builtin-apply.con commit fetch: Speed up fetch of large numbers of refs (b1a01e1)
   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#include "string-list.h"
  16#include "dir.h"
  17#include "parse-options.h"
  18
  19/*
  20 *  --check turns on checking that the working tree matches the
  21 *    files that are being modified, but doesn't apply the patch
  22 *  --stat does just a diffstat, and doesn't actually apply
  23 *  --numstat does numeric diffstat, and doesn't actually apply
  24 *  --index-info shows the old and new index info for paths if available.
  25 *  --index updates the cache as well.
  26 *  --cached updates only the cache without ever touching the working tree.
  27 */
  28static const char *prefix;
  29static int prefix_length = -1;
  30static int newfd = -1;
  31
  32static int unidiff_zero;
  33static int p_value = 1;
  34static int p_value_known;
  35static int check_index;
  36static int update_index;
  37static int cached;
  38static int diffstat;
  39static int numstat;
  40static int summary;
  41static int check;
  42static int apply = 1;
  43static int apply_in_reverse;
  44static int apply_with_reject;
  45static int apply_verbosely;
  46static int no_add;
  47static const char *fake_ancestor;
  48static int line_termination = '\n';
  49static unsigned int p_context = UINT_MAX;
  50static const char * const apply_usage[] = {
  51        "git apply [options] [<patch>...]",
  52        NULL
  53};
  54
  55static enum ws_error_action {
  56        nowarn_ws_error,
  57        warn_on_ws_error,
  58        die_on_ws_error,
  59        correct_ws_error,
  60} ws_error_action = warn_on_ws_error;
  61static int whitespace_error;
  62static int squelch_whitespace_errors = 5;
  63static int applied_after_fixing_ws;
  64
  65static enum ws_ignore {
  66        ignore_ws_none,
  67        ignore_ws_change,
  68} ws_ignore_action = ignore_ws_none;
  69
  70
  71static const char *patch_input_file;
  72static const char *root;
  73static int root_len;
  74static int read_stdin = 1;
  75static int options;
  76
  77static void parse_whitespace_option(const char *option)
  78{
  79        if (!option) {
  80                ws_error_action = warn_on_ws_error;
  81                return;
  82        }
  83        if (!strcmp(option, "warn")) {
  84                ws_error_action = warn_on_ws_error;
  85                return;
  86        }
  87        if (!strcmp(option, "nowarn")) {
  88                ws_error_action = nowarn_ws_error;
  89                return;
  90        }
  91        if (!strcmp(option, "error")) {
  92                ws_error_action = die_on_ws_error;
  93                return;
  94        }
  95        if (!strcmp(option, "error-all")) {
  96                ws_error_action = die_on_ws_error;
  97                squelch_whitespace_errors = 0;
  98                return;
  99        }
 100        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
 101                ws_error_action = correct_ws_error;
 102                return;
 103        }
 104        die("unrecognized whitespace option '%s'", option);
 105}
 106
 107static void parse_ignorewhitespace_option(const char *option)
 108{
 109        if (!option || !strcmp(option, "no") ||
 110            !strcmp(option, "false") || !strcmp(option, "never") ||
 111            !strcmp(option, "none")) {
 112                ws_ignore_action = ignore_ws_none;
 113                return;
 114        }
 115        if (!strcmp(option, "change")) {
 116                ws_ignore_action = ignore_ws_change;
 117                return;
 118        }
 119        die("unrecognized whitespace ignore option '%s'", option);
 120}
 121
 122static void set_default_whitespace_mode(const char *whitespace_option)
 123{
 124        if (!whitespace_option && !apply_default_whitespace)
 125                ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
 126}
 127
 128/*
 129 * For "diff-stat" like behaviour, we keep track of the biggest change
 130 * we've seen, and the longest filename. That allows us to do simple
 131 * scaling.
 132 */
 133static int max_change, max_len;
 134
 135/*
 136 * Various "current state", notably line numbers and what
 137 * file (and how) we're patching right now.. The "is_xxxx"
 138 * things are flags, where -1 means "don't know yet".
 139 */
 140static int linenr = 1;
 141
 142/*
 143 * This represents one "hunk" from a patch, starting with
 144 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 145 * patch text is pointed at by patch, and its byte length
 146 * is stored in size.  leading and trailing are the number
 147 * of context lines.
 148 */
 149struct fragment {
 150        unsigned long leading, trailing;
 151        unsigned long oldpos, oldlines;
 152        unsigned long newpos, newlines;
 153        const char *patch;
 154        int size;
 155        int rejected;
 156        struct fragment *next;
 157};
 158
 159/*
 160 * When dealing with a binary patch, we reuse "leading" field
 161 * to store the type of the binary hunk, either deflated "delta"
 162 * or deflated "literal".
 163 */
 164#define binary_patch_method leading
 165#define BINARY_DELTA_DEFLATED   1
 166#define BINARY_LITERAL_DEFLATED 2
 167
 168/*
 169 * This represents a "patch" to a file, both metainfo changes
 170 * such as creation/deletion, filemode and content changes represented
 171 * as a series of fragments.
 172 */
 173struct patch {
 174        char *new_name, *old_name, *def_name;
 175        unsigned int old_mode, new_mode;
 176        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 177        int rejected;
 178        unsigned ws_rule;
 179        unsigned long deflate_origlen;
 180        int lines_added, lines_deleted;
 181        int score;
 182        unsigned int is_toplevel_relative:1;
 183        unsigned int inaccurate_eof:1;
 184        unsigned int is_binary:1;
 185        unsigned int is_copy:1;
 186        unsigned int is_rename:1;
 187        unsigned int recount:1;
 188        struct fragment *fragments;
 189        char *result;
 190        size_t resultsize;
 191        char old_sha1_prefix[41];
 192        char new_sha1_prefix[41];
 193        struct patch *next;
 194};
 195
 196/*
 197 * A line in a file, len-bytes long (includes the terminating LF,
 198 * except for an incomplete line at the end if the file ends with
 199 * one), and its contents hashes to 'hash'.
 200 */
 201struct line {
 202        size_t len;
 203        unsigned hash : 24;
 204        unsigned flag : 8;
 205#define LINE_COMMON     1
 206};
 207
 208/*
 209 * This represents a "file", which is an array of "lines".
 210 */
 211struct image {
 212        char *buf;
 213        size_t len;
 214        size_t nr;
 215        size_t alloc;
 216        struct line *line_allocated;
 217        struct line *line;
 218};
 219
 220/*
 221 * Records filenames that have been touched, in order to handle
 222 * the case where more than one patches touch the same file.
 223 */
 224
 225static struct string_list fn_table;
 226
 227static uint32_t hash_line(const char *cp, size_t len)
 228{
 229        size_t i;
 230        uint32_t h;
 231        for (i = 0, h = 0; i < len; i++) {
 232                if (!isspace(cp[i])) {
 233                        h = h * 3 + (cp[i] & 0xff);
 234                }
 235        }
 236        return h;
 237}
 238
 239/*
 240 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 241 * whitespace difference. Returns 1 if they match, 0 otherwise
 242 */
 243static int fuzzy_matchlines(const char *s1, size_t n1,
 244                            const char *s2, size_t n2)
 245{
 246        const char *last1 = s1 + n1 - 1;
 247        const char *last2 = s2 + n2 - 1;
 248        int result = 0;
 249
 250        if (n1 < 0 || n2 < 0)
 251                return 0;
 252
 253        /* ignore line endings */
 254        while ((*last1 == '\r') || (*last1 == '\n'))
 255                last1--;
 256        while ((*last2 == '\r') || (*last2 == '\n'))
 257                last2--;
 258
 259        /* skip leading whitespace */
 260        while (isspace(*s1) && (s1 <= last1))
 261                s1++;
 262        while (isspace(*s2) && (s2 <= last2))
 263                s2++;
 264        /* early return if both lines are empty */
 265        if ((s1 > last1) && (s2 > last2))
 266                return 1;
 267        while (!result) {
 268                result = *s1++ - *s2++;
 269                /*
 270                 * Skip whitespace inside. We check for whitespace on
 271                 * both buffers because we don't want "a b" to match
 272                 * "ab"
 273                 */
 274                if (isspace(*s1) && isspace(*s2)) {
 275                        while (isspace(*s1) && s1 <= last1)
 276                                s1++;
 277                        while (isspace(*s2) && s2 <= last2)
 278                                s2++;
 279                }
 280                /*
 281                 * If we reached the end on one side only,
 282                 * lines don't match
 283                 */
 284                if (
 285                    ((s2 > last2) && (s1 <= last1)) ||
 286                    ((s1 > last1) && (s2 <= last2)))
 287                        return 0;
 288                if ((s1 > last1) && (s2 > last2))
 289                        break;
 290        }
 291
 292        return !result;
 293}
 294
 295static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 296{
 297        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 298        img->line_allocated[img->nr].len = len;
 299        img->line_allocated[img->nr].hash = hash_line(bol, len);
 300        img->line_allocated[img->nr].flag = flag;
 301        img->nr++;
 302}
 303
 304static void prepare_image(struct image *image, char *buf, size_t len,
 305                          int prepare_linetable)
 306{
 307        const char *cp, *ep;
 308
 309        memset(image, 0, sizeof(*image));
 310        image->buf = buf;
 311        image->len = len;
 312
 313        if (!prepare_linetable)
 314                return;
 315
 316        ep = image->buf + image->len;
 317        cp = image->buf;
 318        while (cp < ep) {
 319                const char *next;
 320                for (next = cp; next < ep && *next != '\n'; next++)
 321                        ;
 322                if (next < ep)
 323                        next++;
 324                add_line_info(image, cp, next - cp, 0);
 325                cp = next;
 326        }
 327        image->line = image->line_allocated;
 328}
 329
 330static void clear_image(struct image *image)
 331{
 332        free(image->buf);
 333        image->buf = NULL;
 334        image->len = 0;
 335}
 336
 337static void say_patch_name(FILE *output, const char *pre,
 338                           struct patch *patch, const char *post)
 339{
 340        fputs(pre, output);
 341        if (patch->old_name && patch->new_name &&
 342            strcmp(patch->old_name, patch->new_name)) {
 343                quote_c_style(patch->old_name, NULL, output, 0);
 344                fputs(" => ", output);
 345                quote_c_style(patch->new_name, NULL, output, 0);
 346        } else {
 347                const char *n = patch->new_name;
 348                if (!n)
 349                        n = patch->old_name;
 350                quote_c_style(n, NULL, output, 0);
 351        }
 352        fputs(post, output);
 353}
 354
 355#define CHUNKSIZE (8192)
 356#define SLOP (16)
 357
 358static void read_patch_file(struct strbuf *sb, int fd)
 359{
 360        if (strbuf_read(sb, fd, 0) < 0)
 361                die_errno("git apply: failed to read");
 362
 363        /*
 364         * Make sure that we have some slop in the buffer
 365         * so that we can do speculative "memcmp" etc, and
 366         * see to it that it is NUL-filled.
 367         */
 368        strbuf_grow(sb, SLOP);
 369        memset(sb->buf + sb->len, 0, SLOP);
 370}
 371
 372static unsigned long linelen(const char *buffer, unsigned long size)
 373{
 374        unsigned long len = 0;
 375        while (size--) {
 376                len++;
 377                if (*buffer++ == '\n')
 378                        break;
 379        }
 380        return len;
 381}
 382
 383static int is_dev_null(const char *str)
 384{
 385        return !memcmp("/dev/null", str, 9) && isspace(str[9]);
 386}
 387
 388#define TERM_SPACE      1
 389#define TERM_TAB        2
 390
 391static int name_terminate(const char *name, int namelen, int c, int terminate)
 392{
 393        if (c == ' ' && !(terminate & TERM_SPACE))
 394                return 0;
 395        if (c == '\t' && !(terminate & TERM_TAB))
 396                return 0;
 397
 398        return 1;
 399}
 400
 401/* remove double slashes to make --index work with such filenames */
 402static char *squash_slash(char *name)
 403{
 404        int i = 0, j = 0;
 405
 406        while (name[i]) {
 407                if ((name[j++] = name[i++]) == '/')
 408                        while (name[i] == '/')
 409                                i++;
 410        }
 411        name[j] = '\0';
 412        return name;
 413}
 414
 415static char *find_name(const char *line, char *def, int p_value, int terminate)
 416{
 417        int len;
 418        const char *start = line;
 419
 420        if (*line == '"') {
 421                struct strbuf name = STRBUF_INIT;
 422
 423                /*
 424                 * Proposed "new-style" GNU patch/diff format; see
 425                 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
 426                 */
 427                if (!unquote_c_style(&name, line, NULL)) {
 428                        char *cp;
 429
 430                        for (cp = name.buf; p_value; p_value--) {
 431                                cp = strchr(cp, '/');
 432                                if (!cp)
 433                                        break;
 434                                cp++;
 435                        }
 436                        if (cp) {
 437                                /* name can later be freed, so we need
 438                                 * to memmove, not just return cp
 439                                 */
 440                                strbuf_remove(&name, 0, cp - name.buf);
 441                                free(def);
 442                                if (root)
 443                                        strbuf_insert(&name, 0, root, root_len);
 444                                return squash_slash(strbuf_detach(&name, NULL));
 445                        }
 446                }
 447                strbuf_release(&name);
 448        }
 449
 450        for (;;) {
 451                char c = *line;
 452
 453                if (isspace(c)) {
 454                        if (c == '\n')
 455                                break;
 456                        if (name_terminate(start, line-start, c, terminate))
 457                                break;
 458                }
 459                line++;
 460                if (c == '/' && !--p_value)
 461                        start = line;
 462        }
 463        if (!start)
 464                return squash_slash(def);
 465        len = line - start;
 466        if (!len)
 467                return squash_slash(def);
 468
 469        /*
 470         * Generally we prefer the shorter name, especially
 471         * if the other one is just a variation of that with
 472         * something else tacked on to the end (ie "file.orig"
 473         * or "file~").
