() < 0)
3044                        die("unable to read index file");
3045        }
3046
3047        if ((check || apply) &&
3048            check_patch_list(list) < 0 &&
3049            !apply_with_reject)
3050                exit(1);
3051
3052        if (apply && write_out_results(list, skipped_patch))
3053                exit(1);
3054
3055        if (fake_ancestor)
3056                build_fake_ancestor(list, fake_ancestor);
3057
3058        if (diffstat)
3059                stat_patch_list(list);
3060
3061        if (numstat)
3062                numstat_patch_list(list);
3063
3064        if (summary)
3065                summary_patch_list(list);
3066
3067        strbuf_release(&buf);
3068        return 0;
3069}
3070
3071static int git_apply_config(const char *var, const char *value)
3072{
3073        if (!strcmp(var, "apply.whitespace")) {
3074                apply_default_whitespace = xstrdup(value);
3075                return 0;
3076        }
3077        return git_default_config(var, value);
3078}
3079
3080
3081int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3082{
3083        int i;
3084        int read_stdin = 1;
3085        int inaccurate_eof = 0;
3086        int errs = 0;
3087        int is_not_gitdir = 0;
3088
3089        const char *whitespace_option = NULL;
3090
3091        prefix = setup_git_directory_gently(&is_not_gitdir);
3092        prefix_length = prefix ? strlen(prefix) : 0;
3093        git_config(git_apply_config);
3094        if (apply_default_whitespace)
3095                parse_whitespace_option(apply_default_whitespace);
3096
3097        for (i = 1; i < argc; i++) {
3098                const char *arg = argv[i];
3099                char *end;
3100                int fd;
3101
3102                if (!strcmp(arg, "-")) {
3103                        errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3104                        read_stdin = 0;
3105                        continue;
3106                }
3107                if (!prefixcmp(arg, "--exclude=")) {
3108                        struct excludes *x = xmalloc(sizeof(*x));
3109                        x->path = arg + 10;
3110                        x->next = excludes;
3111                        excludes = x;
3112                        continue;
3113                }
3114                if (!prefixcmp(arg, "-p")) {
3115                        p_value = atoi(arg + 2);
3116                        p_value_known = 1;
3117                        continue;
3118                }
3119                if (!strcmp(arg, "--no-add")) {
3120                        no_add = 1;
3121                        continue;
3122                }
3123                if (!strcmp(arg, "--stat")) {
3124                        apply = 0;
3125                        diffstat = 1;
3126                        continue;
3127                }
3128                if (!strcmp(arg, "--allow-binary-replacement") ||
3129                    !strcmp(arg, "--binary")) {
3130                        continue; /* now no-op */
3131                }
3132                if (!strcmp(arg, "--numstat")) {
3133                        apply = 0;
3134                        numstat = 1;
3135                        continue;
3136                }
3137                if (!strcmp(arg, "--summary")) {
3138                        apply = 0;
3139                        summary = 1;
3140                        continue;
3141                }
3142                if (!strcmp(arg, "--check")) {
3143                        apply = 0;
3144                        check = 1;
3145                        continue;
3146                }
3147                if (!strcmp(arg, "--index")) {
3148                        if (is_not_gitdir)
3149                                die("--index outside a repository");
3150                        check_index = 1;
3151                        continue;
3152                }
3153                if (!strcmp(arg, "--cached")) {
3154                        if (is_not_gitdir)
3155                                die("--cached outside a repository");
3156                        check_index = 1;
3157                        cached = 1;
3158                        continue;
3159                }
3160                if (!strcmp(arg, "--apply")) {
3161                        apply = 1;
3162                        continue;
3163                }
3164                if (!strcmp(arg, "--build-fake-ancestor")) {
3165                        apply = 0;
3166                        if (++i >= argc)
3167                                die ("need a filename");
3168                        fake_ancestor = argv[i];
3169                        continue;
3170                }
3171                if (!strcmp(arg, "-z")) {
3172                        line_termination = 0;
3173                        continue;
3174                }
3175                if (!prefixcmp(arg, "-C")) {
3176                        p_context = strtoul(arg + 2, &end, 0);
3177                        if (*end != '\0')
3178                                die("unrecognized context count '%s'", arg + 2);
3179                        continue;
3180                }
3181                if (!prefixcmp(arg, "--whitespace=")) {
3182                        whitespace_option = arg + 13;
3183                        parse_whitespace_option(arg + 13);
3184                        continue;
3185                }
3186                if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3187                        apply_in_reverse = 1;
3188                        continue;
3189                }
3190                if (!strcmp(arg, "--unidiff-zero")) {
3191                        unidiff_zero = 1;
3192                        continue;
3193                }
3194                if (!strcmp(arg, "--reject")) {
3195                        apply = apply_with_reject = apply_verbosely = 1;
3196                        continue;
3197                }
3198                if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3199                        apply_verbosely = 1;
3200                        continue;
3201                }
3202                if (!strcmp(arg, "--inaccurate-eof")) {
3203                        inaccurate_eof = 1;
3204                        continue;
3205                }
3206                if (0 < prefix_length)
3207                        arg = prefix_filename(prefix, prefix_length, arg);
3208
3209                fd = open(arg, O_RDONLY);
3210                if (fd < 0)
3211                        usage(apply_usage);
3212                read_stdin = 0;
3213                set_default_whitespace_mode(whitespace_option);
3214                errs |= apply_patch(fd, arg, inaccurate_eof);
3215                close(fd);
3216        }
3217        set_default_whitespace_mode(whitespace_option);
3218        if (read_stdin)
3219                errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3220        if (whitespace_error) {
3221                if (squelch_whitespace_errors &&
3222                    squelch_whitespace_errors < whitespace_error) {
3223                        int squelched =
3224                                whitespace_error - squelch_whitespace_errors;
3225                        fprintf(stderr, "warning: squelched %d "
3226                                "whitespace error%s\n",
3227                                squelched,
3228                                squelched == 1 ? "" : "s");
3229                }
3230                if (ws_error_action == die_on_ws_error)
3231                        die("%d line%s add%s whitespace errors.",
3232                            whitespace_error,
3233                            whitespace_error == 1 ? "" : "s",
3234                            whitespace_error == 1 ? "s" : "");
3235                if (applied_after_fixing_ws && apply)
3236                        fprintf(stderr, "warning: %d line%s applied after"
3237                                " fixing whitespace errors.\n",
3238                                applied_after_fixing_ws,
3239                                applied_after_fixing_ws == 1 ? "" : "s");
3240                else if (whitespace_error)
3241                        fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3242                                whitespace_error,
3243                                whitespace_error == 1 ? "" : "s",
3244                                whitespace_error == 1 ? "s" : "");
3245        }
3246
3247        if (update_index) {
3248                if (write_cache(newfd, active_cache, active_nr) ||
3249                    commit_locked_index(&lock_file))
3250                        die("Unable to write new index file");
3251        }
3252
3253        return !!errs;
3254}