builtin-apply.con commit git apply: option to ignore whitespace differences (86c91f9)
   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 * Get the name etc info from the ---/+++ lines of a traditional patch header
 540 *
 541 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 542 * files, we can happily check the index for a match, but for creating a
 543 * new file we should try to match whatever "patch" does. I have no idea.
 544 */
 545static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 546{
 547        char *name;
 548
 549        first += 4;     /* skip "--- " */
 550        second += 4;    /* skip "+++ " */
 551        if (!p_value_known) {
 552                int p, q;
 553                p = guess_p_value(first);
 554                q = guess_p_value(second);
 555                if (p < 0) p = q;
 556                if (0 <= p && p == q) {
 557                        p_value = p;
 558                        p_value_known = 1;
 559                }
 560        }
 561        if (is_dev_null(first)) {
 562                patch->is_new = 1;
 563                patch->is_delete = 0;
 564                name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
 565                patch->new_name = name;
 566        } else if (is_dev_null(second)) {
 567                patch->is_new = 0;
 568                patch->is_delete = 1;
 569                name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 570                patch->old_name = name;
 571        } else {
 572                name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 573                name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
 574                patch->old_name = patch->new_name = name;
 575        }
 576        if (!name)
 577                die("unable to find filename in patch at line %d", linenr);
 578}
 579
 580static int gitdiff_hdrend(const char *line, struct patch *patch)
 581{
 582        return -1;
 583}
 584
 585/*
 586 * We're anal about diff header consistency, to make
 587 * sure that we don't end up having strange ambiguous
 588 * patches floating around.
 589 *
 590 * As a result, gitdiff_{old|new}name() will check
 591 * their names against any previous information, just
 592 * to make sure..
 593 */
 594static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
 595{
 596        if (!orig_name && !isnull)
 597                return find_name(line, NULL, p_value, TERM_TAB);
 598
 599        if (orig_name) {
 600                int len;
 601                const char *name;
 602                char *another;
 603                name = orig_name;
 604                len = strlen(name);
 605                if (isnull)
 606                        die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
 607                another = find_name(line, NULL, p_value, TERM_TAB);
 608                if (!another || memcmp(another, name, len))
 609                        die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
 610                free(another);
 611                return orig_name;
 612        }
 613        else {
 614                /* expect "/dev/null" */
 615                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 616                        die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
 617                return NULL;
 618        }
 619}
 620
 621static int gitdiff_oldname(const char *line, struct patch *patch)
 622{
 623        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
 624        return 0;
 625}
 626
 627static int gitdiff_newname(const char *line, struct patch *patch)
 628{
 629        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
 630        return 0;
 631}
 632
 633static int gitdiff_oldmode(const char *line, struct patch *patch)
 634{
 635        patch->old_mode = strtoul(line, NULL, 8);
 636        return 0;
 637}
 638
 639static int gitdiff_newmode(const char *line, struct patch *patch)
 640{
 641        patch->new_mode = strtoul(line, NULL, 8);
 642        return 0;
 643}
 644
 645static int gitdiff_delete(const char *line, struct patch *patch)
 646{
 647        patch->is_delete = 1;
 648        patch->old_name = patch->def_name;
 649        return gitdiff_oldmode(line, patch);
 650}
 651
 652static int gitdiff_newfile(const char *line, struct patch *patch)
 653{
 654        patch->is_new = 1;
 655        patch->new_name = patch->def_name;
 656        return gitdiff_newmode(line, patch);
 657}
 658
 659static int gitdiff_copysrc(const char *line, struct patch *patch)
 660{
 661        patch->is_copy = 1;
 662        patch->old_name = find_name(line, NULL, 0, 0);
 663        return 0;
 664}
 665
 666static int gitdiff_copydst(const char *line, struct patch *patch)
 667{
 668        patch->is_copy = 1;
 669        patch->new_name = find_name(line, NULL, 0, 0);
 670        return 0;
 671}
 672
 673static int gitdiff_renamesrc(const char *line, struct patch *patch)
 674{
 675        patch->is_rename = 1;
 676        patch->old_name = find_name(line, NULL, 0, 0);
 677        return 0;
 678}
 679
 680static int gitdiff_renamedst(const char *line, struct patch *patch)
 681{
 682        patch->is_rename = 1;
 683        patch->new_name = find_name(line, NULL, 0, 0);
 684        return 0;
 685}
 686
 687static int gitdiff_similarity(const char *line, struct patch *patch)
 688{
 689        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 690                patch->score = 0;
 691        return 0;
 692}
 693
 694static int gitdiff_dissimilarity(const char *line, struct patch *patch)
 695{
 696        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 697                patch->score = 0;
 698        return 0;
 699}
 700
 701static int gitdiff_index(const char *line, struct patch *patch)
 702{
 703        /*
 704         * index line is N hexadecimal, "..", N hexadecimal,
 705         * and optional space with octal mode.
 706         */
 707        const char *ptr, *eol;
 708        int len;
 709
 710        ptr = strchr(line, '.');
 711        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
 712                return 0;
 713        len = ptr - line;
 714        memcpy(patch->old_sha1_prefix, line, len);
 715        patch->old_sha1_prefix[len] = 0;
 716
 717        line = ptr + 2;
 718        ptr = strchr(line, ' ');
 719        eol = strchr(line, '\n');
 720
 721        if (!ptr || eol < ptr)
 722                ptr = eol;
 723        len = ptr - line;
 724
 725        if (40 < len)
 726                return 0;
 727        memcpy(patch->new_sha1_prefix, line, len);
 728        patch->new_sha1_prefix[len] = 0;
 729        if (*ptr == ' ')
 730                patch->old_mode = strtoul(ptr+1, NULL, 8);
 731        return 0;
 732}
 733
 734/*
 735 * This is normal for a diff that doesn't change anything: we'll fall through
 736 * into the next diff. Tell the parser to break out.
 737 */
 738static int gitdiff_unrecognized(const char *line, struct patch *patch)
 739{
 740        return -1;
 741}
 742
 743static const char *stop_at_slash(const char *line, int llen)
 744{
 745        int i;
 746
 747        for (i = 0; i < llen; i++) {
 748                int ch = line[i];
 749                if (ch == '/')
 750                        return line + i;
 751        }
 752        return NULL;
 753}
 754
 755/*
 756 * This is to extract the same name that appears on "diff --git"
 757 * line.  We do not find and return anything if it is a rename
 758 * patch, and it is OK because we will find the name elsewhere.
 759 * We need to reliably find name only when it is mode-change only,
 760 * creation or deletion of an empty file.  In any of these cases,
 761 * both sides are the same name under a/ and b/ respectively.
 762 */
 763static char *git_header_name(char *line, int llen)
 764{
 765        const char *name;
 766        const char *second = NULL;
 767        size_t len;
 768
 769        line += strlen("diff --git ");
 770        llen -= strlen("diff --git ");
 771
 772        if (*line == '"') {
 773                const char *cp;
 774                struct strbuf first = STRBUF_INIT;
 775                struct strbuf sp = STRBUF_INIT;
 776
 777                if (unquote_c_style(&first, line, &second))
 778                        goto free_and_fail1;
 779
 780                /* advance to the first slash */
 781                cp = stop_at_slash(first.buf, first.len);
 782                /* we do not accept absolute paths */
 783                if (!cp || cp == first.buf)
 784                        goto free_and_fail1;
 785                strbuf_remove(&first, 0, cp + 1 - first.buf);
 786
 787                /*
 788                 * second points at one past closing dq of name.
 789                 * find the second name.
 790                 */
 791                while ((second < line + llen) && isspace(*second))
 792                        second++;
 793
 794                if (line + llen <= second)
 795                        goto free_and_fail1;
 796                if (*second == '"') {
 797                        if (unquote_c_style(&sp, second, NULL))
 798                                goto free_and_fail1;
 799                        cp = stop_at_slash(sp.buf, sp.len);
 800                        if (!cp || cp == sp.buf)
 801                                goto free_and_fail1;
 802                        /* They must match, otherwise ignore */
 803                        if (strcmp(cp + 1, first.buf))
 804                                goto free_and_fail1;
 805                        strbuf_release(&sp);
 806                        return strbuf_detach(&first, NULL);
 807                }
 808
 809                /* unquoted second */
 810                cp = stop_at_slash(second, line + llen - second);
 811                if (!cp || cp == second)
 812                        goto free_and_fail1;
 813                cp++;
 814                if (line + llen - cp != first.len + 1 ||
 815                    memcmp(first.buf, cp, first.len))
 816                        goto free_and_fail1;
 817                return strbuf_detach(&first, NULL);
 818
 819        free_and_fail1:
 820                strbuf_release(&first);
 821                strbuf_release(&sp);
 822                return NULL;
 823        }
 824
 825        /* unquoted first name */
 826        name = stop_at_slash(line, llen);
 827        if (!name || name == line)
 828                return NULL;
 829        name++;
 830
 831        /*
 832         * since the first name is unquoted, a dq if exists must be
 833         * the beginning of the second name.
 834         */
 835        for (second = name; second < line + llen; second++) {
 836                if (*second == '"') {
 837                        struct strbuf sp = STRBUF_INIT;
 838                        const char *np;
 839
 840                        if (unquote_c_style(&sp, second, NULL))
 841                                goto free_and_fail2;
 842
 843                        np = stop_at_slash(sp.buf, sp.len);
 844                        if (!np || np == sp.buf)
 845                                goto free_and_fail2;
 846                        np++;
 847
 848                        len = sp.buf + sp.len - np;
 849                        if (len < second - name &&
 850                            !strncmp(np, name, len) &&
 851                            isspace(name[len])) {
 852                                /* Good */
 853                                strbuf_remove(&sp, 0, np - sp.buf);
 854                                return strbuf_detach(&sp, NULL);
 855                        }
 856
 857                free_and_fail2:
 858                        strbuf_release(&sp);
 859                        return NULL;
 860                }
 861        }
 862
 863        /*
 864         * Accept a name only if it shows up twice, exactly the same
 865         * form.