 474         */
 475        if (def) {
 476                int deflen = strlen(def);
 477                if (deflen < len && !strncmp(start, def, deflen))
 478                        return squash_slash(def);
 479                free(def);
 480        }
 481
 482        if (root) {
 483                char *ret = xmalloc(root_len + len + 1);
 484                strcpy(ret, root);
 485                memcpy(ret + root_len, start, len);
 486                ret[root_len + len] = '\0';
 487                return squash_slash(ret);
 488        }
 489
 490        return squash_slash(xmemdupz(start, len));
 491}
 492
 493static int count_slashes(const char *cp)
 494{
 495        int cnt = 0;
 496        char ch;
 497
 498        while ((ch = *cp++))
 499                if (ch == '/')
 500                        cnt++;
 501        return cnt;
 502}
 503
 504/*
 505 * Given the string after "--- " or "+++ ", guess the appropriate
 506 * p_value for the given patch.
 507 */
 508static int guess_p_value(const char *nameline)
 509{
 510        char *name, *cp;
 511        int val = -1;
 512
 513        if (is_dev_null(nameline))
 514                return -1;
 515        name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
 516        if (!name)
 517                return -1;
 518        cp = strchr(name, '/');
 519        if (!cp)
 520                val = 0;
 521        else if (prefix) {
 522                /*
 523                 * Does it begin with "a/$our-prefix" and such?  Then this is
 524                 * very likely to apply to our directory.
 525                 */
 526                if (!strncmp(name, prefix, prefix_length))
 527                        val = count_slashes(prefix);
 528                else {
 529                        cp++;
 530                        if (!strncmp(cp, prefix, prefix_length))
 531                                val = count_slashes(prefix) + 1;
 532                }
 533        }
 534        free(name);
 535        return val;
 536}
 537
 538/*
 539 * Does the ---/+++ line has the POSIX timestamp after the last HT?
 540 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 541 * this such a timestamp?
 542 */
 543static int has_epoch_timestamp(const char *nameline)
 544{
 545        /*
 546         * We are only interested in epoch timestamp; any non-zero
 547         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 548         * For the same reason, the date must be either 1969-12-31 or
 549         * 1970-01-01, and the seconds part must be "00".
 550         */
 551        const char stamp_regexp[] =
 552                "^(1969-12-31|1970-01-01)"
 553                " "
 554                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 555                " "
 556                "([-+][0-2][0-9][0-5][0-9])\n";
 557        const char *timestamp = NULL, *cp;
 558        static regex_t *stamp;
 559        regmatch_t m[10];
 560        int zoneoffset;
 561        int hourminute;
 562        int status;
 563
 564        for (cp = nameline; *cp != '\n'; cp++) {
 565                if (*cp == '\t')
 566                        timestamp = cp + 1;
 567        }
 568        if (!timestamp)
 569                return 0;
 570        if (!stamp) {
 571                stamp = xmalloc(sizeof(*stamp));
 572                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 573                        warning("Cannot prepare timestamp regexp %s",
 574                                stamp_regexp);
 575                        return 0;
 576                }
 577        }
 578
 579        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 580        if (status) {
 581                if (status != REG_NOMATCH)
 582                        warning("regexec returned %d for input: %s",
 583                                status, timestamp);
 584                return 0;
 585        }
 586
 587        zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10);
 588        zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 589        if (timestamp[m[3].rm_so] == '-')
 590                zoneoffset = -zoneoffset;
 591
 592        /*
 593         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 594         * (west of GMT) or 1970-01-01 (east of GMT)
 595         */
 596        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 597            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 598                return 0;
 599
 600        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 601                      strtol(timestamp + 14, NULL, 10) -
 602                      zoneoffset);
 603
 604        return ((zoneoffset < 0 && hourminute == 1440) ||
 605                (0 <= zoneoffset && !hourminute));
 606}
 607
 608/*
 609 * Get the name etc info from the ---/+++ lines of a traditional patch header
 610 *
 611 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 612 * files, we can happily check the index for a match, but for creating a
 613 * new file we should try to match whatever "patch" does. I have no idea.
 614 */
 615static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 616{
 617        char *name;
 618
 619        first += 4;     /* skip "--- " */
 620        second += 4;    /* skip "+++ " */
 621        if (!p_value_known) {
 622                int p, q;
 623                p = guess_p_value(first);
 624                q = guess_p_value(second);
 625                if (p < 0) p = q;
 626                if (0 <= p && p == q) {
 627                        p_value = p;
 628                        p_value_known = 1;
 629                }
 630        }
 631        if (is_dev_null(first)) {
 632                patch->is_new = 1;
 633                patch->is_delete = 0;
 634                name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
 635                patch->new_name = name;
 636        } else if (is_dev_null(second)) {
 637                patch->is_new = 0;
 638                patch->is_delete = 1;
 639                name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 640                patch->old_name = name;
 641        } else {
 642                name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 643                name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
 644                if (has_epoch_timestamp(first)) {
 645                        patch->is_new = 1;
 646                        patch->is_delete = 0;
 647                        patch->new_name = name;
 648                } else if (has_epoch_timestamp(second)) {
 649                        patch->is_new = 0;
 650                        patch->is_delete = 1;
 651                        patch->old_name = name;
 652                } else {
 653                        patch->old_name = patch->new_name = name;
 654                }
 655        }
 656        if (!name)
 657                die("unable to find filename in patch at line %d", linenr);
 658}
 659
 660static int gitdiff_hdrend(const char *line, struct patch *patch)
 661{
 662        return -1;
 663}
 664
 665/*
 666 * We're anal about diff header consistency, to make
 667 * sure that we don't end up having strange ambiguous
 668 * patches floating around.
 669 *
 670 * As a result, gitdiff_{old|new}name() will check
 671 * their names against any previous information, just
 672 * to make sure..
 673 */
 674static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
 675{
 676        if (!orig_name && !isnull)
 677                return find_name(line, NULL, p_value, TERM_TAB);
 678
 679        if (orig_name) {
 680                int len;
 681                const char *name;
 682                char *another;
 683                name = orig_name;
 684                len = strlen(name);
 685                if (isnull)
 686                        die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
 687                another = find_name(line, NULL, p_value, TERM_TAB);
 688                if (!another || memcmp(another, name, len))
 689                        die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
 690                free(another);
 691                return orig_name;
 692        }
 693        else {
 694                /* expect "/dev/null" */
 695                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 696                        die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
 697                return NULL;
 698        }
 699}
 700
 701static int gitdiff_oldname(const char *line, struct patch *patch)
 702{
 703        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
 704        return 0;
 705}
 706
 707static int gitdiff_newname(const char *line, struct patch *patch)
 708{
 709        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
 710        return 0;
 711}
 712
 713static int gitdiff_oldmode(const char *line, struct patch *patch)
 714{
 715        patch->old_mode = strtoul(line, NULL, 8);
 716        return 0;
 717}
 718
 719static int gitdiff_newmode(const char *line, struct patch *patch)
 720{
 721        patch->new_mode = strtoul(line, NULL, 8);
 722        return 0;
 723}
 724
 725static int gitdiff_delete(const char *line, struct patch *patch)
 726{
 727        patch->is_delete = 1;
 728        patch->old_name = patch->def_name;
 729        return gitdiff_oldmode(line, patch);
 730}
 731
 732static int gitdiff_newfile(const char *line, struct patch *patch)
 733{
 734        patch->is_new = 1;
 735        patch->new_name = patch->def_name;
 736        return gitdiff_newmode(line, patch);
 737}
 738
 739static int gitdiff_copysrc(const char *line, struct patch *patch)
 740{
 741        patch->is_copy = 1;
 742        patch->old_name = find_name(line, NULL, 0, 0);
 743        return 0;
 744}
 745
 746static int gitdiff_copydst(const char *line, struct patch *patch)
 747{
 748        patch->is_copy = 1;
 749        patch->new_name = find_name(line, NULL, 0, 0);
 750        return 0;
 751}
 752
 753static int gitdiff_renamesrc(const char *line, struct patch *patch)
 754{
 755        patch->is_rename = 1;
 756        patch->old_name = find_name(line, NULL, 0, 0);
 757        return 0;
 758}
 759
 760static int gitdiff_renamedst(const char *line, struct patch *patch)
 761{
 762        patch->is_rename = 1;
 763        patch->new_name = find_name(line, NULL, 0, 0);
 764        return 0;
 765}
 766
 767static int gitdiff_similarity(const char *line, struct patch *patch)
 768{
 769        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 770                patch->score = 0;
 771        return 0;
 772}
 773
 774static int gitdiff_dissimilarity(const char *line, struct patch *patch)
 775{
 776        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 777                patch->score = 0;
 778        return 0;
 779}
 780
 781static int gitdiff_index(const char *line, struct patch *patch)
 782{
 783        /*
 784         * index line is N hexadecimal, "..", N hexadecimal,
 785         * and optional space with octal mode.
 786         */
 787        const char *ptr, *eol;
 788        int len;
 789
 790        ptr = strchr(line, '.');
 791        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
 792                return 0;
 793        len = ptr - line;
 794        memcpy(patch->old_sha1_prefix, line, len);
 795        patch->old_sha1_prefix[len] = 0;
 796
 797        line = ptr + 2;
 798        ptr = strchr(line, ' ');
 799        eol = strchr(line, '\n');
 800
 801        if (!ptr || eol < ptr)
 802                ptr = eol;
 803        len = ptr - line;
 804
 805        if (40 < len)
 806                return 0;
 807        memcpy(patch->new_sha1_prefix, line, len);
 808        patch->new_sha1_prefix[len] = 0;
 809        if (*ptr == ' ')
 810                patch->old_mode = strtoul(ptr+1, NULL, 8);
 811        return 0;
 812}
 813
 814/*
 815 * This is normal for a diff that doesn't change anything: we'll fall through
 816 * into the next diff. Tell the parser to break out.
 817 */
 818static int gitdiff_unrecognized(const char *line, struct patch *patch)
 819{
 820        return -1;
 821}
 822
 823static const char *stop_at_slash(const char *line, int llen)
 824{
 825        int i;
 826
 827        for (i = 0; i < llen; i++) {
 828                int ch = line[i];
 829                if (ch == '/')
 830                        return line + i;
 831        }
 832        return NULL;
 833}
 834
 835/*
 836 * This is to extract the same name that appears on "diff --git"
 837 * line.  We do not find and return anything if it is a rename
 838 * patch, and it is OK because we will find the name elsewhere.
 839 * We need to reliably find name only when it is mode-change only,
 840 * creation or deletion of an empty file.  In any of these cases,
 841 * both sides are the same name under a/ and b/ respectively.
 842 */
 843static char *git_header_name(char *line, int llen)
 844{
 845        const char *name;
 846        const char *second = NULL;
 847        size_t len;
 848
 849        line += strlen("diff --git ");
 850        llen -= strlen("diff --git ");
 851
 852        if (*line == '"') {
 853                const char *cp;
 854                struct strbuf first = STRBUF_INIT;
 855                struct strbuf sp = STRBUF_INIT;
 856
 857                if (unquote_c_style(&first, line, &second))
 858                        goto free_and_fail1;
 859
 860                /* advance to the first slash */
 861                cp = stop_at_slash(first.buf, first.len);
 862                /* we do not accept absolute paths */
 863                if (!cp || cp == first.buf)
 864                        goto free_and_fail1;
 865                strbuf_remove(&first, 0, cp + 1 - first.buf);
 866
 867                /*
 868                 * second points at one past closing dq of name.
 869                 * find the second name.
 870                 */
 871                while ((second < line + llen) && isspace(*second))
 872                        second++;
 873
 874                if (line + llen <= second)
 875                        goto free_and_fail1;
 876                if (*second == '"') {
 877                        if (unquote_c_style(&sp, second, NULL))
 878                                goto free_and_fail1;
 879                        cp = stop_at_slash(sp.buf, sp.len);
 880                        if (!cp || cp == sp.buf)
 881                                goto free_and_fail1;
 882                        /* They must match, otherwise ignore */
 883                        if (strcmp(cp + 1, first.buf))
 884                                goto free_and_fail1;
 885                        strbuf_release(&sp);
 886                        return strbuf_detach(&first, NULL);
 887                }
 888
 889                /* unquoted second */
 890                cp = stop_at_slash(second, line + llen - second);
 891                if (!cp || cp == second)
 892                        goto free_and_fail1;
 893                cp++;
 894                if (line + llen - cp != first.len + 1 ||
 895                    memcmp(first.buf, cp, first.len))
 896                        goto free_and_fail1;
 897                return strbuf_detach(&first, NULL);
 898
 899        free_and_fail1:
 900                strbuf_release(&first);
 901                strbuf_release(&sp);
 902                return NULL;
 903        }
 904
 905        /* unquoted first name */
 906        name = stop_at_slash(line, llen);
 907        if (!name || name == line)
 908                return NULL;
 909        name++;
 910
 911        /*
 912         * since the first name is unquoted, a dq if exists must be
 913         * the beginning of the second name.
 914         */
 915        for (second = name; second < line + llen; second++) {
 916                if (*second == '"') {
 917                        struct strbuf sp = STRBUF_INIT;
 918                        const char *np;
 919
 920                        if (unquote_c_style(&sp, second, NULL))
 921                                goto free_and_fail2;
 922
 923                        np = stop_at_slash(sp.buf, sp.len);
 924                        if (!np || np == sp.buf)
 925                                goto free_and_fail2;
 926                        np++;
 927
 928                        len = sp.buf + sp.len - np;
 929                        if (len < second - name &&
 930                            !strncmp(np, name, len) &&
 931                            isspace(name[len])) {
 932                                /* Good */
 933                                strbuf_remove(&sp, 0, np - sp.buf);
 934                                return strbuf_detach(&sp, NULL);
 935                        }
 936
 937                free_and_fail2:
 938                        strbuf_release(&sp);
 939                        return NULL;
 940                }
 941        }
 942
 943        /*
 944         * Accept a name only if it shows up twice, exactly the same
 945         * form.
 946         */
 947        for (len = 0 ; ; len++) {
 948                switch (name[len]) {
 949                default:
 950                        continue;
 951                case '\n':
 952                        return NULL;
 953                case '\t': case ' ':
 954                        second = name+len;
 955                        for (;;) {
 956                                char c = *second++;
 957                                if (c == '\n')
 958                                        return NULL;
 959                                if (c == '/')
 960                                        break;
 961                        }
 962                        if (second[len] == '\n' && !memcmp(name, second, len)) {
 963                                return xmemdupz(name, len);
 964                        }
 965                }
 966        }
 967}
 968
 969/* Verify that we recognize the lines following a git header */
 970static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
 971{
 972        unsigned long offset;
 973
 974        /* A git diff has explicit new/delete information, so we don't guess */
 975        patch->is_new = 0;
 976        patch->is_delete = 0;
 977
 978        /*
 979         * Some things may not have the old name in the
 980         * rest of the headers anywhere (pure mode changes,
 981         * or removing or adding empty files), so we get
 982         * the default name from the header.