 866         */
 867        for (len = 0 ; ; len++) {
 868                switch (name[len]) {
 869                default:
 870                        continue;
 871                case '\n':
 872                        return NULL;
 873                case '\t': case ' ':
 874                        second = name+len;
 875                        for (;;) {
 876                                char c = *second++;
 877                                if (c == '\n')
 878                                        return NULL;
 879                                if (c == '/')
 880                                        break;
 881                        }
 882                        if (second[len] == '\n' && !memcmp(name, second, len)) {
 883                                return xmemdupz(name, len);
 884                        }
 885                }
 886        }
 887}
 888
 889/* Verify that we recognize the lines following a git header */
 890static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
 891{
 892        unsigned long offset;
 893
 894        /* A git diff has explicit new/delete information, so we don't guess */
 895        patch->is_new = 0;
 896        patch->is_delete = 0;
 897
 898        /*
 899         * Some things may not have the old name in the
 900         * rest of the headers anywhere (pure mode changes,
 901         * or removing or adding empty files), so we get
 902         * the default name from the header.
 903         */
 904        patch->def_name = git_header_name(line, len);
 905        if (patch->def_name && root) {
 906                char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
 907                strcpy(s, root);
 908                strcpy(s + root_len, patch->def_name);
 909                free(patch->def_name);
 910                patch->def_name = s;
 911        }
 912
 913        line += len;
 914        size -= len;
 915        linenr++;
 916        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
 917                static const struct opentry {
 918                        const char *str;
 919                        int (*fn)(const char *, struct patch *);
 920                } optable[] = {
 921                        { "@@ -", gitdiff_hdrend },
 922                        { "--- ", gitdiff_oldname },
 923                        { "+++ ", gitdiff_newname },
 924                        { "old mode ", gitdiff_oldmode },
 925                        { "new mode ", gitdiff_newmode },
 926                        { "deleted file mode ", gitdiff_delete },
 927                        { "new file mode ", gitdiff_newfile },
 928                        { "copy from ", gitdiff_copysrc },
 929                        { "copy to ", gitdiff_copydst },
 930                        { "rename old ", gitdiff_renamesrc },
 931                        { "rename new ", gitdiff_renamedst },
 932                        { "rename from ", gitdiff_renamesrc },
 933                        { "rename to ", gitdiff_renamedst },
 934                        { "similarity index ", gitdiff_similarity },
 935                        { "dissimilarity index ", gitdiff_dissimilarity },
 936                        { "index ", gitdiff_index },
 937                        { "", gitdiff_unrecognized },
 938                };
 939                int i;
 940
 941                len = linelen(line, size);
 942                if (!len || line[len-1] != '\n')
 943                        break;
 944                for (i = 0; i < ARRAY_SIZE(optable); i++) {
 945                        const struct opentry *p = optable + i;
 946                        int oplen = strlen(p->str);
 947                        if (len < oplen || memcmp(p->str, line, oplen))
 948                                continue;
 949                        if (p->fn(line + oplen, patch) < 0)
 950                                return offset;
 951                        break;
 952                }
 953        }
 954
 955        return offset;
 956}
 957
 958static int parse_num(const char *line, unsigned long *p)
 959{
 960        char *ptr;
 961
 962        if (!isdigit(*line))
 963                return 0;
 964        *p = strtoul(line, &ptr, 10);
 965        return ptr - line;
 966}
 967
 968static int parse_range(const char *line, int len, int offset, const char *expect,
 969                       unsigned long *p1, unsigned long *p2)
 970{
 971        int digits, ex;
 972
 973        if (offset < 0 || offset >= len)
 974                return -1;
 975        line += offset;
 976        len -= offset;
 977
 978        digits = parse_num(line, p1);
 979        if (!digits)
 980                return -1;
 981
 982        offset += digits;
 983        line += digits;
 984        len -= digits;
 985
 986        *p2 = 1;
 987        if (*line == ',') {
 988                digits = parse_num(line+1, p2);
 989                if (!digits)
 990                        return -1;
 991
 992                offset += digits+1;
 993                line += digits+1;
 994                len -= digits+1;
 995        }
 996
 997        ex = strlen(expect);
 998        if (ex > len)
 999                return -1;
1000        if (memcmp(line, expect, ex))
1001                return -1;
1002
1003        return offset + ex;
1004}
1005
1006static void recount_diff(char *line, int size, struct fragment *fragment)
1007{
1008        int oldlines = 0, newlines = 0, ret = 0;
1009
1010        if (size < 1) {
1011                warning("recount: ignore empty hunk");
1012                return;
1013        }
1014
1015        for (;;) {
1016                int len = linelen(line, size);
1017                size -= len;
1018                line += len;
1019
1020                if (size < 1)
1021                        break;
1022
1023                switch (*line) {
1024                case ' ': case '\n':
1025                        newlines++;
1026                        /* fall through */
1027                case '-':
1028                        oldlines++;
1029                        continue;
1030                case '+':
1031                        newlines++;
1032                        continue;
1033                case '\\':
1034                        continue;
1035                case '@':
1036                        ret = size < 3 || prefixcmp(line, "@@ ");
1037                        break;
1038                case 'd':
1039                        ret = size < 5 || prefixcmp(line, "diff ");
1040                        break;
1041                default:
1042                        ret = -1;
1043                        break;
1044                }
1045                if (ret) {
1046                        warning("recount: unexpected line: %.*s",
1047                                (int)linelen(line, size), line);
1048                        return;
1049                }
1050                break;
1051        }
1052        fragment->oldlines = oldlines;
1053        fragment->newlines = newlines;
1054}
1055
1056/*
1057 * Parse a unified diff fragment header of the
1058 * form "@@ -a,b +c,d @@"
1059 */
1060static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1061{
1062        int offset;
1063
1064        if (!len || line[len-1] != '\n')
1065                return -1;
1066
1067        /* Figure out the number of lines in a fragment */
1068        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1069        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1070
1071        return offset;
1072}
1073
1074static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1075{
1076        unsigned long offset, len;
1077
1078        patch->is_toplevel_relative = 0;
1079        patch->is_rename = patch->is_copy = 0;
1080        patch->is_new = patch->is_delete = -1;
1081        patch->old_mode = patch->new_mode = 0;
1082        patch->old_name = patch->new_name = NULL;
1083        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1084                unsigned long nextlen;
1085
1086                len = linelen(line, size);
1087                if (!len)
1088                        break;
1089
1090                /* Testing this early allows us to take a few shortcuts.. */
1091                if (len < 6)
1092                        continue;
1093
1094                /*
1095                 * Make sure we don't find any unconnected patch fragments.
1096                 * That's a sign that we didn't find a header, and that a
1097                 * patch has become corrupted/broken up.
1098                 */
1099                if (!memcmp("@@ -", line, 4)) {
1100                        struct fragment dummy;
1101                        if (parse_fragment_header(line, len, &dummy) < 0)
1102                                continue;
1103                        die("patch fragment without header at line %d: %.*s",
1104                            linenr, (int)len-1, line);
1105                }
1106
1107                if (size < len + 6)
1108                        break;
1109
1110                /*
1111                 * Git patch? It might not have a real patch, just a rename
1112                 * or mode change, so we handle that specially
1113                 */
1114                if (!memcmp("diff --git ", line, 11)) {
1115                        int git_hdr_len = parse_git_header(line, len, size, patch);
1116                        if (git_hdr_len <= len)
1117                                continue;
1118                        if (!patch->old_name && !patch->new_name) {
1119                                if (!patch->def_name)
1120                                        die("git diff header lacks filename information (line %d)", linenr);
1121                                patch->old_name = patch->new_name = patch->def_name;
1122                        }
1123                        patch->is_toplevel_relative = 1;
1124                        *hdrsize = git_hdr_len;
1125                        return offset;
1126                }
1127
1128                /* --- followed by +++ ? */
1129                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1130                        continue;
1131
1132                /*
1133                 * We only accept unified patches, so we want it to
1134                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1135                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1136                 */
1137                nextlen = linelen(line + len, size - len);
1138                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1139                        continue;
1140
1141                /* Ok, we'll consider it a patch */
1142                parse_traditional_patch(line, line+len, patch);
1143                *hdrsize = len + nextlen;
1144                linenr += 2;
1145                return offset;
1146        }
1147        return -1;
1148}
1149
1150static void check_whitespace(const char *line, int len, unsigned ws_rule)
1151{
1152        char *err;
1153        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1154        if (!result)
1155                return;
1156
1157        whitespace_error++;
1158        if (squelch_whitespace_errors &&
1159            squelch_whitespace_errors < whitespace_error)
1160                ;
1161        else {
1162                err = whitespace_error_string(result);
1163                fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1164                        patch_input_file, linenr, err, len - 2, line + 1);
1165                free(err);
1166        }
1167}
1168
1169/*
1170 * Parse a unified diff. Note that this really needs to parse each
1171 * fragment separately, since the only way to know the difference
1172 * between a "---" that is part of a patch, and a "---" that starts
1173 * the next patch is to look at the line counts..