 983         */
 984        patch->def_name = git_header_name(line, len);
 985        if (patch->def_name && root) {
 986                char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
 987                strcpy(s, root);
 988                strcpy(s + root_len, patch->def_name);
 989                free(patch->def_name);
 990                patch->def_name = s;
 991        }
 992
 993        line += len;
 994        size -= len;
 995        linenr++;
 996        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
 997                static const struct opentry {
 998                        const char *str;
 999                        int (*fn)(const char *, struct patch *);
1000                } optable[] = {
1001                        { "@@ -", gitdiff_hdrend },
1002                        { "--- ", gitdiff_oldname },
1003                        { "+++ ", gitdiff_newname },
1004                        { "old mode ", gitdiff_oldmode },
1005                        { "new mode ", gitdiff_newmode },
1006                        { "deleted file mode ", gitdiff_delete },
1007                        { "new file mode ", gitdiff_newfile },
1008                        { "copy from ", gitdiff_copysrc },
1009                        { "copy to ", gitdiff_copydst },
1010                        { "rename old ", gitdiff_renamesrc },
1011                        { "rename new ", gitdiff_renamedst },
1012                        { "rename from ", gitdiff_renamesrc },
1013                        { "rename to ", gitdiff_renamedst },
1014                        { "similarity index ", gitdiff_similarity },
1015                        { "dissimilarity index ", gitdiff_dissimilarity },
1016                        { "index ", gitdiff_index },
1017                        { "", gitdiff_unrecognized },
1018                };
1019                int i;
1020
1021                len = linelen(line, size);
1022                if (!len || line[len-1] != '\n')
1023                        break;
1024                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1025                        const struct opentry *p = optable + i;
1026                        int oplen = strlen(p->str);
1027                        if (len < oplen || memcmp(p->str, line, oplen))
1028                                continue;
1029                        if (p->fn(line + oplen, patch) < 0)
1030                                return offset;
1031                        break;
1032                }
1033        }
1034
1035        return offset;
1036}
1037
1038static int parse_num(const char *line, unsigned long *p)
1039{
1040        char *ptr;
1041
1042        if (!isdigit(*line))
1043                return 0;
1044        *p = strtoul(line, &ptr, 10);
1045        return ptr - line;
1046}
1047
1048static int parse_range(const char *line, int len, int offset, const char *expect,
1049                       unsigned long *p1, unsigned long *p2)
1050{
1051        int digits, ex;
1052
1053        if (offset < 0 || offset >= len)
1054                return -1;
1055        line += offset;
1056        len -= offset;
1057
1058        digits = parse_num(line, p1);
1059        if (!digits)
1060                return -1;
1061
1062        offset += digits;
1063        line += digits;
1064        len -= digits;
1065
1066        *p2 = 1;
1067        if (*line == ',') {
1068                digits = parse_num(line+1, p2);
1069                if (!digits)
1070                        return -1;
1071
1072                offset += digits+1;
1073                line += digits+1;
1074                len -= digits+1;
1075        }
1076
1077        ex = strlen(expect);
1078        if (ex > len)
1079                return -1;
1080        if (memcmp(line, expect, ex))
1081                return -1;
1082
1083        return offset + ex;
1084}
1085
1086static void recount_diff(char *line, int size, struct fragment *fragment)
1087{
1088        int oldlines = 0, newlines = 0, ret = 0;
1089
1090        if (size < 1) {
1091                warning("recount: ignore empty hunk");
1092                return;
1093        }
1094
1095        for (;;) {
1096                int len = linelen(line, size);
1097                size -= len;
1098                line += len;
1099
1100                if (size < 1)
1101                        break;
1102
1103                switch (*line) {
1104                case ' ': case '\n':
1105                        newlines++;
1106                        /* fall through */
1107                case '-':
1108                        oldlines++;
1109                        continue;
1110                case '+':
1111                        newlines++;
1112                        continue;
1113                case '\\':
1114                        continue;
1115                case '@':
1116                        ret = size < 3 || prefixcmp(line, "@@ ");
1117                        break;
1118                case 'd':
1119                        ret = size < 5 || prefixcmp(line, "diff ");
1120                        break;
1121                default:
1122                        ret = -1;
1123                        break;
1124                }
1125                if (ret) {
1126                        warning("recount: unexpected line: %.*s",
1127                                (int)linelen(line, size), line);
1128                        return;
1129                }
1130                break;
1131        }
1132        fragment->oldlines = oldlines;
1133        fragment->newlines = newlines;
1134}
1135
1136/*
1137 * Parse a unified diff fragment header of the
1138 * form "@@ -a,b +c,d @@"
1139 */
1140static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1141{
1142        int offset;
1143
1144        if (!len || line[len-1] != '\n')
1145                return -1;
1146
1147        /* Figure out the number of lines in a fragment */
1148        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1149        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1150
1151        return offset;
1152}
1153
1154static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1155{
1156        unsigned long offset, len;
1157
1158        patch->is_toplevel_relative = 0;
1159        patch->is_rename = patch->is_copy = 0;
1160        patch->is_new = patch->is_delete = -1;
1161        patch->old_mode = patch->new_mode = 0;
1162        patch->old_name = patch->new_name = NULL;
1163        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1164                unsigned long nextlen;
1165
1166                len = linelen(line, size);
1167                if (!len)
1168                        break;
1169
1170                /* Testing this early allows us to take a few shortcuts.. */
1171                if (len < 6)
1172                        continue;
1173
1174                /*
1175                 * Make sure we don't find any unconnected patch fragments.
1176                 * That's a sign that we didn't find a header, and that a
1177                 * patch has become corrupted/broken up.
1178                 */
1179                if (!memcmp("@@ -", line, 4)) {
1180                        struct fragment dummy;
1181                        if (parse_fragment_header(line, len, &dummy) < 0)
1182                                continue;
1183                        die("patch fragment without header at line %d: %.*s",
1184                            linenr, (int)len-1, line);
1185                }
1186
1187                if (size < len + 6)
1188                        break;
1189
1190                /*
1191                 * Git patch? It might not have a real patch, just a rename
1192                 * or mode change, so we handle that specially
1193                 */
1194                if (!memcmp("diff --git ", line, 11)) {
1195                        int git_hdr_len = parse_git_header(line, len, size, patch);
1196                        if (git_hdr_len <= len)
1197                                continue;
1198                        if (!patch->old_name && !patch->new_name) {
1199                                if (!patch->def_name)
1200                                        die("git diff header lacks filename information (line %d)", linenr);
1201                                patch->old_name = patch->new_name = patch->def_name;
1202                        }
1203                        patch->is_toplevel_relative = 1;
1204                        *hdrsize = git_hdr_len;
1205                        return offset;
1206                }
1207
1208                /* --- followed by +++ ? */
1209                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1210                        continue;
1211
1212                /*
1213                 * We only accept unified patches, so we want it to
1214                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1215                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1216                 */
1217                nextlen = linelen(line + len, size - len);
1218                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1219                        continue;
1220
1221                /* Ok, we'll consider it a patch */
1222                parse_traditional_patch(line, line+len, patch);
1223                *hdrsize = len + nextlen;
1224                linenr += 2;
1225                return offset;
1226        }
1227        return -1;
1228}
1229
1230static void check_whitespace(const char *line, int len, unsigned ws_rule)
1231{
1232        char *err;
1233        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1234        if (!result)
1235                return;
1236
1237        whitespace_error++;
1238        if (squelch_whitespace_errors &&
1239            squelch_whitespace_errors < whitespace_error)
1240                ;
1241        else {
1242                err = whitespace_error_string(result);
1243                fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1244                        patch_input_file, linenr, err, len - 2, line + 1);
1245                free(err);
1246        }
1247}
1248
1249/*
1250 * Parse a unified diff. Note that this really needs to parse each
1251 * fragment separately, since the only way to know the difference
1252 * between a "---" that is part of a patch, and a "---" that starts
1253 * the next patch is to look at the line counts..
1254 */
1255static int parse_fragment(char *line, unsigned long size,
1256                          struct patch *patch, struct fragment *fragment)
1257{
1258        int added, deleted;
1259        int len = linelen(line, size), offset;
1260        unsigned long oldlines, newlines;
1261        unsigned long leading, trailing;
1262
1263        offset = parse_fragment_header(line, len, fragment);
1264        if (offset < 0)
1265                return -1;
1266        if (offset > 0 && patch->recount)
1267                recount_diff(line + offset, size - offset, fragment);
1268        oldlines = fragment->oldlines;
1269        newlines = fragment->newlines;
1270        leading = 0;
1271        trailing = 0;
1272
1273        /* Parse the thing.. */
1274        line += len;
1275        size -= len;
1276        linenr++;
1277        added = deleted = 0;
1278        for (offset = len;
1279             0 < size;
1280             offset += len, size -= len, line += len, linenr++) {
1281                if (!oldlines && !newlines)
1282                        break;
1283                len = linelen(line, size);
1284                if (!len || line[len-1] != '\n')
1285                        return -1;
1286                switch (*line) {
1287                default:
1288                        return -1;
1289                case '\n': /* newer GNU diff, an empty context line */
1290                case ' ':
1291                        oldlines--;
1292                        newlines--;
1293                        if (!deleted && !added)
1294                                leading++;
1295                        trailing++;
1296                        break;
1297                case '-':
1298                        if (apply_in_reverse &&
1299                            ws_error_action != nowarn_ws_error)
1300                                check_whitespace(line, len, patch->ws_rule);
1301                        deleted++;
1302                        oldlines--;
1303                        trailing = 0;
1304                        break;
1305                case '+':
1306                        if (!apply_in_reverse &&
1307                            ws_error_action != nowarn_ws_error)
1308                                check_whitespace(line, len, patch->ws_rule);
1309                        added++;
1310                        newlines--;
1311                        trailing = 0;
1312                        break;
1313
1314                /*
1315                 * We allow "\ No newline at end of file". Depending
1316                 * on locale settings when the patch was produced we
1317                 * don't know what this line looks like. The only
1318                 * thing we do know is that it begins with "\ ".
1319                 * Checking for 12 is just for sanity check -- any
1320                 * l10n of "\ No newline..." is at least that long.
1321                 */
1322                case '\\':
1323                        if (len < 12 || memcmp(line, "\\ ", 2))
1324                                return -1;
1325                        break;
1326                }
1327        }
1328        if (oldlines || newlines)
1329                return -1;
1330        fragment->leading = leading;
1331        fragment->trailing = trailing;
1332
1333        /*
1334         * If a fragment ends with an incomplete line, we failed to include
1335         * it in the above loop because we hit oldlines == newlines == 0
1336         * before seeing it.
1337         */
1338        if (12 < size && !memcmp(line, "\\ ", 2))
1339                offset += linelen(line, size);
1340
1341        patch->lines_added += added;
1342        patch->lines_deleted += deleted;
1343
1344        if (0 < patch->is_new && oldlines)
1345                return error("new file depends on old contents");
1346        if (0 < patch->is_delete && newlines)
1347                return error("deleted file still has contents");
1348        return offset;
1349}
1350
1351static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1352{
1353        unsigned long offset = 0;
1354        unsigned long oldlines = 0, newlines = 0, context = 0;
1355        struct fragment **fragp = &patch->fragments;
1356
1357        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1358                struct fragment *fragment;
1359                int len;
1360
1361                fragment = xcalloc(1, sizeof(*fragment));
1362                len = parse_fragment(line, size, patch, fragment);
1363                if (len <= 0)
1364                        die("corrupt patch at line %d", linenr);
1365                fragment->patch = line;
1366                fragment->size = len;
1367                oldlines += fragment->oldlines;
1368                newlines += fragment->newlines;
1369                context += fragment->leading + fragment->trailing;
1370
1371                *fragp = fragment;
1372                fragp = &fragment->next;
1373
1374                offset += len;
1375                line += len;
1376                size -= len;
1377        }
1378
1379        /*
1380         * If something was removed (i.e. we have old-lines) it cannot
1381         * be creation, and if something was added it cannot be
1382         * deletion.  However, the reverse is not true; --unified=0
1383         * patches that only add are not necessarily creation even
1384         * though they do not have any old lines, and ones that only
1385         * delete are not necessarily deletion.
1386         *
1387         * Unfortunately, a real creation/deletion patch do _not_ have
1388         * any context line by definition, so we cannot safely tell it
1389         * apart with --unified=0 insanity.  At least if the patch has
1390         * more than one hunk it is not creation or deletion.
1391         */
1392        if (patch->is_new < 0 &&
1393            (oldlines || (patch->fragments && patch->fragments->next)))
1394                patch->is_new = 0;
1395        if (patch->is_delete < 0 &&
1396            (newlines || (patch->fragments && patch->fragments->next)))
1397                patch->is_delete = 0;
1398
1399        if (0 < patch->is_new && oldlines)
1400                die("new file %s depends on old contents", patch->new_name);
1401        if (0 < patch->is_delete && newlines)
1402                die("deleted file %s still has contents", patch->old_name);
1403        if (!patch->is_delete && !newlines && context)
1404                fprintf(stderr, "** warning: file %s becomes empty but "
1405                        "is not deleted\n", patch->new_name);
1406
1407        return offset;
1408}
1409
1410static inline int metadata_changes(struct patch *patch)
1411{
1412        return  patch->is_rename > 0 ||
1413                patch->is_copy > 0 ||
1414                patch->is_new > 0 ||
1415                patch->is_delete ||
1416                (patch->old_mode && patch->new_mode &&
1417                 patch->old_mode != patch->new_mode);
1418}
1419
1420static char *inflate_it(const void *data, unsigned long size,
1421                        unsigned long inflated_size)
1422{
1423        z_stream stream;
1424        void *out;
1425        int st;
1426
1427        memset(&stream, 0, sizeof(stream));
1428
1429        stream.next_in = (unsigned char *)data;
1430        stream.avail_in = size;
1431        stream.next_out = out = xmalloc(inflated_size);
1432        stream.avail_out = inflated_size;
1433        git_inflate_init(&stream);
1434        st = git_inflate(&stream, Z_FINISH);
1435        git_inflate_end(&stream);
1436        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1437                free(out);
1438                return NULL;
1439        }
1440        return out;
1441}
1442
1443static struct fragment *parse_binary_hunk(char **buf_p,
1444                                          unsigned long *sz_p,
1445                                          int *status_p,
1446                                          int *used_p)
1447{
1448        /*
1449         * Expect a line that begins with binary patch method ("literal"
1450         * or "delta"), followed by the length of data before deflating.
1451         * a sequence of 'length-byte' followed by base-85 encoded data
1452         * should follow, terminated by a newline.
1453         *
1454         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1455         * and we would limit the patch line to 66 characters,
1456         * so one line can fit up to 13 groups that would decode
1457         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1458         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1459         */
1460        int llen, used;
1461        unsigned long size = *sz_p;
1462        char *buffer = *buf_p;
1463        int patch_method;
1464        unsigned long origlen;
1465        char *data = NULL;
1466        int hunk_size = 0;
1467        struct fragment *frag;
1468
1469        llen = linelen(buffer, size);
1470        used = llen;
1471
1472        *status_p = 0;
1473
1474        if (!prefixcmp(buffer, "delta ")) {
1475                patch_method = BINARY_DELTA_DEFLATED;
1476                origlen = strtoul(buffer + 6, NULL, 10);
1477        }
1478        else if (!prefixcmp(buffer, "literal ")) {
1479                patch_method = BINARY_LITERAL_DEFLATED;
1480                origlen = strtoul(buffer + 8, NULL, 10);
1481        }
1482        else
1483                return NULL;
1484
1485        linenr++;
1486        buffer += llen;
1487        while (1) {
1488                int byte_length, max_byte_length, newsize;
1489                llen = linelen(buffer, size);
1490                used += llen;
1491                linenr++;
1492                if (llen == 1) {
1493                        /* consume the blank line */
1494                        buffer++;
1495                        size--;
1496                        break;
1497                }
1498                /*
1499                 * Minimum line is "A00000\n" which is 7-byte long,
1500                 * and the line length must be multiple of 5 plus 2.
1501                 */
1502                if ((llen < 7) || (llen-2) % 5)
1503                        goto corrupt;
1504                max_byte_length = (llen - 2) / 5 * 4;
1505                byte_length = *buffer;
1506                if ('A' <= byte_length && byte_length <= 'Z')
1507                        byte_length = byte_length - 'A' + 1;
1508                else if ('a' <= byte_length && byte_length <= 'z')
1509                        byte_length = byte_length - 'a' + 27;
1510                else
1511                        goto corrupt;
1512                /* if the input length was not multiple of 4, we would
1513                 * have filler at the end but the filler should never
1514                 * exceed 3 bytes
1515                 */
1516                if (max_byte_length < byte_length ||
1517                    byte_length <= max_byte_length - 4)
1518                        goto corrupt;
1519                newsize = hunk_size + byte_length;
1520                data = xrealloc(data, newsize);
1521                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1522                        goto corrupt;
1523                hunk_size = newsize;
1524                buffer += llen;
1525                size -= llen;
1526        }
1527
1528        frag = xcalloc(1, sizeof(*frag));
1529        frag->patch = inflate_it(data, hunk_size, origlen);
1530        if (!frag->patch)
1531                goto corrupt;
1532        free(data);
1533        frag->size = origlen;
1534        *buf_p = buffer;
1535        *sz_p = size;
1536        *used_p = used;
1537        frag->binary_patch_method = patch_method;
1538        return frag;
1539
1540 corrupt:
1541        free(data);
1542        *status_p = -1;
1543        error("corrupt binary patch at line %d: %.*s",
1544              linenr-1, llen-1, buffer);
1545        return NULL;
1546}
1547
1548static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1549{
1550        /*
1551         * We have read "GIT binary patch\n"; what follows is a line
1552         * that says the patch method (currently, either "literal" or
1553         * "delta") and the length of data before deflating; a
1554         * sequence of 'length-byte' followed by base-85 encoded data
1555         * follows.
1556         *
1557         * When a binary patch is reversible, there is another binary
1558         * hunk in the same format, starting with patch method (either
1559         * "literal" or "delta") with the length of data, and a sequence
1560         * of length-byte + base-85 encoded data, terminated with another
1561         * empty line.  This data, when applied to the postimage, produces
1562         * the preimage.
1563         */
1564        struct fragment *forward;
1565        struct fragment *reverse;
1566        int status;
1567        int used, used_1;
1568
1569        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1570        if (!forward && !status)
1571                /* there has to be one hunk (forward hunk) */
1572                return error("unrecognized binary patch at line %d", linenr-1);
1573        if (status)
1574                /* otherwise we already gave an error message */
1575                return status;
1576
1577        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1578        if (reverse)
1579                used += used_1;
1580        else if (status) {
1581                /*
1582                 * Not having reverse hunk is not an error, but having
1583                 * a corrupt reverse hunk is.
1584                 */
1585                free((void*) forward->patch);
1586                free(forward);
1587                return status;
1588        }
1589        forward->next = reverse;
1590        patch->fragments = forward;
1591        patch->is_binary = 1;
1592        return used;
1593}
1594
1595static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1596{
1597        int hdrsize, patchsize;
1598        int offset = find_header(buffer, size, &hdrsize, patch);
1599
1600        if (offset < 0)
1601                return offset;
1602
1603        patch->ws_rule = whitespace_rule(patch->new_name
1604                                         ? patch->new_name
1605                                         : patch->old_name);
1606
1607        patchsize = parse_single_patch(buffer + offset + hdrsize,
1608                                       size - offset - hdrsize, patch);
1609
1610        if (!patchsize) {
1611                static const char *binhdr[] = {
1612                        "Binary files ",
1613                        "Files ",
1614                        NULL,
1615                };
1616                static const char git_binary[] = "GIT binary patch\n";
1617                int i;
1618                int hd = hdrsize + offset;
1619                unsigned long llen = linelen(buffer + hd, size - hd);
1620
1621                if (llen == sizeof(git_binary) - 1 &&
1622                    !memcmp(git_binary, buffer + hd, llen)) {
1623                        int used;
1624                        linenr++;
1625                        used = parse_binary(buffer + hd + llen,
1626                                            size - hd - llen, patch);
1627                        if (used)
1628                                patchsize = used + llen;
1629                        else
1630                                patchsize = 0;
1631                }
1632                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1633                        for (i = 0; binhdr[i]; i++) {
1634                                int len = strlen(binhdr[i]);
1635                                if (len < size - hd &&
1636                                    !memcmp(binhdr[i], buffer + hd, len)) {
1637                                        linenr++;
1638                                        patch->is_binary = 1;
1639                                        patchsize = llen;
1640                                        break;
1641                                }
1642                        }
1643                }
1644
1645                /* Empty patch cannot be applied if it is a text patch
1646                 * without metadata change.  A binary patch appears
1647                 * empty to us here.
1648                 */
1649                if ((apply || check) &&
1650                    (!patch->is_binary && !metadata_changes(patch)))
1651                        die("patch with only garbage at line %d", linenr);
1652        }
1653
1654        return offset + hdrsize + patchsize;
1655}
1656
1657#define swap(a,b) myswap((a),(b),sizeof(a))
1658
1659#define myswap(a, b, size) do {         \
1660        unsigned char mytmp[size];      \
1661        memcpy(mytmp, &a, size);                \
1662        memcpy(&a, &b, size);           \
1663        memcpy(&b, mytmp, size);                \
1664} while (0)
1665
1666static void reverse_patches(struct patch *p)
1667{
1668        for (; p; p = p->next) {
1669                struct fragment *frag = p->fragments;
1670
1671                swap(p->new_name, p->old_name);
1672                swap(p->new_mode, p->old_mode);
1673                swap(p->is_new, p->is_delete);
1674                swap(p->lines_added, p->lines_deleted);
1675                swap(p->old_sha1_prefix, p->new_sha1_prefix);
1676
1677                for (; frag; frag = frag->next) {
1678                        swap(frag->newpos, frag->oldpos);
1679                        swap(frag->newlines, frag->oldlines);
1680                }
1681        }
1682}
1683
1684static const char pluses[] =
1685"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1686static const char minuses[]=
1687"----------------------------------------------------------------------";
1688
1689static void show_stats(struct patch *patch)
1690{
1691        struct strbuf qname = STRBUF_INIT;
1692        char *cp = patch->new_name ? patch->new_name : patch->old_name;
1693        int max, add, del;
1694
1695        quote_c_style(cp, &qname, NULL, 0);
1696
1697        /*
1698         * "scale" the filename
1699         */
1700        max = max_len;
1701        if (max > 50)
1702                max = 50;
1703
1704        if (qname.len > max) {
1705                cp = strchr(qname.buf + qname.len + 3 - max, '/');
1706                if (!cp)
1707                        cp = qname.buf + qname.len + 3 - max;
1708                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1709        }
1710
1711        if (patch->is_binary) {
1712                printf(" %-*s |  Bin\n", max, qname.buf);
1713                strbuf_release(&qname);
1714                return;
1715        }
1716
1717        printf(" %-*s |", max, qname.buf);
1718        strbuf_release(&qname);
1719
1720        /*
1721         * scale the add/delete
1722         */
1723        max = max + max_change > 70 ? 70 - max : max_change;
1724        add = patch->lines_added;
1725        del = patch->lines_deleted;
1726
1727        if (max_change > 0) {
1728                int total = ((add + del) * max + max_change / 2) / max_change;
1729                add = (add * max + max_change / 2) / max_change;
1730                del = total - add;
1731        }
1732        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1733                add, pluses, del, minuses);
1734}
1735
1736static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1737{
1738        switch (st->st_mode & S_IFMT) {
1739        case S_IFLNK:
1740                if (strbuf_readlink(buf, path, st->st_size) < 0)
1741                        return error("unable to read symlink %s", path);
1742                return 0;
1743        case S_IFREG:
1744                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1745                        return error("unable to open or read %s", path);
1746                convert_to_git(path, buf->buf, buf->len, buf, 0);
1747                return 0;
1748        default:
1749                return -1;
1750        }
1751}
1752
1753/*
1754 * Update the preimage, and the common lines in postimage,
1755 * from buffer buf of length len. If postlen is 0 the postimage
1756 * is updated in place, otherwise it's updated on a new buffer
1757 * of length postlen
1758 */
1759
1760static void update_pre_post_images(struct image *preimage,
1761                                   struct image *postimage,
1762                                   char *buf,
1763                                   size_t len, size_t postlen)
1764{
1765        int i, ctx;
1766        char *new, *old, *fixed;
1767        struct image fixed_preimage;
1768
1769        /*
1770         * Update the preimage with whitespace fixes.  Note that we
1771         * are not losing preimage->buf -- apply_one_fragment() will
1772         * free "oldlines".
1773         */
1774        prepare_image(&fixed_preimage, buf, len, 1);
1775        assert(fixed_preimage.nr == preimage->nr);
1776        for (i = 0; i < preimage->nr; i++)
1777                fixed_preimage.line[i].flag = preimage->line[i].flag;
1778        free(preimage->line_allocated);
1779        *preimage = fixed_preimage;
1780
1781        /*
1782         * Adjust the common context lines in postimage. This can be
1783         * done in-place when we are just doing whitespace fixing,
1784         * which does not make the string grow, but needs a new buffer
1785         * when ignoring whitespace causes the update, since in this case
1786         * we could have e.g. tabs converted to multiple spaces.
1787         * We trust the caller to tell us if the update can be done
1788         * in place (postlen==0) or not.
1789         */
1790        old = postimage->buf;
1791        if (postlen)
1792                new = postimage->buf = xmalloc(postlen);
1793        else
1794                new = old;
1795        fixed = preimage->buf;
1796        for (i = ctx = 0; i < postimage->nr; i++) {
1797                size_t len = postimage->line[i].len;
1798                if (!(postimage->line[i].flag & LINE_COMMON)) {
1799                        /* an added line -- no counterparts in preimage */
1800                        memmove(new, old, len);
1801                        old += len;
1802                        new += len;
1803                        continue;
1804                }
1805
1806                /* a common context -- skip it in the original postimage */
1807                old += len;
1808
1809                /* and find the corresponding one in the fixed preimage */
1810                while (ctx < preimage->nr &&
1811                       !(preimage->line[ctx].flag & LINE_COMMON)) {
1812                        fixed += preimage->line[ctx].len;
1813                        ctx++;
1814                }
1815                if (preimage->nr <= ctx)
1816                        die("oops");
1817
1818                /* and copy it in, while fixing the line length */
1819                len = preimage->line[ctx].len;
1820                memcpy(new, fixed, len);
1821                new += len;
1822                fixed += len;
1823                postimage->line[i].len = len;
1824                ctx++;
1825        }
1826
1827        /* Fix the length of the whole thing */
1828        postimage->len = new - postimage->buf;
1829}
1830
1831static int match_fragment(struct image *img,
1832                          struct image *preimage,
1833                          struct image *postimage,
1834                          unsigned long try,
1835                          int try_lno,
1836                          unsigned ws_rule,
1837                          int match_beginning, int match_end)
1838{
1839        int i;
1840        char *fixed_buf, *buf, *orig, *target;
1841
1842        if (preimage->nr + try_lno > img->nr)
1843                return 0;
1844
1845        if (match_beginning && try_lno)
1846                return 0;
1847
1848        if (match_end && preimage->nr + try_lno != img->nr)
1849                return 0;
1850
1851        /* Quick hash check */
1852        for (i = 0; i < preimage->nr; i++)
1853                if (preimage->line[i].hash != img->line[try_lno + i].hash)
1854                        return 0;
1855
1856        /*
1857         * Do we have an exact match?  If we were told to match
1858         * at the end, size must be exactly at try+fragsize,
1859         * otherwise try+fragsize must be still within the preimage,
1860         * and either case, the old piece should match the preimage
1861         * exactly.
1862         */
1863        if ((match_end
1864             ? (try + preimage->len == img->len)
1865             : (try + preimage->len <= img->len)) &&
1866            !memcmp(img->buf + try, preimage->buf, preimage->len))
1867                return 1;
1868
1869        /*
1870         * No exact match. If we are ignoring whitespace, run a line-by-line
1871         * fuzzy matching. We collect all the line length information because
1872         * we need it to adjust whitespace if we match.