1174 */
1175static int parse_fragment(char *line, unsigned long size,
1176                          struct patch *patch, struct fragment *fragment)
1177{
1178        int added, deleted;
1179        int len = linelen(line, size), offset;
1180        unsigned long oldlines, newlines;
1181        unsigned long leading, trailing;
1182
1183        offset = parse_fragment_header(line, len, fragment);
1184        if (offset < 0)
1185                return -1;
1186        if (offset > 0 && patch->recount)
1187                recount_diff(line + offset, size - offset, fragment);
1188        oldlines = fragment->oldlines;
1189        newlines = fragment->newlines;
1190        leading = 0;
1191        trailing = 0;
1192
1193        /* Parse the thing.. */
1194        line += len;
1195        size -= len;
1196        linenr++;
1197        added = deleted = 0;
1198        for (offset = len;
1199             0 < size;
1200             offset += len, size -= len, line += len, linenr++) {
1201                if (!oldlines && !newlines)
1202                        break;
1203                len = linelen(line, size);
1204                if (!len || line[len-1] != '\n')
1205                        return -1;
1206                switch (*line) {
1207                default:
1208                        return -1;
1209                case '\n': /* newer GNU diff, an empty context line */
1210                case ' ':
1211                        oldlines--;
1212                        newlines--;
1213                        if (!deleted && !added)
1214                                leading++;
1215                        trailing++;
1216                        break;
1217                case '-':
1218                        if (apply_in_reverse &&
1219                            ws_error_action != nowarn_ws_error)
1220                                check_whitespace(line, len, patch->ws_rule);
1221                        deleted++;
1222                        oldlines--;
1223                        trailing = 0;
1224                        break;
1225                case '+':
1226                        if (!apply_in_reverse &&
1227                            ws_error_action != nowarn_ws_error)
1228                                check_whitespace(line, len, patch->ws_rule);
1229                        added++;
1230                        newlines--;
1231                        trailing = 0;
1232                        break;
1233
1234                /*
1235                 * We allow "\ No newline at end of file". Depending
1236                 * on locale settings when the patch was produced we
1237                 * don't know what this line looks like. The only
1238                 * thing we do know is that it begins with "\ ".
1239                 * Checking for 12 is just for sanity check -- any
1240                 * l10n of "\ No newline..." is at least that long.
1241                 */
1242                case '\\':
1243                        if (len < 12 || memcmp(line, "\\ ", 2))
1244                                return -1;
1245                        break;
1246                }
1247        }
1248        if (oldlines || newlines)
1249                return -1;
1250        fragment->leading = leading;
1251        fragment->trailing = trailing;
1252
1253        /*
1254         * If a fragment ends with an incomplete line, we failed to include
1255         * it in the above loop because we hit oldlines == newlines == 0
1256         * before seeing it.
1257         */
1258        if (12 < size && !memcmp(line, "\\ ", 2))
1259                offset += linelen(line, size);
1260
1261        patch->lines_added += added;
1262        patch->lines_deleted += deleted;
1263
1264        if (0 < patch->is_new && oldlines)
1265                return error("new file depends on old contents");
1266        if (0 < patch->is_delete && newlines)
1267                return error("deleted file still has contents");
1268        return offset;
1269}
1270
1271static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1272{
1273        unsigned long offset = 0;
1274        unsigned long oldlines = 0, newlines = 0, context = 0;
1275        struct fragment **fragp = &patch->fragments;
1276
1277        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1278                struct fragment *fragment;
1279                int len;
1280
1281                fragment = xcalloc(1, sizeof(*fragment));
1282                len = parse_fragment(line, size, patch, fragment);
1283                if (len <= 0)
1284                        die("corrupt patch at line %d", linenr);
1285                fragment->patch = line;
1286                fragment->size = len;
1287                oldlines += fragment->oldlines;
1288                newlines += fragment->newlines;
1289                context += fragment->leading + fragment->trailing;
1290
1291                *fragp = fragment;
1292                fragp = &fragment->next;
1293
1294                offset += len;
1295                line += len;
1296                size -= len;
1297        }
1298
1299        /*
1300         * If something was removed (i.e. we have old-lines) it cannot
1301         * be creation, and if something was added it cannot be
1302         * deletion.  However, the reverse is not true; --unified=0
1303         * patches that only add are not necessarily creation even
1304         * though they do not have any old lines, and ones that only
1305         * delete are not necessarily deletion.
1306         *
1307         * Unfortunately, a real creation/deletion patch do _not_ have
1308         * any context line by definition, so we cannot safely tell it
1309         * apart with --unified=0 insanity.  At least if the patch has
1310         * more than one hunk it is not creation or deletion.
1311         */
1312        if (patch->is_new < 0 &&
1313            (oldlines || (patch->fragments && patch->fragments->next)))
1314                patch->is_new = 0;
1315        if (patch->is_delete < 0 &&
1316            (newlines || (patch->fragments && patch->fragments->next)))
1317                patch->is_delete = 0;
1318
1319        if (0 < patch->is_new && oldlines)
1320                die("new file %s depends on old contents", patch->new_name);
1321        if (0 < patch->is_delete && newlines)
1322                die("deleted file %s still has contents", patch->old_name);
1323        if (!patch->is_delete && !newlines && context)
1324                fprintf(stderr, "** warning: file %s becomes empty but "
1325                        "is not deleted\n", patch->new_name);
1326
1327        return offset;
1328}
1329
1330static inline int metadata_changes(struct patch *patch)
1331{
1332        return  patch->is_rename > 0 ||
1333                patch->is_copy > 0 ||
1334                patch->is_new > 0 ||
1335                patch->is_delete ||
1336                (patch->old_mode && patch->new_mode &&
1337                 patch->old_mode != patch->new_mode);
1338}
1339
1340static char *inflate_it(const void *data, unsigned long size,
1341                        unsigned long inflated_size)
1342{
1343        z_stream stream;
1344        void *out;
1345        int st;
1346
1347        memset(&stream, 0, sizeof(stream));
1348
1349        stream.next_in = (unsigned char *)data;
1350        stream.avail_in = size;
1351        stream.next_out = out = xmalloc(inflated_size);
1352        stream.avail_out = inflated_size;
1353        git_inflate_init(&stream);
1354        st = git_inflate(&stream, Z_FINISH);
1355        git_inflate_end(&stream);
1356        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1357                free(out);
1358                return NULL;
1359        }
1360        return out;
1361}
1362
1363static struct fragment *parse_binary_hunk(char **buf_p,
1364                                          unsigned long *sz_p,
1365                                          int *status_p,
1366                                          int *used_p)
1367{
1368        /*
1369         * Expect a line that begins with binary patch method ("literal"
1370         * or "delta"), followed by the length of data before deflating.
1371         * a sequence of 'length-byte' followed by base-85 encoded data
1372         * should follow, terminated by a newline.
1373         *
1374         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1375         * and we would limit the patch line to 66 characters,
1376         * so one line can fit up to 13 groups that would decode
1377         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1378         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1379         */
1380        int llen, used;
1381        unsigned long size = *sz_p;
1382        char *buffer = *buf_p;
1383        int patch_method;
1384        unsigned long origlen;
1385        char *data = NULL;
1386        int hunk_size = 0;
1387        struct fragment *frag;
1388
1389        llen = linelen(buffer, size);
1390        used = llen;
1391
1392        *status_p = 0;
1393
1394        if (!prefixcmp(buffer, "delta ")) {
1395                patch_method = BINARY_DELTA_DEFLATED;
1396                origlen = strtoul(buffer + 6, NULL, 10);
1397        }
1398        else if (!prefixcmp(buffer, "literal ")) {
1399                patch_method = BINARY_LITERAL_DEFLATED;
1400                origlen = strtoul(buffer + 8, NULL, 10);
1401        }
1402        else
1403                return NULL;
1404
1405        linenr++;
1406        buffer += llen;
1407        while (1) {
1408                int byte_length, max_byte_length, newsize;
1409                llen = linelen(buffer, size);
1410                used += llen;
1411                linenr++;
1412                if (llen == 1) {
1413                        /* consume the blank line */
1414                        buffer++;
1415                        size--;
1416                        break;
1417                }
1418                /*
1419                 * Minimum line is "A00000\n" which is 7-byte long,
1420                 * and the line length must be multiple of 5 plus 2.
1421                 */
1422                if ((llen < 7) || (llen-2) % 5)
1423                        goto corrupt;
1424                max_byte_length = (llen - 2) / 5 * 4;
1425                byte_length = *buffer;
1426                if ('A' <= byte_length && byte_length <= 'Z')
1427                        byte_length = byte_length - 'A' + 1;
1428                else if ('a' <= byte_length && byte_length <= 'z')
1429                        byte_length = byte_length - 'a' + 27;
1430                else
1431                        goto corrupt;
1432                /* if the input length was not multiple of 4, we would
1433                 * have filler at the end but the filler should never
1434                 * exceed 3 bytes
1435                 */
1436                if (max_byte_length < byte_length ||
1437                    byte_length <= max_byte_length - 4)
1438                        goto corrupt;
1439                newsize = hunk_size + byte_length;
1440                data = xrealloc(data, newsize);
1441                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1442                        goto corrupt;
1443                hunk_size = newsize;
1444                buffer += llen;
1445                size -= llen;
1446        }
1447
1448        frag = xcalloc(1, sizeof(*frag));
1449        frag->patch = inflate_it(data, hunk_size, origlen);
1450        if (!frag->patch)
1451                goto corrupt;
1452        free(data);
1453        frag->size = origlen;
1454        *buf_p = buffer;
1455        *sz_p = size;
1456        *used_p = used;
1457        frag->binary_patch_method = patch_method;
1458        return frag;
1459
1460 corrupt:
1461        free(data);
1462        *status_p = -1;
1463        error("corrupt binary patch at line %d: %.*s",
1464              linenr-1, llen-1, buffer);
1465        return NULL;
1466}
1467
1468static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1469{
1470        /*
1471         * We have read "GIT binary patch\n"; what follows is a line
1472         * that says the patch method (currently, either "literal" or
1473         * "delta") and the length of data before deflating; a
1474         * sequence of 'length-byte' followed by base-85 encoded data
1475         * follows.
1476         *
1477         * When a binary patch is reversible, there is another binary
1478         * hunk in the same format, starting with patch method (either
1479         * "literal" or "delta") with the length of data, and a sequence
1480         * of length-byte + base-85 encoded data, terminated with another
1481         * empty line.  This data, when applied to the postimage, produces
1482         * the preimage.
1483         */
1484        struct fragment *forward;
1485        struct fragment *reverse;
1486        int status;
1487        int used, used_1;
1488
1489        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1490        if (!forward && !status)
1491                /* there has to be one hunk (forward hunk) */
1492                return error("unrecognized binary patch at line %d", linenr-1);
1493        if (status)
1494                /* otherwise we already gave an error message */
1495                return status;
1496
1497        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1498        if (reverse)
1499                used += used_1;
1500        else if (status) {
1501                /*
1502                 * Not having reverse hunk is not an error, but having
1503                 * a corrupt reverse hunk is.
1504                 */
1505                free((void*) forward->patch);
1506                free(forward);
1507                return status;
1508        }
1509        forward->next = reverse;
1510        patch->fragments = forward;
1511        patch->is_binary = 1;
1512        return used;
1513}
1514
1515static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1516{
1517        int hdrsize, patchsize;
1518        int offset = find_header(buffer, size, &hdrsize, patch);
1519
1520        if (offset < 0)
1521                return offset;
1522
1523        patch->ws_rule = whitespace_rule(patch->new_name
1524                                         ? patch->new_name
1525                                         : patch->old_name);
1526
1527        patchsize = parse_single_patch(buffer + offset + hdrsize,
1528                                       size - offset - hdrsize, patch);
1529
1530        if (!patchsize) {
1531                static const char *binhdr[] = {
1532                        "Binary files ",
1533                        "Files ",
1534                        NULL,
1535                };
1536                static const char git_binary[] = "GIT binary patch\n";
1537                int i;
1538                int hd = hdrsize + offset;
1539                unsigned long llen = linelen(buffer + hd, size - hd);
1540
1541                if (llen == sizeof(git_binary) - 1 &&
1542                    !memcmp(git_binary, buffer + hd, llen)) {
1543                        int used;
1544                        linenr++;
1545                        used = parse_binary(buffer + hd + llen,
1546                                            size - hd - llen, patch);
1547                        if (used)
1548                                patchsize = used + llen;
1549                        else
1550                                patchsize = 0;
1551                }
1552                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1553                        for (i = 0; binhdr[i]; i++) {
1554                                int len = strlen(binhdr[i]);
1555                                if (len < size - hd &&
1556                                    !memcmp(binhdr[i], buffer + hd, len)) {
1557                                        linenr++;
1558                                        patch->is_binary = 1;
1559                                        patchsize = llen;
1560                                        break;
1561                                }
1562                        }
1563                }
1564
1565                /* Empty patch cannot be applied if it is a text patch
1566                 * without metadata change.  A binary patch appears
1567                 * empty to us here.
1568                 */
1569                if ((apply || check) &&
1570                    (!patch->is_binary && !metadata_changes(patch)))
1571                        die("patch with only garbage at line %d", linenr);
1572        }
1573
1574        return offset + hdrsize + patchsize;
1575}
1576
1577#define swap(a,b) myswap((a),(b),sizeof(a))
1578
1579#define myswap(a, b, size) do {         \
1580        unsigned char mytmp[size];      \
1581        memcpy(mytmp, &a, size);                \
1582        memcpy(&a, &b, size);           \
1583        memcpy(&b, mytmp, size);                \
1584} while (0)
1585
1586static void reverse_patches(struct patch *p)
1587{
1588        for (; p; p = p->next) {
1589                struct fragment *frag = p->fragments;
1590
1591                swap(p->new_name, p->old_name);
1592                swap(p->new_mode, p->old_mode);
1593                swap(p->is_new, p->is_delete);
1594                swap(p->lines_added, p->lines_deleted);
1595                swap(p->old_sha1_prefix, p->new_sha1_prefix);
1596
1597                for (; frag; frag = frag->next) {
1598                        swap(frag->newpos, frag->oldpos);
1599                        swap(frag->newlines, frag->oldlines);
1600                }
1601        }
1602}
1603
1604static const char pluses[] =
1605"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1606static const char minuses[]=
1607"----------------------------------------------------------------------";
1608
1609static void show_stats(struct patch *patch)
1610{
1611        struct strbuf qname = STRBUF_INIT;
1612        char *cp = patch->new_name ? patch->new_name : patch->old_name;
1613        int max, add, del;
1614
1615        quote_c_style(cp, &qname, NULL, 0);
1616
1617        /*
1618         * "scale" the filename
1619         */
1620        max = max_len;
1621        if (max > 50)
1622                max = 50;
1623
1624        if (qname.len > max) {
1625                cp = strchr(qname.buf + qname.len + 3 - max, '/');
1626                if (!cp)
1627                        cp = qname.buf + qname.len + 3 - max;
1628                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1629        }
1630
1631        if (patch->is_binary) {
1632                printf(" %-*s |  Bin\n", max, qname.buf);
1633                strbuf_release(&qname);
1634                return;
1635        }
1636
1637        printf(" %-*s |", max, qname.buf);
1638        strbuf_release(&qname);
1639
1640        /*
1641         * scale the add/delete
1642         */
1643        max = max + max_change > 70 ? 70 - max : max_change;
1644        add = patch->lines_added;
1645        del = patch->lines_deleted;
1646
1647        if (max_change > 0) {
1648                int total = ((add + del) * max + max_change / 2) / max_change;
1649                add = (add * max + max_change / 2) / max_change;
1650                del = total - add;
1651        }
1652        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1653                add, pluses, del, minuses);
1654}
1655
1656static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1657{
1658        switch (st->st_mode & S_IFMT) {
1659        case S_IFLNK:
1660                if (strbuf_readlink(buf, path, st->st_size) < 0)
1661                        return error("unable to read symlink %s", path);
1662                return 0;
1663        case S_IFREG:
1664                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1665                        return error("unable to open or read %s", path);
1666                convert_to_git(path, buf->buf, buf->len, buf, 0);
1667                return 0;
1668        default:
1669                return -1;
1670        }
1671}
1672
1673/*
1674 * Update the preimage, and the common lines in postimage,
1675 * from buffer buf of length len. If postlen is 0 the postimage
1676 * is updated in place, otherwise it's updated on a new buffer
1677 * of length postlen
1678 */
1679
1680static void update_pre_post_images(struct image *preimage,
1681                                   struct image *postimage,
1682                                   char *buf,
1683                                   size_t len, size_t postlen)
1684{
1685        int i, ctx;
1686        char *new, *old, *fixed;
1687        struct image fixed_preimage;
1688
1689        /*
1690         * Update the preimage with whitespace fixes.  Note that we
1691         * are not losing preimage->buf -- apply_one_fragment() will
1692         * free "oldlines".
1693         */
1694        prepare_image(&fixed_preimage, buf, len, 1);
1695        assert(fixed_preimage.nr == preimage->nr);
1696        for (i = 0; i < preimage->nr; i++)
1697                fixed_preimage.line[i].flag = preimage->line[i].flag;
1698        free(preimage->line_allocated);
1699        *preimage = fixed_preimage;
1700
1701        /*
1702         * Adjust the common context lines in postimage. This can be
1703         * done in-place when we are just doing whitespace fixing,
1704         * which does not make the string grow, but needs a new buffer
1705         * when ignoring whitespace causes the update, since in this case
1706         * we could have e.g. tabs converted to multiple spaces.
1707         * We trust the caller to tell us if the update can be done
1708         * in place (postlen==0) or not.
1709         */
1710        old = postimage->buf;
1711        if (postlen)
1712                new = postimage->buf = xmalloc(postlen);
1713        else
1714                new = old;
1715        fixed = preimage->buf;
1716        for (i = ctx = 0; i < postimage->nr; i++) {
1717                size_t len = postimage->line[i].len;
1718                if (!(postimage->line[i].flag & LINE_COMMON)) {
1719                        /* an added line -- no counterparts in preimage */
1720                        memmove(new, old, len);
1721                        old += len;
1722                        new += len;
1723                        continue;
1724                }
1725
1726                /* a common context -- skip it in the original postimage */
1727                old += len;
1728
1729                /* and find the corresponding one in the fixed preimage */
1730                while (ctx < preimage->nr &&
1731                       !(preimage->line[ctx].flag & LINE_COMMON)) {
1732                        fixed += preimage->line[ctx].len;
1733                        ctx++;
1734                }
1735                if (preimage->nr <= ctx)
1736                        die("oops");
1737
1738                /* and copy it in, while fixing the line length */
1739                len = preimage->line[ctx].len;
1740                memcpy(new, fixed, len);
1741                new += len;
1742                fixed += len;
1743                postimage->line[i].len = len;
1744                ctx++;
1745        }
1746
1747        /* Fix the length of the whole thing */
1748        postimage->len = new - postimage->buf;
1749}
1750
1751static int match_fragment(struct image *img,
1752                          struct image *preimage,
1753                          struct image *postimage,
1754                          unsigned long try,
1755                          int try_lno,
1756                          unsigned ws_rule,
1757                          int match_beginning, int match_end)
1758{
1759        int i;
1760        char *fixed_buf, *buf, *orig, *target;
1761
1762        if (preimage->nr + try_lno > img->nr)
1763                return 0;
1764
1765        if (match_beginning && try_lno)
1766                return 0;
1767
1768        if (match_end && preimage->nr + try_lno != img->nr)
1769                return 0;
1770
1771        /* Quick hash check */
1772        for (i = 0; i < preimage->nr; i++)
1773                if (preimage->line[i].hash != img->line[try_lno + i].hash)
1774                        return 0;
1775
1776        /*
1777         * Do we have an exact match?  If we were told to match
1778         * at the end, size must be exactly at try+fragsize,
1779         * otherwise try+fragsize must be still within the preimage,
1780         * and either case, the old piece should match the preimage
1781         * exactly.
1782         */
1783        if ((match_end
1784             ? (try + preimage->len == img->len)
1785             : (try + preimage->len <= img->len)) &&
1786            !memcmp(img->buf + try, preimage->buf, preimage->len))
1787                return 1;
1788
1789        /*
1790         * No exact match. If we are ignoring whitespace, run a line-by-line
1791         * fuzzy matching. We collect all the line length information because
1792         * we need it to adjust whitespace if we match.