1873         */
1874        if (ws_ignore_action == ignore_ws_change) {
1875                size_t imgoff = 0;
1876                size_t preoff = 0;
1877                size_t postlen = postimage->len;
1878                for (i = 0; i < preimage->nr; i++) {
1879                        size_t prelen = preimage->line[i].len;
1880                        size_t imglen = img->line[try_lno+i].len;
1881
1882                        if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
1883                                              preimage->buf + preoff, prelen))
1884                                return 0;
1885                        if (preimage->line[i].flag & LINE_COMMON)
1886                                postlen += imglen - prelen;
1887                        imgoff += imglen;
1888                        preoff += prelen;
1889                }
1890
1891                /*
1892                 * Ok, the preimage matches with whitespace fuzz. Update it and
1893                 * the common postimage lines to use the same whitespace as the
1894                 * target. imgoff now holds the true length of the target that
1895                 * matches the preimage, and we need to update the line lengths
1896                 * of the preimage to match the target ones.
1897                 */
1898                fixed_buf = xmalloc(imgoff);
1899                memcpy(fixed_buf, img->buf + try, imgoff);
1900                for (i = 0; i < preimage->nr; i++)
1901                        preimage->line[i].len = img->line[try_lno+i].len;
1902
1903                /*
1904                 * Update the preimage buffer and the postimage context lines.
1905                 */
1906                update_pre_post_images(preimage, postimage,
1907                                fixed_buf, imgoff, postlen);
1908                return 1;
1909        }
1910
1911        if (ws_error_action != correct_ws_error)
1912                return 0;
1913
1914        /*
1915         * The hunk does not apply byte-by-byte, but the hash says
1916         * it might with whitespace fuzz. We haven't been asked to
1917         * ignore whitespace, we were asked to correct whitespace
1918         * errors, so let's try matching after whitespace correction.
1919         */
1920        fixed_buf = xmalloc(preimage->len + 1);
1921        buf = fixed_buf;
1922        orig = preimage->buf;
1923        target = img->buf + try;
1924        for (i = 0; i < preimage->nr; i++) {
1925                size_t fixlen; /* length after fixing the preimage */
1926                size_t oldlen = preimage->line[i].len;
1927                size_t tgtlen = img->line[try_lno + i].len;
1928                size_t tgtfixlen; /* length after fixing the target line */
1929                char tgtfixbuf[1024], *tgtfix;
1930                int match;
1931
1932                /* Try fixing the line in the preimage */
1933                fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1934
1935                /* Try fixing the line in the target */
1936                if (sizeof(tgtfixbuf) > tgtlen)
1937                        tgtfix = tgtfixbuf;
1938                else
1939                        tgtfix = xmalloc(tgtlen);
1940                tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1941
1942                /*
1943                 * If they match, either the preimage was based on
1944                 * a version before our tree fixed whitespace breakage,
1945                 * or we are lacking a whitespace-fix patch the tree
1946                 * the preimage was based on already had (i.e. target
1947                 * has whitespace breakage, the preimage doesn't).
1948                 * In either case, we are fixing the whitespace breakages
1949                 * so we might as well take the fix together with their
1950                 * real change.
1951                 */
1952                match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1953
1954                if (tgtfix != tgtfixbuf)
1955                        free(tgtfix);
1956                if (!match)
1957                        goto unmatch_exit;
1958
1959                orig += oldlen;
1960                buf += fixlen;
1961                target += tgtlen;
1962        }
1963
1964        /*
1965         * Yes, the preimage is based on an older version that still
1966         * has whitespace breakages unfixed, and fixing them makes the
1967         * hunk match.  Update the context lines in the postimage.
1968         */
1969        update_pre_post_images(preimage, postimage,
1970                               fixed_buf, buf - fixed_buf, 0);
1971        return 1;
1972
1973 unmatch_exit:
1974        free(fixed_buf);
1975        return 0;
1976}
1977
1978static int find_pos(struct image *img,
1979                    struct image *preimage,
1980                    struct image *postimage,
1981                    int line,
1982                    unsigned ws_rule,
1983                    int match_beginning, int match_end)
1984{
1985        int i;
1986        unsigned long backwards, forwards, try;
1987        int backwards_lno, forwards_lno, try_lno;
1988
1989        if (preimage->nr > img->nr)
1990                return -1;
1991
1992        /*
1993         * If match_begining or match_end is specified, there is no
1994         * point starting from a wrong line that will never match and
1995         * wander around and wait for a match at the specified end.
1996         */
1997        if (match_beginning)
1998                line = 0;
1999        else if (match_end)
2000                line = img->nr - preimage->nr;
2001
2002        if (line > img->nr)
2003                line = img->nr;
2004
2005        try = 0;
2006        for (i = 0; i < line; i++)
2007                try += img->line[i].len;
2008
2009        /*
2010         * There's probably some smart way to do this, but I'll leave
2011         * that to the smart and beautiful people. I'm simple and stupid.
2012         */
2013        backwards = try;
2014        backwards_lno = line;
2015        forwards = try;
2016        forwards_lno = line;
2017        try_lno = line;
2018
2019        for (i = 0; ; i++) {
2020                if (match_fragment(img, preimage, postimage,
2021                                   try, try_lno, ws_rule,
2022                                   match_beginning, match_end))
2023                        return try_lno;
2024
2025        again:
2026                if (backwards_lno == 0 && forwards_lno == img->nr)
2027                        break;
2028
2029                if (i & 1) {
2030                        if (backwards_lno == 0) {
2031                                i++;
2032                                goto again;
2033                        }
2034                        backwards_lno--;
2035                        backwards -= img->line[backwards_lno].len;
2036                        try = backwards;
2037                        try_lno = backwards_lno;
2038                } else {
2039                        if (forwards_lno == img->nr) {
2040                                i++;
2041                                goto again;
2042                        }
2043                        forwards += img->line[forwards_lno].len;
2044                        forwards_lno++;
2045                        try = forwards;
2046                        try_lno = forwards_lno;
2047                }
2048
2049        }
2050        return -1;
2051}
2052
2053static void remove_first_line(struct image *img)
2054{
2055        img->buf += img->line[0].len;
2056        img->len -= img->line[0].len;
2057        img->line++;
2058        img->nr--;
2059}
2060
2061static void remove_last_line(struct image *img)
2062{
2063        img->len -= img->line[--img->nr].len;
2064}
2065
2066static void update_image(struct image *img,
2067                         int applied_pos,
2068                         struct image *preimage,
2069                         struct image *postimage)
2070{
2071        /*
2072         * remove the copy of preimage at offset in img
2073         * and replace it with postimage
2074         */
2075        int i, nr;
2076        size_t remove_count, insert_count, applied_at = 0;
2077        char *result;
2078
2079        for (i = 0; i < applied_pos; i++)
2080                applied_at += img->line[i].len;
2081
2082        remove_count = 0;
2083        for (i = 0; i < preimage->nr; i++)
2084                remove_count += img->line[applied_pos + i].len;
2085        insert_count = postimage->len;
2086
2087        /* Adjust the contents */
2088        result = xmalloc(img->len + insert_count - remove_count + 1);
2089        memcpy(result, img->buf, applied_at);
2090        memcpy(result + applied_at, postimage->buf, postimage->len);
2091        memcpy(result + applied_at + postimage->len,
2092               img->buf + (applied_at + remove_count),
2093               img->len - (applied_at + remove_count));
2094        free(img->buf);
2095        img->buf = result;
2096        img->len += insert_count - remove_count;
2097        result[img->len] = '\0';
2098
2099        /* Adjust the line table */
2100        nr = img->nr + postimage->nr - preimage->nr;
2101        if (preimage->nr < postimage->nr) {
2102                /*
2103                 * NOTE: this knows that we never call remove_first_line()
2104                 * on anything other than pre/post image.
2105                 */
2106                img->line = xrealloc(img->line, nr * sizeof(*img->line));
2107                img->line_allocated = img->line;
2108        }
2109        if (preimage->nr != postimage->nr)
2110                memmove(img->line + applied_pos + postimage->nr,
2111                        img->line + applied_pos + preimage->nr,
2112                        (img->nr - (applied_pos + preimage->nr)) *
2113                        sizeof(*img->line));
2114        memcpy(img->line + applied_pos,
2115               postimage->line,
2116               postimage->nr * sizeof(*img->line));
2117        img->nr = nr;
2118}
2119
2120static int apply_one_fragment(struct image *img, struct fragment *frag,
2121                              int inaccurate_eof, unsigned ws_rule)
2122{
2123        int match_beginning, match_end;
2124        const char *patch = frag->patch;
2125        int size = frag->size;
2126        char *old, *new, *oldlines, *newlines;
2127        int new_blank_lines_at_end = 0;
2128        unsigned long leading, trailing;
2129        int pos, applied_pos;
2130        struct image preimage;
2131        struct image postimage;
2132
2133        memset(&preimage, 0, sizeof(preimage));
2134        memset(&postimage, 0, sizeof(postimage));
2135        oldlines = xmalloc(size);
2136        newlines = xmalloc(size);
2137
2138        old = oldlines;
2139        new = newlines;
2140        while (size > 0) {
2141                char first;
2142                int len = linelen(patch, size);
2143                int plen, added;
2144                int added_blank_line = 0;
2145
2146                if (!len)
2147                        break;
2148
2149                /*
2150                 * "plen" is how much of the line we should use for
2151                 * the actual patch data. Normally we just remove the
2152                 * first character on the line, but if the line is
2153                 * followed by "\ No newline", then we also remove the
2154                 * last one (which is the newline, of course).
2155                 */
2156                plen = len - 1;
2157                if (len < size && patch[len] == '\\')
2158                        plen--;
2159                first = *patch;
2160                if (apply_in_reverse) {
2161                        if (first == '-')
2162                                first = '+';
2163                        else if (first == '+')
2164                                first = '-';
2165                }
2166
2167                switch (first) {
2168                case '\n':
2169                        /* Newer GNU diff, empty context line */
2170                        if (plen < 0)
2171                                /* ... followed by '\No newline'; nothing */
2172                                break;
2173                        *old++ = '\n';
2174                        *new++ = '\n';
2175                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2176                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2177                        break;
2178                case ' ':
2179                case '-':
2180                        memcpy(old, patch + 1, plen);
2181                        add_line_info(&preimage, old, plen,
2182                                      (first == ' ' ? LINE_COMMON : 0));
2183                        old += plen;
2184                        if (first == '-')
2185                                break;
2186                /* Fall-through for ' ' */
2187                case '+':
2188                        /* --no-add does not add new lines */
2189                        if (first == '+' && no_add)
2190                                break;
2191
2192                        if (first != '+' ||
2193                            !whitespace_error ||
2194                            ws_error_action != correct_ws_error) {
2195                                memcpy(new, patch + 1, plen);
2196                                added = plen;
2197                        }
2198                        else {
2199                                added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2200                        }
2201                        add_line_info(&postimage, new, added,
2202                                      (first == '+' ? 0 : LINE_COMMON));
2203                        new += added;
2204                        if (first == '+' &&
2205                            added == 1 && new[-1] == '\n')
2206                                added_blank_line = 1;
2207                        break;
2208                case '@': case '\\':
2209                        /* Ignore it, we already handled it */
2210                        break;
2211                default:
2212                        if (apply_verbosely)
2213                                error("invalid start of line: '%c'", first);
2214                        return -1;
2215                }
2216                if (added_blank_line)
2217                        new_blank_lines_at_end++;
2218                else
2219                        new_blank_lines_at_end = 0;
2220                patch += len;
2221                size -= len;
2222        }
2223        if (inaccurate_eof &&
2224            old > oldlines && old[-1] == '\n' &&
2225            new > newlines && new[-1] == '\n') {
2226                old--;
2227                new--;
2228        }
2229
2230        leading = frag->leading;
2231        trailing = frag->trailing;
2232
2233        /*
2234         * A hunk to change lines at the beginning would begin with
2235         * @@ -1,L +N,M @@
2236         * but we need to be careful.  -U0 that inserts before the second
2237         * line also has this pattern.
2238         *
2239         * And a hunk to add to an empty file would begin with
2240         * @@ -0,0 +N,M @@
2241         *
2242         * In other words, a hunk that is (frag->oldpos <= 1) with or
2243         * without leading context must match at the beginning.
2244         */
2245        match_beginning = (!frag->oldpos ||
2246                           (frag->oldpos == 1 && !unidiff_zero));
2247
2248        /*
2249         * A hunk without trailing lines must match at the end.
2250         * However, we simply cannot tell if a hunk must match end
2251         * from the lack of trailing lines if the patch was generated
2252         * with unidiff without any context.
2253         */
2254        match_end = !unidiff_zero && !trailing;
2255
2256        pos = frag->newpos ? (frag->newpos - 1) : 0;
2257        preimage.buf = oldlines;
2258        preimage.len = old - oldlines;
2259        postimage.buf = newlines;
2260        postimage.len = new - newlines;
2261        preimage.line = preimage.line_allocated;
2262        postimage.line = postimage.line_allocated;
2263
2264        for (;;) {
2265
2266                applied_pos = find_pos(img, &preimage, &postimage, pos,
2267                                       ws_rule, match_beginning, match_end);
2268
2269                if (applied_pos >= 0)
2270                        break;
2271
2272                /* Am I at my context limits? */
2273                if ((leading <= p_context) && (trailing <= p_context))
2274                        break;
2275                if (match_beginning || match_end) {
2276                        match_beginning = match_end = 0;
2277                        continue;
2278                }
2279
2280                /*
2281                 * Reduce the number of context lines; reduce both
2282                 * leading and trailing if they are equal otherwise
2283                 * just reduce the larger context.
2284                 */
2285                if (leading >= trailing) {
2286                        remove_first_line(&preimage);
2287                        remove_first_line(&postimage);
2288                        pos--;
2289                        leading--;
2290                }
2291                if (trailing > leading) {
2292                        remove_last_line(&preimage);
2293                        remove_last_line(&postimage);
2294                        trailing--;
2295                }
2296        }
2297
2298        if (applied_pos >= 0) {
2299                if (ws_error_action == correct_ws_error &&
2300                    new_blank_lines_at_end &&
2301                    postimage.nr + applied_pos == img->nr) {
2302                        /*
2303                         * If the patch application adds blank lines
2304                         * at the end, and if the patch applies at the
2305                         * end of the image, remove those added blank
2306                         * lines.
2307                         */
2308                        while (new_blank_lines_at_end--)
2309                                remove_last_line(&postimage);
2310                }
2311
2312                /*
2313                 * Warn if it was necessary to reduce the number
2314                 * of context lines.
2315                 */
2316                if ((leading != frag->leading) ||
2317                    (trailing != frag->trailing))
2318                        fprintf(stderr, "Context reduced to (%ld/%ld)"
2319                                " to apply fragment at %d\n",
2320                                leading, trailing, applied_pos+1);
2321                update_image(img, applied_pos, &preimage, &postimage);
2322        } else {
2323                if (apply_verbosely)
2324                        error("while searching for:\n%.*s",
2325                              (int)(old - oldlines), oldlines);
2326        }
2327
2328        free(oldlines);
2329        free(newlines);
2330        free(preimage.line_allocated);
2331        free(postimage.line_allocated);
2332
2333        return (applied_pos < 0);
2334}
2335
2336static int apply_binary_fragment(struct image *img, struct patch *patch)
2337{
2338        struct fragment *fragment = patch->fragments;
2339        unsigned long len;
2340        void *dst;
2341
2342        /* Binary patch is irreversible without the optional second hunk */
2343        if (apply_in_reverse) {
2344                if (!fragment->next)
2345                        return error("cannot reverse-apply a binary patch "
2346                                     "without the reverse hunk to '%s'",
2347                                     patch->new_name
2348                                     ? patch->new_name : patch->old_name);
2349                fragment = fragment->next;
2350        }
2351        switch (fragment->binary_patch_method) {
2352        case BINARY_DELTA_DEFLATED:
2353                dst = patch_delta(img->buf, img->len, fragment->patch,
2354                                  fragment->size, &len);
2355                if (!dst)
2356                        return -1;
2357                clear_image(img);
2358                img->buf = dst;
2359                img->len = len;
2360                return 0;
2361        case BINARY_LITERAL_DEFLATED:
2362                clear_image(img);
2363                img->len = fragment->size;
2364                img->buf = xmalloc(img->len+1);
2365                memcpy(img->buf, fragment->patch, img->len);
2366                img->buf[img->len] = '\0';
2367                return 0;
2368        }
2369        return -1;
2370}
2371
2372static int apply_binary(struct image *img, struct patch *patch)
2373{
2374        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2375        unsigned char sha1[20];
2376
2377        /*
2378         * For safety, we require patch index line to contain
2379         * full 40-byte textual SHA1 for old and new, at least for now.
2380         */
2381        if (strlen(patch->old_sha1_prefix) != 40 ||
2382            strlen(patch->new_sha1_prefix) != 40 ||
2383            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2384            get_sha1_hex(patch->new_sha1_prefix, sha1))
2385                return error("cannot apply binary patch to '%s' "
2386                             "without full index line", name);
2387
2388        if (patch->old_name) {
2389                /*
2390                 * See if the old one matches what the patch
2391                 * applies to.
2392                 */
2393                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2394                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2395                        return error("the patch applies to '%s' (%s), "
2396                                     "which does not match the "
2397                                     "current contents.",
2398                                     name, sha1_to_hex(sha1));
2399        }
2400        else {
2401                /* Otherwise, the old one must be empty. */
2402                if (img->len)
2403                        return error("the patch applies to an empty "
2404                                     "'%s' but it is not empty", name);
2405        }
2406
2407        get_sha1_hex(patch->new_sha1_prefix, sha1);
2408        if (is_null_sha1(sha1)) {
2409                clear_image(img);
2410                return 0; /* deletion patch */
2411        }
2412
2413        if (has_sha1_file(sha1)) {
2414                /* We already have the postimage */
2415                enum object_type type;
2416                unsigned long size;
2417                char *result;
2418
2419                result = read_sha1_file(sha1, &type, &size);
2420                if (!result)
2421                        return error("the necessary postimage %s for "
2422                                     "'%s' cannot be read",
2423                                     patch->new_sha1_prefix, name);
2424                clear_image(img);
2425                img->buf = result;
2426                img->len = size;
2427        } else {
2428                /*
2429                 * We have verified buf matches the preimage;
2430                 * apply the patch data to it, which is stored
2431                 * in the patch->fragments->{patch,size}.
2432                 */
2433                if (apply_binary_fragment(img, patch))
2434                        return error("binary patch does not apply to '%s'",
2435                                     name);
2436
2437                /* verify that the result matches */
2438                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2439                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2440                        return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2441                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2442        }
2443
2444        return 0;
2445}
2446
2447static int apply_fragments(struct image *img, struct patch *patch)
2448{
2449        struct fragment *frag = patch->fragments;
2450        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2451        unsigned ws_rule = patch->ws_rule;
2452        unsigned inaccurate_eof = patch->inaccurate_eof;
2453
2454        if (patch->is_binary)
2455                return apply_binary(img, patch);
2456
2457        while (frag) {
2458                if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2459                        error("patch failed: %s:%ld", name, frag->oldpos);
2460                        if (!apply_with_reject)
2461                                return -1;
2462                        frag->rejected = 1;
2463                }
2464                frag = frag->next;
2465        }
2466        return 0;
2467}
2468
2469static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2470{
2471        if (!ce)
2472                return 0;
2473
2474        if (S_ISGITLINK(ce->ce_mode)) {
2475                strbuf_grow(buf, 100);
2476                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2477        } else {
2478                enum object_type type;
2479                unsigned long sz;
2480                char *result;
2481
2482                result = read_sha1_file(ce->sha1, &type, &sz);
2483                if (!result)
2484                        return -1;
2485                /* XXX read_sha1_file NUL-terminates */
2486                strbuf_attach(buf, result, sz, sz + 1);
2487        }
2488        return 0;
2489}
2490
2491static struct patch *in_fn_table(const char *name)
2492{
2493        struct string_list_item *item;
2494
2495        if (name == NULL)
2496                return NULL;
2497
2498        item = string_list_lookup(name, &fn_table);
2499        if (item != NULL)
2500                return (struct patch *)item->util;
2501
2502        return NULL;
2503}
2504
2505/*
2506 * item->util in the filename table records the status of the path.
2507 * Usually it points at a patch (whose result records the contents
2508 * of it after applying it), but it could be PATH_WAS_DELETED for a
2509 * path that a previously applied patch has already removed.
2510 */
2511 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2512#define PATH_WAS_DELETED ((struct patch *) -1)
2513
2514static int to_be_deleted(struct patch *patch)
2515{
2516        return patch == PATH_TO_BE_DELETED;
2517}
2518
2519static int was_deleted(struct patch *patch)
2520{
2521        return patch == PATH_WAS_DELETED;
2522}
2523
2524static void add_to_fn_table(struct patch *patch)
2525{
2526        struct string_list_item *item;
2527
2528        /*
2529         * Always add new_name unless patch is a deletion
2530         * This should cover the cases for normal diffs,
2531         * file creations and copies
2532         */
2533        if (patch->new_name != NULL) {
2534                item = string_list_insert(patch->new_name, &fn_table);
2535                item->util = patch;
2536        }
2537
2538        /*
2539         * store a failure on rename/deletion cases because
2540         * later chunks shouldn't patch old names
2541         */
2542        if ((patch->new_name == NULL) || (patch->is_rename)) {
2543                item = string_list_insert(patch->old_name, &fn_table);
2544                item->util = PATH_WAS_DELETED;
2545        }
2546}
2547
2548static void prepare_fn_table(struct patch *patch)
2549{
2550        /*
2551         * store information about incoming file deletion
2552         */
2553        while (patch) {
2554                if ((patch->new_name == NULL) || (patch->is_rename)) {
2555                        struct string_list_item *item;
2556                        item = string_list_insert(patch->old_name, &fn_table);
2557                        item->util = PATH_TO_BE_DELETED;
2558                }
2559                patch = patch->next;
2560        }
2561}
2562
2563static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2564{
2565        struct strbuf buf = STRBUF_INIT;
2566        struct image image;
2567        size_t len;
2568        char *img;
2569        struct patch *tpatch;
2570
2571        if (!(patch->is_copy || patch->is_rename) &&
2572            (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2573                if (was_deleted(tpatch)) {
2574                        return error("patch %s has been renamed/deleted",
2575                                patch->old_name);
2576                }
2577                /* We have a patched copy in memory use that */
2578                strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2579        } else if (cached) {
2580                if (read_file_or_gitlink(ce, &buf))
2581                        return error("read of %s failed", patch->old_name);
2582        } else if (patch->old_name) {
2583                if (S_ISGITLINK(patch->old_mode)) {
2584                        if (ce) {
2585                                read_file_or_gitlink(ce, &buf);
2586                        } else {
2587                                /*
2588                                 * There is no way to apply subproject
2589                                 * patch without looking at the index.
2590                                 */
2591                                patch->fragments = NULL;
2592                        }
2593                } else {
2594                        if (read_old_data(st, patch->old_name, &buf))
2595                                return error("read of %s failed", patch->old_name);
2596                }
2597        }
2598
2599        img = strbuf_detach(&buf, &len);
2600        prepare_image(&image, img, len, !patch->is_binary);
2601
2602        if (apply_fragments(&image, patch) < 0)
2603                return -1; /* note with --reject this succeeds. */
2604        patch->result = image.buf;
2605        patch->resultsize = image.len;
2606        add_to_fn_table(patch);
2607        free(image.line_allocated);
2608
2609        if (0 < patch->is_delete && patch->resultsize)
2610                return error("removal patch leaves file contents");
2611
2612        return 0;
2613}
2614
2615static int check_to_create_blob(const char *new_name, int ok_if_exists)
2616{
2617        struct stat nst;
2618        if (!lstat(new_name, &nst)) {
2619                if (S_ISDIR(nst.st_mode) || ok_if_exists)
2620                        return 0;
2621                /*
2622                 * A leading component of new_name might be a symlink
2623                 * that is going to be removed with this patch, but
2624                 * still pointing at somewhere that has the path.
2625                 * In such a case, path "new_name" does not exist as
2626                 * far as git is concerned.
2627                 */
2628                if (has_symlink_leading_path(new_name, strlen(new_name)))
2629                        return 0;
2630
2631                return error("%s: already exists in working directory", new_name);
2632        }
2633        else if ((errno != ENOENT) && (errno != ENOTDIR))
2634                return error("%s: %s", new_name, strerror(errno));
2635        return 0;
2636}
2637
2638static int verify_index_match(struct cache_entry *ce, struct stat *st)
2639{
2640        if (S_ISGITLINK(ce->ce_mode)) {
2641                if (!S_ISDIR(st->st_mode))
2642                        return -1;
2643                return 0;
2644        }
2645        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2646}
2647
2648static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2649{
2650        const char *old_name = patch->old_name;
2651        struct patch *tpatch = NULL;
2652        int stat_ret = 0;
2653        unsigned st_mode = 0;
2654
2655        /*
2656         * Make sure that we do not have local modifications from the
2657         * index when we are looking at the index.  Also make sure
2658         * we have the preimage file to be patched in the work tree,
2659         * unless --cached, which tells git to apply only in the index.
2660         */
2661        if (!old_name)
2662                return 0;
2663
2664        assert(patch->is_new <= 0);
2665
2666        if (!(patch->is_copy || patch->is_rename) &&
2667            (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2668                if (was_deleted(tpatch))
2669                        return error("%s: has been deleted/renamed", old_name);
2670                st_mode = tpatch->new_mode;
2671        } else if (!cached) {
2672                stat_ret = lstat(old_name, st);
2673                if (stat_ret && errno != ENOENT)
2674                        return error("%s: %s", old_name, strerror(errno));
2675        }
2676
2677        if (to_be_deleted(tpatch))
2678                tpatch = NULL;
2679
2680        if (check_index && !tpatch) {
2681                int pos = cache_name_pos(old_name, strlen(old_name));
2682                if (pos < 0) {
2683                        if (patch->is_new < 0)
2684                                goto is_new;
2685                        return error("%s: does not exist in index", old_name);
2686                }
2687                *ce = active_cache[pos];
2688                if (stat_ret < 0) {
2689                        struct checkout costate;
2690                        /* checkout */
2691                        costate.base_dir = "";
2692                        costate.base_dir_len = 0;
2693                        costate.force = 0;
2694                        costate.quiet = 0;
2695                        costate.not_new = 0;
2696                        costate.refresh_cache = 1;
2697                        if (checkout_entry(*ce, &costate, NULL) ||
2698                            lstat(old_name, st))
2699                                return -1;
2700                }
2701                if (!cached && verify_index_match(*ce, st))
2702                        return error("%s: does not match index", old_name);
2703                if (cached)
2704                        st_mode = (*ce)->ce_mode;
2705        } else if (stat_ret < 0) {
2706                if (patch->is_new < 0)
2707                        goto is_new;
2708                return error("%s: %s", old_name, strerror(errno));
2709        }
2710
2711        if (!cached && !tpatch)
2712                st_mode = ce_mode_from_stat(*ce, st->st_mode);
2713
2714        if (patch->is_new < 0)
2715                patch->is_new = 0;
2716        if (!patch->old_mode)
2717                patch->old_mode = st_mode;
2718        if ((st_mode ^ patch->old_mode) & S_IFMT)
2719                return error("%s: wrong type", old_name);
2720        if (st_mode != patch->old_mode)
2721                warning("%s has type %o, expected %o",
2722                        old_name, st_mode, patch->old_mode);
2723        if (!patch->new_mode && !patch->is_delete)
2724                patch->new_mode = st_mode;
2725        return 0;
2726
2727 is_new:
2728        patch->is_new = 1;
2729        patch->is_delete = 0;
2730        patch->old_name = NULL;
2731        return 0;
2732}
2733
2734static int check_patch(struct patch *patch)
2735{
2736        struct stat st;
2737        const char *old_name = patch->old_name;
2738        const char *new_name = patch->new_name;
2739        const char *name = old_name ? old_name : new_name;
2740        struct cache_entry *ce = NULL;
2741        struct patch *tpatch;
2742        int ok_if_exists;
2743        int status;
2744
2745        patch->rejected = 1; /* we will drop this after we succeed */
2746
2747        status = check_preimage(patch, &ce, &st);
2748        if (status)
2749                return status;
2750        old_name = patch->old_name;
2751
2752        if ((tpatch = in_fn_table(new_name)) &&
2753                        (was_deleted(tpatch) || to_be_deleted(tpatch)))
2754                /*
2755                 * A type-change diff is always split into a patch to
2756                 * delete old, immediately followed by a patch to
2757                 * create new (see diff.c::run_diff()); in such a case
2758                 * it is Ok that the entry to be deleted by the
2759                 * previous patch is still in the working tree and in
2760                 * the index.