1793         */
1794        if (ws_ignore_action == ignore_ws_change) {
1795                size_t imgoff = 0;
1796                size_t preoff = 0;
1797                size_t postlen = postimage->len;
1798                size_t imglen[preimage->nr];
1799                for (i = 0; i < preimage->nr; i++) {
1800                        size_t prelen = preimage->line[i].len;
1801
1802                        imglen[i] = img->line[try_lno+i].len;
1803                        if (!fuzzy_matchlines(
1804                                img->buf + try + imgoff, imglen[i],
1805                                preimage->buf + preoff, prelen))
1806                                return 0;
1807                        if (preimage->line[i].flag & LINE_COMMON)
1808                                postlen += imglen[i] - prelen;
1809                        imgoff += imglen[i];
1810                        preoff += prelen;
1811                }
1812
1813                /*
1814                 * Ok, the preimage matches with whitespace fuzz. Update it and
1815                 * the common postimage lines to use the same whitespace as the
1816                 * target. imgoff now holds the true length of the target that
1817                 * matches the preimage, and we need to update the line lengths
1818                 * of the preimage to match the target ones.
1819                 */
1820                fixed_buf = xmalloc(imgoff);
1821                memcpy(fixed_buf, img->buf + try, imgoff);
1822                for (i = 0; i < preimage->nr; i++)
1823                        preimage->line[i].len = imglen[i];
1824
1825                /*
1826                 * Update the preimage buffer and the postimage context lines.
1827                 */
1828                update_pre_post_images(preimage, postimage,
1829                                fixed_buf, imgoff, postlen);
1830                return 1;
1831        }
1832
1833        if (ws_error_action != correct_ws_error)
1834                return 0;
1835
1836        /*
1837         * The hunk does not apply byte-by-byte, but the hash says
1838         * it might with whitespace fuzz. We haven't been asked to
1839         * ignore whitespace, we were asked to correct whitespace
1840         * errors, so let's try matching after whitespace correction.
1841         */
1842        fixed_buf = xmalloc(preimage->len + 1);
1843        buf = fixed_buf;
1844        orig = preimage->buf;
1845        target = img->buf + try;
1846        for (i = 0; i < preimage->nr; i++) {
1847                size_t fixlen; /* length after fixing the preimage */
1848                size_t oldlen = preimage->line[i].len;
1849                size_t tgtlen = img->line[try_lno + i].len;
1850                size_t tgtfixlen; /* length after fixing the target line */
1851                char tgtfixbuf[1024], *tgtfix;
1852                int match;
1853
1854                /* Try fixing the line in the preimage */
1855                fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1856
1857                /* Try fixing the line in the target */
1858                if (sizeof(tgtfixbuf) > tgtlen)
1859                        tgtfix = tgtfixbuf;
1860                else
1861                        tgtfix = xmalloc(tgtlen);
1862                tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1863
1864                /*
1865                 * If they match, either the preimage was based on
1866                 * a version before our tree fixed whitespace breakage,
1867                 * or we are lacking a whitespace-fix patch the tree
1868                 * the preimage was based on already had (i.e. target
1869                 * has whitespace breakage, the preimage doesn't).
1870                 * In either case, we are fixing the whitespace breakages
1871                 * so we might as well take the fix together with their
1872                 * real change.
1873                 */
1874                match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1875
1876                if (tgtfix != tgtfixbuf)
1877                        free(tgtfix);
1878                if (!match)
1879                        goto unmatch_exit;
1880
1881                orig += oldlen;
1882                buf += fixlen;
1883                target += tgtlen;
1884        }
1885
1886        /*
1887         * Yes, the preimage is based on an older version that still
1888         * has whitespace breakages unfixed, and fixing them makes the
1889         * hunk match.  Update the context lines in the postimage.
1890         */
1891        update_pre_post_images(preimage, postimage,
1892                               fixed_buf, buf - fixed_buf, 0);
1893        return 1;
1894
1895 unmatch_exit:
1896        free(fixed_buf);
1897        return 0;
1898}
1899
1900static int find_pos(struct image *img,
1901                    struct image *preimage,
1902                    struct image *postimage,
1903                    int line,
1904                    unsigned ws_rule,
1905                    int match_beginning, int match_end)
1906{
1907        int i;
1908        unsigned long backwards, forwards, try;
1909        int backwards_lno, forwards_lno, try_lno;
1910
1911        if (preimage->nr > img->nr)
1912                return -1;
1913
1914        /*
1915         * If match_begining or match_end is specified, there is no
1916         * point starting from a wrong line that will never match and
1917         * wander around and wait for a match at the specified end.
1918         */
1919        if (match_beginning)
1920                line = 0;
1921        else if (match_end)
1922                line = img->nr - preimage->nr;
1923
1924        if (line > img->nr)
1925                line = img->nr;
1926
1927        try = 0;
1928        for (i = 0; i < line; i++)
1929                try += img->line[i].len;
1930
1931        /*
1932         * There's probably some smart way to do this, but I'll leave
1933         * that to the smart and beautiful people. I'm simple and stupid.
1934         */
1935        backwards = try;
1936        backwards_lno = line;
1937        forwards = try;
1938        forwards_lno = line;
1939        try_lno = line;
1940
1941        for (i = 0; ; i++) {
1942                if (match_fragment(img, preimage, postimage,
1943                                   try, try_lno, ws_rule,
1944                                   match_beginning, match_end))
1945                        return try_lno;
1946
1947        again:
1948                if (backwards_lno == 0 && forwards_lno == img->nr)
1949                        break;
1950
1951                if (i & 1) {
1952                        if (backwards_lno == 0) {
1953                                i++;
1954                                goto again;
1955                        }
1956                        backwards_lno--;
1957                        backwards -= img->line[backwards_lno].len;
1958                        try = backwards;
1959                        try_lno = backwards_lno;
1960                } else {
1961                        if (forwards_lno == img->nr) {
1962                                i++;
1963                                goto again;
1964                        }
1965                        forwards += img->line[forwards_lno].len;
1966                        forwards_lno++;
1967                        try = forwards;
1968                        try_lno = forwards_lno;
1969                }
1970
1971        }
1972        return -1;
1973}
1974
1975static void remove_first_line(struct image *img)
1976{
1977        img->buf += img->line[0].len;
1978        img->len -= img->line[0].len;
1979        img->line++;
1980        img->nr--;
1981}
1982
1983static void remove_last_line(struct image *img)
1984{
1985        img->len -= img->line[--img->nr].len;
1986}
1987
1988static void update_image(struct image *img,
1989                         int applied_pos,
1990                         struct image *preimage,
1991                         struct image *postimage)
1992{
1993        /*
1994         * remove the copy of preimage at offset in img
1995         * and replace it with postimage
1996         */
1997        int i, nr;
1998        size_t remove_count, insert_count, applied_at = 0;
1999        char *result;
2000
2001        for (i = 0; i < applied_pos; i++)
2002                applied_at += img->line[i].len;
2003
2004        remove_count = 0;
2005        for (i = 0; i < preimage->nr; i++)
2006                remove_count += img->line[applied_pos + i].len;
2007        insert_count = postimage->len;
2008
2009        /* Adjust the contents */
2010        result = xmalloc(img->len + insert_count - remove_count + 1);
2011        memcpy(result, img->buf, applied_at);
2012        memcpy(result + applied_at, postimage->buf, postimage->len);
2013        memcpy(result + applied_at + postimage->len,
2014               img->buf + (applied_at + remove_count),
2015               img->len - (applied_at + remove_count));
2016        free(img->buf);
2017        img->buf = result;
2018        img->len += insert_count - remove_count;
2019        result[img->len] = '\0';
2020
2021        /* Adjust the line table */
2022        nr = img->nr + postimage->nr - preimage->nr;
2023        if (preimage->nr < postimage->nr) {
2024                /*
2025                 * NOTE: this knows that we never call remove_first_line()
2026                 * on anything other than pre/post image.
2027                 */
2028                img->line = xrealloc(img->line, nr * sizeof(*img->line));
2029                img->line_allocated = img->line;
2030        }
2031        if (preimage->nr != postimage->nr)
2032                memmove(img->line + applied_pos + postimage->nr,
2033                        img->line + applied_pos + preimage->nr,
2034                        (img->nr - (applied_pos + preimage->nr)) *
2035                        sizeof(*img->line));
2036        memcpy(img->line + applied_pos,
2037               postimage->line,
2038               postimage->nr * sizeof(*img->line));
2039        img->nr = nr;
2040}
2041
2042static int apply_one_fragment(struct image *img, struct fragment *frag,
2043                              int inaccurate_eof, unsigned ws_rule)
2044{
2045        int match_beginning, match_end;
2046        const char *patch = frag->patch;
2047        int size = frag->size;
2048        char *old, *new, *oldlines, *newlines;
2049        int new_blank_lines_at_end = 0;
2050        unsigned long leading, trailing;
2051        int pos, applied_pos;
2052        struct image preimage;
2053        struct image postimage;
2054
2055        memset(&preimage, 0, sizeof(preimage));
2056        memset(&postimage, 0, sizeof(postimage));
2057        oldlines = xmalloc(size);
2058        newlines = xmalloc(size);
2059
2060        old = oldlines;
2061        new = newlines;
2062        while (size > 0) {
2063                char first;
2064                int len = linelen(patch, size);
2065                int plen, added;
2066                int added_blank_line = 0;
2067
2068                if (!len)
2069                        break;
2070
2071                /*
2072                 * "plen" is how much of the line we should use for
2073                 * the actual patch data. Normally we just remove the
2074                 * first character on the line, but if the line is
2075                 * followed by "\ No newline", then we also remove the
2076                 * last one (which is the newline, of course).