2761                 */
2762                ok_if_exists = 1;
2763        else
2764                ok_if_exists = 0;
2765
2766        if (new_name &&
2767            ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2768                if (check_index &&
2769                    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2770                    !ok_if_exists)
2771                        return error("%s: already exists in index", new_name);
2772                if (!cached) {
2773                        int err = check_to_create_blob(new_name, ok_if_exists);
2774                        if (err)
2775                                return err;
2776                }
2777                if (!patch->new_mode) {
2778                        if (0 < patch->is_new)
2779                                patch->new_mode = S_IFREG | 0644;
2780                        else
2781                                patch->new_mode = patch->old_mode;
2782                }
2783        }
2784
2785        if (new_name && old_name) {
2786                int same = !strcmp(old_name, new_name);
2787                if (!patch->new_mode)
2788                        patch->new_mode = patch->old_mode;
2789                if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2790                        return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2791                                patch->new_mode, new_name, patch->old_mode,
2792                                same ? "" : " of ", same ? "" : old_name);
2793        }
2794
2795        if (apply_data(patch, &st, ce) < 0)
2796                return error("%s: patch does not apply", name);
2797        patch->rejected = 0;
2798        return 0;
2799}
2800
2801static int check_patch_list(struct patch *patch)
2802{
2803        int err = 0;
2804
2805        prepare_fn_table(patch);
2806        while (patch) {
2807                if (apply_verbosely)
2808                        say_patch_name(stderr,
2809                                       "Checking patch ", patch, "...\n");
2810                err |= check_patch(patch);
2811                patch = patch->next;
2812        }
2813        return err;
2814}
2815
2816/* This function tries to read the sha1 from the current index */
2817static int get_current_sha1(const char *path, unsigned char *sha1)
2818{
2819        int pos;
2820
2821        if (read_cache() < 0)
2822                return -1;
2823        pos = cache_name_pos(path, strlen(path));
2824        if (pos < 0)
2825                return -1;
2826        hashcpy(sha1, active_cache[pos]->sha1);
2827        return 0;
2828}
2829
2830/* Build an index that contains the just the files needed for a 3way merge */
2831static void build_fake_ancestor(struct patch *list, const char *filename)
2832{
2833        struct patch *patch;
2834        struct index_state result = { NULL };
2835        int fd;
2836
2837        /* Once we start supporting the reverse patch, it may be
2838         * worth showing the new sha1 prefix, but until then...
2839         */
2840        for (patch = list; patch; patch = patch->next) {
2841                const unsigned char *sha1_ptr;
2842                unsigned char sha1[20];
2843                struct cache_entry *ce;
2844                const char *name;
2845
2846                name = patch->old_name ? patch->old_name : patch->new_name;
2847                if (0 < patch->is_new)
2848                        continue;
2849                else if (get_sha1(patch->old_sha1_prefix, sha1))
2850                        /* git diff has no index line for mode/type changes */
2851                        if (!patch->lines_added && !patch->lines_deleted) {
2852                                if (get_current_sha1(patch->new_name, sha1) ||
2853                                    get_current_sha1(patch->old_name, sha1))
2854                                        die("mode change for %s, which is not "
2855                                                "in current HEAD", name);
2856                                sha1_ptr = sha1;
2857                        } else
2858                                die("sha1 information is lacking or useless "
2859                                        "(%s).", name);
2860                else
2861                        sha1_ptr = sha1;
2862
2863                ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2864                if (!ce)
2865                        die("make_cache_entry failed for path '%s'", name);
2866                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2867                        die ("Could not add %s to temporary index", name);
2868        }
2869
2870        fd = open(filename, O_WRONLY | O_CREAT, 0666);
2871        if (fd < 0 || write_index(&result, fd) || close(fd))
2872                die ("Could not write temporary index to %s", filename);
2873
2874        discard_index(&result);
2875}
2876
2877static void stat_patch_list(struct patch *patch)
2878{
2879        int files, adds, dels;
2880
2881        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2882                files++;
2883                adds += patch->lines_added;
2884                dels += patch->lines_deleted;
2885                show_stats(patch);
2886        }
2887
2888        printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2889}
2890
2891static void numstat_patch_list(struct patch *patch)
2892{
2893        for ( ; patch; patch = patch->next) {
2894                const char *name;
2895                name = patch->new_name ? patch->new_name : patch->old_name;
2896                if (patch->is_binary)
2897                        printf("-\t-\t");
2898                else
2899                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2900                write_name_quoted(name, stdout, line_termination);
2901        }
2902}
2903
2904static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2905{
2906        if (mode)
2907                printf(" %s mode %06o %s\n", newdelete, mode, name);
2908        else
2909                printf(" %s %s\n", newdelete, name);
2910}
2911
2912static void show_mode_change(struct patch *p, int show_name)
2913{
2914        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2915                if (show_name)
2916                        printf(" mode change %06o => %06o %s\n",
2917                               p->old_mode, p->new_mode, p->new_name);
2918                else
2919                        printf(" mode change %06o => %06o\n",
2920                               p->old_mode, p->new_mode);
2921        }
2922}
2923
2924static void show_rename_copy(struct patch *p)
2925{
2926        const char *renamecopy = p->is_rename ? "rename" : "copy";
2927        const char *old, *new;
2928
2929        /* Find common prefix */
2930        old = p->old_name;
2931        new = p->new_name;
2932        while (1) {
2933                const char *slash_old, *slash_new;
2934                slash_old = strchr(old, '/');
2935                slash_new = strchr(new, '/');
2936                if (!slash_old ||
2937                    !slash_new ||
2938                    slash_old - old != slash_new - new ||
2939                    memcmp(old, new, slash_new - new))
2940                        break;
2941                old = slash_old + 1;
2942                new = slash_new + 1;
2943        }
2944        /* p->old_name thru old is the common prefix, and old and new
2945         * through the end of names are renames
2946         */
2947        if (old != p->old_name)
2948                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2949                       (int)(old - p->old_name), p->old_name,
2950                       old, new, p->score);
2951        else
2952                printf(" %s %s => %s (%d%%)\n", renamecopy,
2953                       p->old_name, p->new_name, p->score);
2954        show_mode_change(p, 0);
2955}
2956
2957static void summary_patch_list(struct patch *patch)
2958{
2959        struct patch *p;
2960
2961        for (p = patch; p; p = p->next) {
2962                if (p->is_new)
2963                        show_file_mode_name("create", p->new_mode, p->new_name);
2964                else if (p->is_delete)
2965                        show_file_mode_name("delete", p->old_mode, p->old_name);
2966                else {
2967                        if (p->is_rename || p->is_copy)
2968                                show_rename_copy(p);
2969                        else {
2970                                if (p->score) {
2971                                        printf(" rewrite %s (%d%%)\n",
2972                                               p->new_name, p->score);
2973                                        show_mode_change(p, 0);
2974                                }
2975                                else
2976                                        show_mode_change(p, 1);
2977                        }
2978                }
2979        }
2980}
2981
2982static void patch_stats(struct patch *patch)
2983{
2984        int lines = patch->lines_added + patch->lines_deleted;
2985
2986        if (lines > max_change)
2987                max_change = lines;
2988        if (patch->old_name) {
2989                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2990                if (!len)
2991                        len = strlen(patch->old_name);
2992                if (len > max_len)
2993                        max_len = len;
2994        }
2995        if (patch->new_name) {
2996                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2997                if (!len)
2998                        len = strlen(patch->new_name);
2999                if (len > max_len)
3000                        max_len = len;
3001        }
3002}
3003
3004static void remove_file(struct patch *patch, int rmdir_empty)
3005{
3006        if (update_index) {
3007                if (remove_file_from_cache(patch->old_name) < 0)
3008                        die("unable to remove %s from index", patch->old_name);
3009        }
3010        if (!cached) {
3011                if (S_ISGITLINK(patch->old_mode)) {
3012                        if (rmdir(patch->old_name))
3013                                warning("unable to remove submodule %s",
3014                                        patch->old_name);
3015                } else if (!unlink_or_warn(patch->old_name) && rmdir_empty) {
3016                        remove_path(patch->old_name);
3017                }
3018        }
3019}
3020
3021static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3022{
3023        struct stat st;
3024        struct cache_entry *ce;
3025        int namelen = strlen(path);
3026        unsigned ce_size = cache_entry_size(namelen);
3027
3028        if (!update_index)
3029                return;
3030
3031        ce = xcalloc(1, ce_size);
3032        memcpy(ce->name, path, namelen);
3033        ce->ce_mode = create_ce_mode(mode);
3034        ce->ce_flags = namelen;
3035        if (S_ISGITLINK(mode)) {
3036                const char *s = buf;
3037
3038                if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3039                        die("corrupt patch for subproject %s", path);
3040        } else {
3041                if (!cached) {
3042                        if (lstat(path, &st) < 0)
3043                                die_errno("unable to stat newly created file '%s'",
3044                                          path);
3045                        fill_stat_cache_info(ce, &st);
3046                }
3047                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3048                        die("unable to create backing store for newly created file %s", path);
3049        }
3050        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3051                die("unable to add cache entry for %s", path);
3052}
3053
3054static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3055{
3056        int fd;
3057        struct strbuf nbuf = STRBUF_INIT;
3058
3059        if (S_ISGITLINK(mode)) {
3060                struct stat st;
3061                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3062                        return 0;
3063                return mkdir(path, 0777);
3064        }
3065
3066        if (has_symlinks && S_ISLNK(mode))
3067                /* Although buf:size is counted string, it also is NUL
3068                 * terminated.
3069                 */
3070                return symlink(buf, path);
3071
3072        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3073        if (fd < 0)
3074                return -1;
3075
3076        if (convert_to_working_tree(path, buf, size, &nbuf)) {
3077                size = nbuf.len;
3078                buf  = nbuf.buf;
3079        }
3080        write_or_die(fd, buf, size);
3081        strbuf_release(&nbuf);
3082
3083        if (close(fd) < 0)
3084                die_errno("closing file '%s'", path);
3085        return 0;
3086}
3087
3088/*
3089 * We optimistically assume that the directories exist,
3090 * which is true 99% of the time anyway. If they don't,
3091 * we create them and try again.
3092 */
3093static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3094{
3095        if (cached)
3096                return;
3097        if (!try_create_file(path, mode, buf, size))
3098                return;
3099
3100        if (errno == ENOENT) {
3101                if (safe_create_leading_directories(path))
3102                        return;
3103                if (!try_create_file(path, mode, buf, size))
3104                        return;
3105        }
3106
3107        if (errno == EEXIST || errno == EACCES) {
3108                /* We may be trying to create a file where a directory
3109                 * used to be.
3110                 */
3111                struct stat st;
3112                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3113                        errno = EEXIST;
3114        }
3115
3116        if (errno == EEXIST) {
3117                unsigned int nr = getpid();
3118
3119                for (;;) {
3120                        char newpath[PATH_MAX];
3121                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3122                        if (!try_create_file(newpath, mode, buf, size)) {
3123                                if (!rename(newpath, path))
3124                                        return;
3125                                unlink_or_warn(newpath);
3126                                break;
3127                        }
3128                        if (errno != EEXIST)
3129                                break;
3130                        ++nr;
3131                }
3132        }
3133        die_errno("unable to write file '%s' mode %o", path, mode);
3134}
3135
3136static void create_file(struct patch *patch)
3137{
3138        char *path = patch->new_name;
3139        unsigned mode = patch->new_mode;
3140        unsigned long size = patch->resultsize;
3141        char *buf = patch->result;
3142
3143        if (!mode)
3144                mode = S_IFREG | 0644;
3145        create_one_file(path, mode, buf, size);
3146        add_index_file(path, mode, buf, size);
3147}
3148
3149/* phase zero is to remove, phase one is to create */
3150static void write_out_one_result(struct patch *patch, int phase)
3151{
3152        if (patch->is_delete > 0) {
3153                if (phase == 0)
3154                        remove_file(patch, 1);
3155                return;
3156        }
3157        if (patch->is_new > 0 || patch->is_copy) {
3158                if (phase == 1)
3159                        create_file(patch);
3160                return;
3161        }
3162        /*
3163         * Rename or modification boils down to the same
3164         * thing: remove the old, write the new
3165         */
3166        if (phase == 0)
3167                remove_file(patch, patch->is_rename);
3168        if (phase == 1)
3169                create_file(patch);
3170}
3171
3172static int write_out_one_reject(struct patch *patch)
3173{
3174        FILE *rej;
3175        char namebuf[PATH_MAX];
3176        struct fragment *frag;
3177        int cnt = 0;
3178
3179        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3180                if (!frag->rejected)
3181                        continue;
3182                cnt++;
3183        }
3184
3185        if (!cnt) {
3186                if (apply_verbosely)
3187                        say_patch_name(stderr,
3188                                       "Applied patch ", patch, " cleanly.\n");
3189                return 0;
3190        }
3191
3192        /* This should not happen, because a removal patch that leaves
3193         * contents are marked "rejected" at the patch level.
3194         */
3195        if (!patch->new_name)
3196                die("internal error");
3197
3198        /* Say this even without --verbose */
3199        say_patch_name(stderr, "Applying patch ", patch, " with");
3200        fprintf(stderr, " %d rejects...\n", cnt);
3201
3202        cnt = strlen(patch->new_name);
3203        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3204                cnt = ARRAY_SIZE(namebuf) - 5;
3205                warning("truncating .rej filename to %.*s.rej",
3206                        cnt - 1, patch->new_name);
3207        }
3208        memcpy(namebuf, patch->new_name, cnt);
3209        memcpy(namebuf + cnt, ".rej", 5);
3210
3211        rej = fopen(namebuf, "w");
3212        if (!rej)
3213                return error("cannot open %s: %s", namebuf, strerror(errno));
3214
3215        /* Normal git tools never deal with .rej, so do not pretend
3216         * this is a git patch by saying --git nor give extended
3217         * headers.  While at it, maybe please "kompare" that wants
3218         * the trailing TAB and some garbage at the end of line ;-).
3219         */
3220        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3221                patch->new_name, patch->new_name);
3222        for (cnt = 1, frag = patch->fragments;
3223             frag;
3224             cnt++, frag = frag->next) {
3225                if (!frag->rejected) {
3226                        fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3227                        continue;
3228                }
3229                fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3230                fprintf(rej, "%.*s", frag->size, frag->patch);
3231                if (frag->patch[frag->size-1] != '\n')
3232                        fputc('\n', rej);
3233        }
3234        fclose(rej);
3235        return -1;
3236}
3237
3238static int write_out_results(struct patch *list, int skipped_patch)
3239{
3240        int phase;
3241        int errs = 0;
3242        struct patch *l;
3243
3244        if (!list && !skipped_patch)
3245                return error("No changes");
3246
3247        for (phase = 0; phase < 2; phase++) {
3248                l = list;
3249                while (l) {
3250                        if (l->rejected)
3251                                errs = 1;
3252                        else {
3253                                write_out_one_result(l, phase);
3254                                if (phase == 1 && write_out_one_reject(l))
3255                                        errs = 1;
3256                        }
3257                        l = l->next;
3258                }
3259        }
3260        return errs;
3261}
3262
3263static struct lock_file lock_file;
3264
3265static struct string_list limit_by_name;
3266static int has_include;
3267static void add_name_limit(const char *name, int exclude)
3268{
3269        struct string_list_item *it;
3270
3271        it = string_list_append(name, &limit_by_name);
3272        it->util = exclude ? NULL : (void *) 1;
3273}
3274
3275static int use_patch(struct patch *p)
3276{
3277        const char *pathname = p->new_name ? p->new_name : p->old_name;
3278        int i;
3279
3280        /* Paths outside are not touched regardless of "--include" */
3281        if (0 < prefix_length) {
3282                int pathlen = strlen(pathname);
3283                if (pathlen <= prefix_length ||
3284                    memcmp(prefix, pathname, prefix_length))
3285                        return 0;
3286        }
3287
3288        /* See if it matches any of exclude/include rule */
3289        for (i = 0; i < limit_by_name.nr; i++) {
3290                struct string_list_item *it = &limit_by_name.items[i];
3291                if (!fnmatch(it->string, pathname, 0))
3292                        return (it->util != NULL);
3293        }
3294
3295        /*
3296         * If we had any include, a path that does not match any rule is
3297         * not used.  Otherwise, we saw bunch of exclude rules (or none)
3298         * and such a path is used.
3299         */
3300        return !has_include;
3301}
3302
3303
3304static void prefix_one(char **name)
3305{
3306        char *old_name = *name;
3307        if (!old_name)
3308                return;
3309        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3310        free(old_name);
3311}
3312
3313static void prefix_patches(struct patch *p)
3314{
3315        if (!prefix || p->is_toplevel_relative)
3316                return;
3317        for ( ; p; p = p->next) {
3318                if (p->new_name == p->old_name) {
3319                        char *prefixed = p->new_name;
3320                        prefix_one(&prefixed);
3321                        p->new_name = p->old_name = prefixed;
3322                }
3323                else {
3324                        prefix_one(&p->new_name);
3325                        prefix_one(&p->old_name);
3326                }
3327        }
3328}
3329
3330#define INACCURATE_EOF  (1<<0)
3331#define RECOUNT         (1<<1)
3332
3333static int apply_patch(int fd, const char *filename, int options)
3334{
3335        size_t offset;
3336        struct strbuf buf = STRBUF_INIT;
3337        struct patch *list = NULL, **listp = &list;
3338        int skipped_patch = 0;
3339
3340        /* FIXME - memory leak when using multiple patch files as inputs */
3341        memset(&fn_table, 0, sizeof(struct string_list));
3342        patch_input_file = filename;
3343        read_patch_file(&buf, fd);
3344        offset = 0;
3345        while (offset < buf.len) {
3346                struct patch *patch;
3347                int nr;
3348
3349                patch = xcalloc(1, sizeof(*patch));
3350                patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3351                patch->recount =  !!(options & RECOUNT);
3352                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3353                if (nr < 0)
3354                        break;
3355                if (apply_in_reverse)
3356                        reverse_patches(patch);
3357                if (prefix)
3358                        prefix_patches(patch);
3359                if (use_patch(patch)) {
3360                        patch_stats(patch);
3361                        *listp = patch;
3362                        listp = &patch->next;
3363                }
3364                else {
3365                        /* perhaps free it a bit better? */
3366                        free(patch);
3367                        skipped_patch++;
3368                }
3369                offset += nr;
3370        }
3371
3372        if (whitespace_error && (ws_error_action == die_on_ws_error))
3373                apply = 0;
3374
3375        update_index = check_index && apply;
3376        if (update_index && newfd < 0)
3377                newfd = hold_locked_index(&lock_file, 1);
3378
3379        if (check_index) {
3380                if (read_cache() < 0)
3381                        die("unable to read index file");
3382        }
3383
3384        if ((check || apply) &&
3385            check_patch_list(list) < 0 &&
3386            !apply_with_reject)
3387                exit(1);
3388
3389        if (apply && write_out_results(list, skipped_patch))
3390                exit(1);
3391
3392        if (fake_ancestor)
3393                build_fake_ancestor(list, fake_ancestor);
3394
3395        if (diffstat)
3396                stat_patch_list(list);
3397
3398        if (numstat)
3399                numstat_patch_list(list);
3400
3401        if (summary)
3402                summary_patch_list(list);
3403
3404        strbuf_release(&buf);
3405        return 0;
3406}
3407
3408static int git_apply_config(const char *var, const char *value, void *cb)
3409{
3410        if (!strcmp(var, "apply.whitespace"))
3411                return git_config_string(&apply_default_whitespace, var, value);
3412        else if (!strcmp(var, "apply.ignorewhitespace"))
3413                return git_config_string(&apply_default_ignorewhitespace, var, value);
3414        return git_default_config(var, value, cb);
3415}
3416
3417static int option_parse_exclude(const struct option *opt,
3418                                const char *arg, int unset)
3419{
3420        add_name_limit(arg, 1);
3421        return 0;
3422}
3423
3424static int option_parse_include(const struct option *opt,
3425                                const char *arg, int unset)
3426{
3427        add_name_limit(arg, 0);
3428        has_include = 1;
3429        return 0;
3430}
3431
3432static int option_parse_p(const struct option *opt,
3433                          const char *arg, int unset)
3434{
3435        p_value = atoi(arg);
3436        p_value_known = 1;
3437        return 0;
3438}
3439
3440static int option_parse_z(const struct option *opt,
3441                          const char *arg, int unset)
3442{
3443        if (unset)
3444                line_termination = '\n';
3445        else
3446                line_termination = 0;
3447        return 0;
3448}
3449
3450static int option_parse_space_change(const struct option *opt,
3451                          const char *arg, int unset)
3452{
3453        if (unset)
3454                ws_ignore_action = ignore_ws_none;
3455        else
3456                ws_ignore_action = ignore_ws_change;
3457        return 0;
3458}
3459
3460static int option_parse_whitespace(const struct option *opt,
3461                                   const char *arg, int unset)
3462{
3463        const char **whitespace_option = opt->value;
3464
3465        *whitespace_option = arg;
3466        parse_whitespace_option(arg);
3467        return 0;
3468}
3469
3470static int option_parse_directory(const struct option *opt,
3471                                  const char *arg, int unset)
3472{
3473        root_len = strlen(arg);
3474        if (root_len && arg[root_len - 1] != '/') {
3475                char *new_root;
3476                root = new_root = xmalloc(root_len + 2);
3477                strcpy(new_root, arg);
3478                strcpy(new_root + root_len++, "/");
3479        } else
3480                root = arg;
3481        return 0;
3482}
3483
3484int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3485{
3486        int i;
3487        int errs = 0;
3488        int is_not_gitdir;
3489        int binary;
3490        int force_apply = 0;
3491
3492        const char *whitespace_option = NULL;
3493
3494        struct option builtin_apply_options[] = {
3495                { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3496                        "don't apply changes matching the given path",
3497                        0, option_parse_exclude },
3498                { OPTION_CALLBACK, 0, "include", NULL, "path",
3499                        "apply changes matching the given path",
3500                        0, option_parse_include },
3501                { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3502                        "remove <num> leading slashes from traditional diff paths",
3503                        0, option_parse_p },
3504                OPT_BOOLEAN(0, "no-add", &no_add,
3505                        "ignore additions made by the patch"),
3506                OPT_BOOLEAN(0, "stat", &diffstat,
3507                        "instead of applying the patch, output diffstat for the input"),
3508                { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3509                  NULL, "old option, now no-op",
3510                  PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3511                { OPTION_BOOLEAN, 0, "binary", &binary,
3512                  NULL, "old option, now no-op",
3513                  PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3514                OPT_BOOLEAN(0, "numstat", &numstat,
3515                        "shows number of added and deleted lines in decimal notation"),
3516                OPT_BOOLEAN(0, "summary", &summary,
3517                        "instead of applying the patch, output a summary for the input"),
3518                OPT_BOOLEAN(0, "check", &check,
3519                        "instead of applying the patch, see if the patch is applicable"),
3520                OPT_BOOLEAN(0, "index", &check_index,
3521                        "make sure the patch is applicable to the current index"),
3522                OPT_BOOLEAN(0, "cached", &cached,
3523                        "apply a patch without touching the working tree"),
3524                OPT_BOOLEAN(0, "apply", &force_apply,
3525                        "also apply the patch (use with --stat/--summary/--check)"),
3526                OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3527                        "build a temporary index based on embedded index information"),
3528                { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3529                        "paths are separated with NUL character",
3530                        PARSE_OPT_NOARG, option_parse_z },
3531                OPT_INTEGER('C', NULL, &p_context,
3532                                "ensure at least <n> lines of context match"),
3533                { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3534                        "detect new or modified lines that have whitespace errors",
3535                        0, option_parse_whitespace },
3536                { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3537                        "ignore changes in whitespace when finding context",
3538                        PARSE_OPT_NOARG, option_parse_space_change },
3539                { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3540                        "ignore changes in whitespace when finding context",
3541                        PARSE_OPT_NOARG, option_parse_space_change },
3542                OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3543                        "apply the patch in reverse"),
3544                OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3545                        "don't expect at least one line of context"),
3546                OPT_BOOLEAN(0, "reject", &apply_with_reject,
3547                        "leave the rejected hunks in corresponding *.rej files"),
3548                OPT__VERBOSE(&apply_verbosely),
3549                OPT_BIT(0, "inaccurate-eof", &options,
3550                        "tolerate incorrectly detected missing new-line at the end of file",
3551                        INACCURATE_EOF),
3552                OPT_BIT(0, "recount", &options,
3553                        "do not trust the line counts in the hunk headers",
3554                        RECOUNT),
3555                { OPTION_CALLBACK, 0, "directory", NULL, "root",
3556                        "prepend <root> to all filenames",
3557                        0, option_parse_directory },
3558                OPT_END()
3559        };
3560
3561        prefix = setup_git_directory_gently(&is_not_gitdir);
3562        prefix_length = prefix ? strlen(prefix) : 0;
3563        git_config(git_apply_config, NULL);
3564        if (apply_default_whitespace)
3565                parse_whitespace_option(apply_default_whitespace);
3566        if (apply_default_ignorewhitespace)
3567                parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3568
3569        argc = parse_options(argc, argv, prefix, builtin_apply_options,
3570                        apply_usage, 0);
3571
3572        if (apply_with_reject)
3573                apply = apply_verbosely = 1;
3574        if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3575                apply = 0;
3576        if (check_index && is_not_gitdir)
3577                die("--index outside a repository");
3578        if (cached) {
3579                if (is_not_gitdir)
3580                        die("--cached outside a repository");
3581                check_index = 1;
3582        }
3583        for (i = 0; i < argc; i++) {
3584                const char *arg = argv[i];
3585                int fd;
3586
3587                if (!strcmp(arg, "-")) {
3588                        errs |= apply_patch(0, "<stdin>", options);
3589                        read_stdin = 0;
3590                        continue;
3591                } else if (0 < prefix_length)
3592                        arg = prefix_filename(prefix, prefix_length, arg);
3593
3594                fd = open(arg, O_RDONLY);
3595                if (fd < 0)
3596                        die_errno("can't open patch '%s'", arg);
3597                read_stdin = 0;
3598                set_default_whitespace_mode(whitespace_option);
3599                errs |= apply_patch(fd, arg, options);
3600                close(fd);
3601        }
3602        set_default_whitespace_mode(whitespace_option);
3603        if (read_stdin)
3604                errs |= apply_patch(0, "<stdin>", options);
3605        if (whitespace_error) {
3606                if (squelch_whitespace_errors &&
3607                    squelch_whitespace_errors < whitespace_error) {
3608                        int squelched =
3609                                whitespace_error - squelch_whitespace_errors;
3610                        warning("squelched %d "
3611                                "whitespace error%s",
3612                                squelched,
3613                                squelched == 1 ? "" : "s");
3614                }
3615                if (ws_error_action == die_on_ws_error)
3616                        die("%d line%s add%s whitespace errors.",
3617                            whitespace_error,
3618                            whitespace_error == 1 ? "" : "s",
3619                            whitespace_error == 1 ? "s" : "");
3620                if (applied_after_fixing_ws && apply)
3621                        warning("%d line%s applied after"
3622                                " fixing whitespace errors.",
3623                                applied_after_fixing_ws,
3624                                applied_after_fixing_ws == 1 ? "" : "s");
3625                else if (whitespace_error)
3626                        warning("%d line%s add%s whitespace errors.",
3627                                whitespace_error,
3628                                whitespace_error == 1 ? "" : "s",
3629                                whitespace_error == 1 ? "s" : "");
3630        }
3631
3632        if (update_index) {
3633                if (write_cache(newfd, active_cache, active_nr) ||
3634                    commit_locked_index(&lock_file))
3635                        die("Unable to write new index file");
3636        }
3637
3638        return !!errs;
3639}