2077                 */
2078                plen = len - 1;
2079                if (len < size && patch[len] == '\\')
2080                        plen--;
2081                first = *patch;
2082                if (apply_in_reverse) {
2083                        if (first == '-')
2084                                first = '+';
2085                        else if (first == '+')
2086                                first = '-';
2087                }
2088
2089                switch (first) {
2090                case '\n':
2091                        /* Newer GNU diff, empty context line */
2092                        if (plen < 0)
2093                                /* ... followed by '\No newline'; nothing */
2094                                break;
2095                        *old++ = '\n';
2096                        *new++ = '\n';
2097                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2098                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2099                        break;
2100                case ' ':
2101                case '-':
2102                        memcpy(old, patch + 1, plen);
2103                        add_line_info(&preimage, old, plen,
2104                                      (first == ' ' ? LINE_COMMON : 0));
2105                        old += plen;
2106                        if (first == '-')
2107                                break;
2108                /* Fall-through for ' ' */
2109                case '+':
2110                        /* --no-add does not add new lines */
2111                        if (first == '+' && no_add)
2112                                break;
2113
2114                        if (first != '+' ||
2115                            !whitespace_error ||
2116                            ws_error_action != correct_ws_error) {
2117                                memcpy(new, patch + 1, plen);
2118                                added = plen;
2119                        }
2120                        else {
2121                                added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2122                        }
2123                        add_line_info(&postimage, new, added,
2124                                      (first == '+' ? 0 : LINE_COMMON));
2125                        new += added;
2126                        if (first == '+' &&
2127                            added == 1 && new[-1] == '\n')
2128                                added_blank_line = 1;
2129                        break;
2130                case '@': case '\\':
2131                        /* Ignore it, we already handled it */
2132                        break;
2133                default:
2134                        if (apply_verbosely)
2135                                error("invalid start of line: '%c'", first);
2136                        return -1;
2137                }
2138                if (added_blank_line)
2139                        new_blank_lines_at_end++;
2140                else
2141                        new_blank_lines_at_end = 0;
2142                patch += len;
2143                size -= len;
2144        }
2145        if (inaccurate_eof &&
2146            old > oldlines && old[-1] == '\n' &&
2147            new > newlines && new[-1] == '\n') {
2148                old--;
2149                new--;
2150        }
2151
2152        leading = frag->leading;
2153        trailing = frag->trailing;
2154
2155        /*
2156         * A hunk to change lines at the beginning would begin with
2157         * @@ -1,L +N,M @@
2158         * but we need to be careful.  -U0 that inserts before the second
2159         * line also has this pattern.
2160         *
2161         * And a hunk to add to an empty file would begin with
2162         * @@ -0,0 +N,M @@
2163         *
2164         * In other words, a hunk that is (frag->oldpos <= 1) with or
2165         * without leading context must match at the beginning.
2166         */
2167        match_beginning = (!frag->oldpos ||
2168                           (frag->oldpos == 1 && !unidiff_zero));
2169
2170        /*
2171         * A hunk without trailing lines must match at the end.
2172         * However, we simply cannot tell if a hunk must match end
2173         * from the lack of trailing lines if the patch was generated
2174         * with unidiff without any context.
2175         */
2176        match_end = !unidiff_zero && !trailing;
2177
2178        pos = frag->newpos ? (frag->newpos - 1) : 0;
2179        preimage.buf = oldlines;
2180        preimage.len = old - oldlines;
2181        postimage.buf = newlines;
2182        postimage.len = new - newlines;
2183        preimage.line = preimage.line_allocated;
2184        postimage.line = postimage.line_allocated;
2185
2186        for (;;) {
2187
2188                applied_pos = find_pos(img, &preimage, &postimage, pos,
2189                                       ws_rule, match_beginning, match_end);
2190
2191                if (applied_pos >= 0)
2192                        break;
2193
2194                /* Am I at my context limits? */
2195                if ((leading <= p_context) && (trailing <= p_context))
2196                        break;
2197                if (match_beginning || match_end) {
2198                        match_beginning = match_end = 0;
2199                        continue;
2200                }
2201
2202                /*
2203                 * Reduce the number of context lines; reduce both
2204                 * leading and trailing if they are equal otherwise
2205                 * just reduce the larger context.
2206                 */
2207                if (leading >= trailing) {
2208                        remove_first_line(&preimage);
2209                        remove_first_line(&postimage);
2210                        pos--;
2211                        leading--;
2212                }
2213                if (trailing > leading) {
2214                        remove_last_line(&preimage);
2215                        remove_last_line(&postimage);
2216                        trailing--;
2217                }
2218        }
2219
2220        if (applied_pos >= 0) {
2221                if (ws_error_action == correct_ws_error &&
2222                    new_blank_lines_at_end &&
2223                    postimage.nr + applied_pos == img->nr) {
2224                        /*
2225                         * If the patch application adds blank lines
2226                         * at the end, and if the patch applies at the
2227                         * end of the image, remove those added blank
2228                         * lines.
2229                         */
2230                        while (new_blank_lines_at_end--)
2231                                remove_last_line(&postimage);
2232                }
2233
2234                /*
2235                 * Warn if it was necessary to reduce the number
2236                 * of context lines.
2237                 */
2238                if ((leading != frag->leading) ||
2239                    (trailing != frag->trailing))
2240                        fprintf(stderr, "Context reduced to (%ld/%ld)"
2241                                " to apply fragment at %d\n",
2242                                leading, trailing, applied_pos+1);
2243                update_image(img, applied_pos, &preimage, &postimage);
2244        } else {
2245                if (apply_verbosely)
2246                        error("while searching for:\n%.*s",
2247                              (int)(old - oldlines), oldlines);
2248        }
2249
2250        free(oldlines);
2251        free(newlines);
2252        free(preimage.line_allocated);
2253        free(postimage.line_allocated);
2254
2255        return (applied_pos < 0);
2256}
2257
2258static int apply_binary_fragment(struct image *img, struct patch *patch)
2259{
2260        struct fragment *fragment = patch->fragments;
2261        unsigned long len;
2262        void *dst;
2263
2264        /* Binary patch is irreversible without the optional second hunk */
2265        if (apply_in_reverse) {
2266                if (!fragment->next)
2267                        return error("cannot reverse-apply a binary patch "
2268                                     "without the reverse hunk to '%s'",
2269                                     patch->new_name
2270                                     ? patch->new_name : patch->old_name);
2271                fragment = fragment->next;
2272        }
2273        switch (fragment->binary_patch_method) {
2274        case BINARY_DELTA_DEFLATED:
2275                dst = patch_delta(img->buf, img->len, fragment->patch,
2276                                  fragment->size, &len);
2277                if (!dst)
2278                        return -1;
2279                clear_image(img);
2280                img->buf = dst;
2281                img->len = len;
2282                return 0;
2283        case BINARY_LITERAL_DEFLATED:
2284                clear_image(img);
2285                img->len = fragment->size;
2286                img->buf = xmalloc(img->len+1);
2287                memcpy(img->buf, fragment->patch, img->len);
2288                img->buf[img->len] = '\0';
2289                return 0;
2290        }
2291        return -1;
2292}
2293
2294static int apply_binary(struct image *img, struct patch *patch)
2295{
2296        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2297        unsigned char sha1[20];
2298
2299        /*
2300         * For safety, we require patch index line to contain
2301         * full 40-byte textual SHA1 for old and new, at least for now.
2302         */
2303        if (strlen(patch->old_sha1_prefix) != 40 ||
2304            strlen(patch->new_sha1_prefix) != 40 ||
2305            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2306            get_sha1_hex(patch->new_sha1_prefix, sha1))
2307                return error("cannot apply binary patch to '%s' "
2308                             "without full index line", name);
2309
2310        if (patch->old_name) {
2311                /*
2312                 * See if the old one matches what the patch
2313                 * applies to.
2314                 */
2315                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2316                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2317                        return error("the patch applies to '%s' (%s), "
2318                                     "which does not match the "
2319                                     "current contents.",
2320                                     name, sha1_to_hex(sha1));
2321        }
2322        else {
2323                /* Otherwise, the old one must be empty. */
2324                if (img->len)
2325                        return error("the patch applies to an empty "
2326                                     "'%s' but it is not empty", name);
2327        }
2328
2329        get_sha1_hex(patch->new_sha1_prefix, sha1);
2330        if (is_null_sha1(sha1)) {
2331                clear_image(img);
2332                return 0; /* deletion patch */
2333        }
2334
2335        if (has_sha1_file(sha1)) {
2336                /* We already have the postimage */
2337                enum object_type type;
2338                unsigned long size;
2339                char *result;
2340
2341                result = read_sha1_file(sha1, &type, &size);
2342                if (!result)
2343                        return error("the necessary postimage %s for "
2344                                     "'%s' cannot be read",
2345                                     patch->new_sha1_prefix, name);
2346                clear_image(img);
2347                img->buf = result;
2348                img->len = size;
2349        } else {
2350                /*
2351                 * We have verified buf matches the preimage;
2352                 * apply the patch data to it, which is stored
2353                 * in the patch->fragments->{patch,size}.
2354                 */
2355                if (apply_binary_fragment(img, patch))
2356                        return error("binary patch does not apply to '%s'",
2357                                     name);
2358
2359                /* verify that the result matches */
2360                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2361                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2362                        return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2363                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2364        }
2365
2366        return 0;
2367}
2368
2369static int apply_fragments(struct image *img, struct patch *patch)
2370{
2371        struct fragment *frag = patch->fragments;
2372        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2373        unsigned ws_rule = patch->ws_rule;
2374        unsigned inaccurate_eof = patch->inaccurate_eof;
2375
2376        if (patch->is_binary)
2377                return apply_binary(img, patch);
2378
2379        while (frag) {
2380                if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2381                        error("patch failed: %s:%ld", name, frag->oldpos);
2382                        if (!apply_with_reject)
2383                                return -1;
2384                        frag->rejected = 1;
2385                }
2386                frag = frag->next;
2387        }
2388        return 0;
2389}
2390
2391static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2392{
2393        if (!ce)
2394                return 0;
2395
2396        if (S_ISGITLINK(ce->ce_mode)) {
2397                strbuf_grow(buf, 100);
2398                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2399        } else {
2400                enum object_type type;
2401                unsigned long sz;
2402                char *result;
2403
2404                result = read_sha1_file(ce->sha1, &type, &sz);
2405                if (!result)
2406                        return -1;
2407                /* XXX read_sha1_file NUL-terminates */
2408                strbuf_attach(buf, result, sz, sz + 1);
2409        }
2410        return 0;
2411}
2412
2413static struct patch *in_fn_table(const char *name)
2414{
2415        struct string_list_item *item;
2416
2417        if (name == NULL)
2418                return NULL;
2419
2420        item = string_list_lookup(name, &fn_table);
2421        if (item != NULL)
2422                return (struct patch *)item->util;
2423
2424        return NULL;
2425}
2426
2427/*
2428 * item->util in the filename table records the status of the path.
2429 * Usually it points at a patch (whose result records the contents
2430 * of it after applying it), but it could be PATH_WAS_DELETED for a
2431 * path that a previously applied patch has already removed.
2432 */
2433 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2434#define PATH_WAS_DELETED ((struct patch *) -1)
2435
2436static int to_be_deleted(struct patch *patch)
2437{
2438        return patch == PATH_TO_BE_DELETED;
2439}
2440
2441static int was_deleted(struct patch *patch)
2442{
2443        return patch == PATH_WAS_DELETED;
2444}
2445
2446static void add_to_fn_table(struct patch *patch)
2447{
2448        struct string_list_item *item;
2449
2450        /*
2451         * Always add new_name unless patch is a deletion
2452         * This should cover the cases for normal diffs,
2453         * file creations and copies
2454         */
2455        if (patch->new_name != NULL) {
2456                item = string_list_insert(patch->new_name, &fn_table);
2457                item->util = patch;
2458        }
2459
2460        /*
2461         * store a failure on rename/deletion cases because
2462         * later chunks shouldn't patch old names
2463         */
2464        if ((patch->new_name == NULL) || (patch->is_rename)) {
2465                item = string_list_insert(patch->old_name, &fn_table);
2466                item->util = PATH_WAS_DELETED;
2467        }
2468}
2469
2470static void prepare_fn_table(struct patch *patch)
2471{
2472        /*
2473         * store information about incoming file deletion
2474         */
2475        while (patch) {
2476                if ((patch->new_name == NULL) || (patch->is_rename)) {
2477                        struct string_list_item *item;
2478                        item = string_list_insert(patch->old_name, &fn_table);
2479                        item->util = PATH_TO_BE_DELETED;
2480                }
2481                patch = patch->next;
2482        }
2483}
2484
2485static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2486{
2487        struct strbuf buf = STRBUF_INIT;
2488        struct image image;
2489        size_t len;
2490        char *img;
2491        struct patch *tpatch;
2492
2493        if (!(patch->is_copy || patch->is_rename) &&
2494            (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2495                if (was_deleted(tpatch)) {
2496                        return error("patch %s has been renamed/deleted",
2497                                patch->old_name);
2498                }
2499                /* We have a patched copy in memory use that */
2500                strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2501        } else if (cached) {
2502                if (read_file_or_gitlink(ce, &buf))
2503                        return error("read of %s failed", patch->old_name);
2504        } else if (patch->old_name) {
2505                if (S_ISGITLINK(patch->old_mode)) {
2506                        if (ce) {
2507                                read_file_or_gitlink(ce, &buf);
2508                        } else {
2509                                /*
2510                                 * There is no way to apply subproject
2511                                 * patch without looking at the index.
2512                                 */
2513                                patch->fragments = NULL;
2514                        }
2515                } else {
2516                        if (read_old_data(st, patch->old_name, &buf))
2517                                return error("read of %s failed", patch->old_name);
2518                }
2519        }
2520
2521        img = strbuf_detach(&buf, &len);
2522        prepare_image(&image, img, len, !patch->is_binary);
2523
2524        if (apply_fragments(&image, patch) < 0)
2525                return -1; /* note with --reject this succeeds. */
2526        patch->result = image.buf;
2527        patch->resultsize = image.len;
2528        add_to_fn_table(patch);
2529        free(image.line_allocated);
2530
2531        if (0 < patch->is_delete && patch->resultsize)
2532                return error("removal patch leaves file contents");
2533
2534        return 0;
2535}
2536
2537static int check_to_create_blob(const char *new_name, int ok_if_exists)
2538{
2539        struct stat nst;
2540        if (!lstat(new_name, &nst)) {
2541                if (S_ISDIR(nst.st_mode) || ok_if_exists)
2542                        return 0;
2543                /*
2544                 * A leading component of new_name might be a symlink
2545                 * that is going to be removed with this patch, but
2546                 * still pointing at somewhere that has the path.
2547                 * In such a case, path "new_name" does not exist as
2548                 * far as git is concerned.
2549                 */
2550                if (has_symlink_leading_path(new_name, strlen(new_name)))
2551                        return 0;
2552
2553                return error("%s: already exists in working directory", new_name);
2554        }
2555        else if ((errno != ENOENT) && (errno != ENOTDIR))
2556                return error("%s: %s", new_name, strerror(errno));
2557        return 0;
2558}
2559
2560static int verify_index_match(struct cache_entry *ce, struct stat *st)
2561{
2562        if (S_ISGITLINK(ce->ce_mode)) {
2563                if (!S_ISDIR(st->st_mode))
2564                        return -1;
2565                return 0;
2566        }
2567        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2568}
2569
2570static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2571{
2572        const char *old_name = patch->old_name;
2573        struct patch *tpatch = NULL;
2574        int stat_ret = 0;
2575        unsigned st_mode = 0;
2576
2577        /*
2578         * Make sure that we do not have local modifications from the
2579         * index when we are looking at the index.  Also make sure
2580         * we have the preimage file to be patched in the work tree,
2581         * unless --cached, which tells git to apply only in the index.
2582         */
2583        if (!old_name)
2584                return 0;
2585
2586        assert(patch->is_new <= 0);
2587
2588        if (!(patch->is_copy || patch->is_rename) &&
2589            (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2590                if (was_deleted(tpatch))
2591                        return error("%s: has been deleted/renamed", old_name);
2592                st_mode = tpatch->new_mode;
2593        } else if (!cached) {
2594                stat_ret = lstat(old_name, st);
2595                if (stat_ret && errno != ENOENT)
2596                        return error("%s: %s", old_name, strerror(errno));
2597        }
2598
2599        if (to_be_deleted(tpatch))
2600                tpatch = NULL;
2601
2602        if (check_index && !tpatch) {
2603                int pos = cache_name_pos(old_name, strlen(old_name));
2604                if (pos < 0) {
2605                        if (patch->is_new < 0)
2606                                goto is_new;
2607                        return error("%s: does not exist in index", old_name);
2608                }
2609                *ce = active_cache[pos];
2610                if (stat_ret < 0) {
2611                        struct checkout costate;
2612                        /* checkout */
2613                        costate.base_dir = "";
2614                        costate.base_dir_len = 0;
2615                        costate.force = 0;
2616                        costate.quiet = 0;
2617                        costate.not_new = 0;
2618                        costate.refresh_cache = 1;
2619                        if (checkout_entry(*ce, &costate, NULL) ||
2620                            lstat(old_name, st))
2621                                return -1;
2622                }
2623                if (!cached && verify_index_match(*ce, st))
2624                        return error("%s: does not match index", old_name);
2625                if (cached)
2626                        st_mode = (*ce)->ce_mode;
2627        } else if (stat_ret < 0) {
2628                if (patch->is_new < 0)
2629                        goto is_new;
2630                return error("%s: %s", old_name, strerror(errno));
2631        }
2632
2633        if (!cached && !tpatch)
2634                st_mode = ce_mode_from_stat(*ce, st->st_mode);
2635
2636        if (patch->is_new < 0)
2637                patch->is_new = 0;
2638        if (!patch->old_mode)
2639                patch->old_mode = st_mode;
2640        if ((st_mode ^ patch->old_mode) & S_IFMT)
2641                return error("%s: wrong type", old_name);
2642        if (st_mode != patch->old_mode)
2643                warning("%s has type %o, expected %o",
2644                        old_name, st_mode, patch->old_mode);
2645        if (!patch->new_mode && !patch->is_delete)
2646                patch->new_mode = st_mode;
2647        return 0;
2648
2649 is_new:
2650        patch->is_new = 1;
2651        patch->is_delete = 0;
2652        patch->old_name = NULL;
2653        return 0;
2654}
2655
2656static int check_patch(struct patch *patch)
2657{
2658        struct stat st;
2659        const char *old_name = patch->old_name;
2660        const char *new_name = patch->new_name;
2661        const char *name = old_name ? old_name : new_name;
2662        struct cache_entry *ce = NULL;
2663        struct patch *tpatch;
2664        int ok_if_exists;
2665        int status;
2666
2667        patch->rejected = 1; /* we will drop this after we succeed */
2668
2669        status = check_preimage(patch, &ce, &st);
2670        if (status)
2671                return status;
2672        old_name = patch->old_name;
2673
2674        if ((tpatch = in_fn_table(new_name)) &&
2675                        (was_deleted(tpatch) || to_be_deleted(tpatch)))
2676                /*
2677                 * A type-change diff is always split into a patch to
2678                 * delete old, immediately followed by a patch to
2679                 * create new (see diff.c::run_diff()); in such a case
2680                 * it is Ok that the entry to be deleted by the
2681                 * previous patch is still in the working tree and in
2682                 * the index.
2683                 */
2684                ok_if_exists = 1;
2685        else
2686                ok_if_exists = 0;
2687
2688        if (new_name &&
2689            ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2690                if (check_index &&
2691                    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2692                    !ok_if_exists)
2693                        return error("%s: already exists in index", new_name);
2694                if (!cached) {
2695                        int err = check_to_create_blob(new_name, ok_if_exists);
2696                        if (err)
2697                                return err;
2698                }
2699                if (!patch->new_mode) {
2700                        if (0 < patch->is_new)
2701                                patch->new_mode = S_IFREG | 0644;
2702                        else
2703                                patch->new_mode = patch->old_mode;
2704                }
2705        }
2706
2707        if (new_name && old_name) {
2708                int same = !strcmp(old_name, new_name);
2709                if (!patch->new_mode)
2710                        patch->new_mode = patch->old_mode;
2711                if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2712                        return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2713                                patch->new_mode, new_name, patch->old_mode,
2714                                same ? "" : " of ", same ? "" : old_name);
2715        }
2716
2717        if (apply_data(patch, &st, ce) < 0)
2718                return error("%s: patch does not apply", name);
2719        patch->rejected = 0;
2720        return 0;
2721}
2722
2723static int check_patch_list(struct patch *patch)
2724{
2725        int err = 0;
2726
2727        prepare_fn_table(patch);
2728        while (patch) {
2729                if (apply_verbosely)
2730                        say_patch_name(stderr,
2731                                       "Checking patch ", patch, "...\n");
2732                err |= check_patch(patch);
2733                patch = patch->next;
2734        }
2735        return err;
2736}
2737
2738/* This function tries to read the sha1 from the current index */
2739static int get_current_sha1(const char *path, unsigned char *sha1)
2740{
2741        int pos;
2742
2743        if (read_cache() < 0)
2744                return -1;
2745        pos = cache_name_pos(path, strlen(path));
2746        if (pos < 0)
2747                return -1;
2748        hashcpy(sha1, active_cache[pos]->sha1);
2749        return 0;
2750}
2751
2752/* Build an index that contains the just the files needed for a 3way merge */
2753static void build_fake_ancestor(struct patch *list, const char *filename)
2754{
2755        struct patch *patch;
2756        struct index_state result = { NULL };
2757        int fd;
2758
2759        /* Once we start supporting the reverse patch, it may be
2760         * worth showing the new sha1 prefix, but until then...
2761         */
2762        for (patch = list; patch; patch = patch->next) {
2763                const unsigned char *sha1_ptr;
2764                unsigned char sha1[20];
2765                struct cache_entry *ce;
2766                const char *name;
2767
2768                name = patch->old_name ? patch->old_name : patch->new_name;
2769                if (0 < patch->is_new)
2770                        continue;
2771                else if (get_sha1(patch->old_sha1_prefix, sha1))
2772                        /* git diff has no index line for mode/type changes */
2773                        if (!patch->lines_added && !patch->lines_deleted) {
2774                                if (get_current_sha1(patch->new_name, sha1) ||
2775                                    get_current_sha1(patch->old_name, sha1))
2776                                        die("mode change for %s, which is not "
2777                                                "in current HEAD", name);
2778                                sha1_ptr = sha1;
2779                        } else
2780                                die("sha1 information is lacking or useless "
2781                                        "(%s).", name);
2782                else
2783                        sha1_ptr = sha1;
2784
2785                ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2786                if (!ce)
2787                        die("make_cache_entry failed for path '%s'", name);
2788                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2789                        die ("Could not add %s to temporary index", name);
2790        }
2791
2792        fd = open(filename, O_WRONLY | O_CREAT, 0666);
2793        if (fd < 0 || write_index(&result, fd) || close(fd))
2794                die ("Could not write temporary index to %s", filename);
2795
2796        discard_index(&result);
2797}
2798
2799static void stat_patch_list(struct patch *patch)
2800{
2801        int files, adds, dels;
2802
2803        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2804                files++;
2805                adds += patch->lines_added;
2806                dels += patch->lines_deleted;
2807                show_stats(patch);
2808        }
2809
2810        printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2811}
2812
2813static void numstat_patch_list(struct patch *patch)
2814{
2815        for ( ; patch; patch = patch->next) {
2816                const char *name;
2817                name = patch->new_name ? patch->new_name : patch->old_name;
2818                if (patch->is_binary)
2819                        printf("-\t-\t");
2820                else
2821                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2822                write_name_quoted(name, stdout, line_termination);
2823        }
2824}
2825
2826static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2827{
2828        if (mode)
2829                printf(" %s mode %06o %s\n", newdelete, mode, name);
2830        else
2831                printf(" %s %s\n", newdelete, name);
2832}
2833
2834static void show_mode_change(struct patch *p, int show_name)
2835{
2836        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2837                if (show_name)
2838                        printf(" mode change %06o => %06o %s\n",
2839                               p->old_mode, p->new_mode, p->new_name);
2840                else
2841                        printf(" mode change %06o => %06o\n",
2842                               p->old_mode, p->new_mode);
2843        }
2844}
2845
2846static void show_rename_copy(struct patch *p)
2847{
2848        const char *renamecopy = p->is_rename ? "rename" : "copy";
2849        const char *old, *new;
2850
2851        /* Find common prefix */
2852        old = p->old_name;
2853        new = p->new_name;
2854        while (1) {
2855                const char *slash_old, *slash_new;
2856                slash_old = strchr(old, '/');
2857                slash_new = strchr(new, '/');
2858                if (!slash_old ||
2859                    !slash_new ||
2860                    slash_old - old != slash_new - new ||
2861                    memcmp(old, new, slash_new - new))
2862                        break;
2863                old = slash_old + 1;
2864                new = slash_new + 1;
2865        }
2866        /* p->old_name thru old is the common prefix, and old and new
2867         * through the end of names are renames
2868         */
2869        if (old != p->old_name)
2870                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2871                       (int)(old - p->old_name), p->old_name,
2872                       old, new, p->score);
2873        else
2874                printf(" %s %s => %s (%d%%)\n", renamecopy,
2875                       p->old_name, p->new_name, p->score);
2876        show_mode_change(p, 0);
2877}
2878
2879static void summary_patch_list(struct patch *patch)
2880{
2881        struct patch *p;
2882
2883        for (p = patch; p; p = p->next) {
2884                if (p->is_new)
2885                        show_file_mode_name("create", p->new_mode, p->new_name);
2886                else if (p->is_delete)
2887                        show_file_mode_name("delete", p->old_mode, p->old_name);
2888                else {
2889                        if (p->is_rename || p->is_copy)
2890                                show_rename_copy(p);
2891                        else {
2892                                if (p->score) {
2893                                        printf(" rewrite %s (%d%%)\n",
2894                                               p->new_name, p->score);
2895                                        show_mode_change(p, 0);
2896                                }
2897                                else
2898                                        show_mode_change(p, 1);
2899                        }
2900                }
2901        }
2902}
2903
2904static void patch_stats(struct patch *patch)
2905{
2906        int lines = patch->lines_added + patch->lines_deleted;
2907
2908        if (lines > max_change)
2909                max_change = lines;
2910        if (patch->old_name) {
2911                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2912                if (!len)
2913                        len = strlen(patch->old_name);
2914                if (len > max_len)
2915                        max_len = len;
2916        }
2917        if (patch->new_name) {
2918                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2919                if (!len)
2920                        len = strlen(patch->new_name);
2921                if (len > max_len)
2922                        max_len = len;
2923        }
2924}
2925
2926static void remove_file(struct patch *patch, int rmdir_empty)
2927{
2928        if (update_index) {
2929                if (remove_file_from_cache(patch->old_name) < 0)
2930                        die("unable to remove %s from index", patch->old_name);
2931        }
2932        if (!cached) {
2933                if (S_ISGITLINK(patch->old_mode)) {
2934                        if (rmdir(patch->old_name))
2935                                warning("unable to remove submodule %s",
2936                                        patch->old_name);
2937                } else if (!unlink_or_warn(patch->old_name) && rmdir_empty) {
2938                        remove_path(patch->old_name);
2939                }
2940        }
2941}
2942
2943static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2944{
2945        struct stat st;
2946        struct cache_entry *ce;
2947        int namelen = strlen(path);
2948        unsigned ce_size = cache_entry_size(namelen);
2949
2950        if (!update_index)
2951                return;
2952
2953        ce = xcalloc(1, ce_size);
2954        memcpy(ce->name, path, namelen);
2955        ce->ce_mode = create_ce_mode(mode);
2956        ce->ce_flags = namelen;
2957        if (S_ISGITLINK(mode)) {
2958                const char *s = buf;
2959
2960                if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2961                        die("corrupt patch for subproject %s", path);
2962        } else {
2963                if (!cached) {
2964                        if (lstat(path, &st) < 0)
2965                                die_errno("unable to stat newly created file '%s'",
2966                                          path);
2967                        fill_stat_cache_info(ce, &st);
2968                }
2969                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2970                        die("unable to create backing store for newly created file %s", path);
2971        }
2972        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2973                die("unable to add cache entry for %s", path);
2974}
2975
2976static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2977{
2978        int fd;
2979        struct strbuf nbuf = STRBUF_INIT;
2980
2981        if (S_ISGITLINK(mode)) {
2982                struct stat st;
2983                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2984                        return 0;
2985                return mkdir(path, 0777);
2986        }
2987
2988        if (has_symlinks && S_ISLNK(mode))
2989                /* Although buf:size is counted string, it also is NUL
2990                 * terminated.
2991                 */
2992                return symlink(buf, path);
2993
2994        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2995        if (fd < 0)
2996                return -1;
2997
2998        if (convert_to_working_tree(path, buf, size, &nbuf)) {
2999                size = nbuf.len;
3000                buf  = nbuf.buf;
3001        }
3002        write_or_die(fd, buf, size);
3003        strbuf_release(&nbuf);
3004
3005        if (close(fd) < 0)
3006                die_errno("closing file '%s'", path);
3007        return 0;
3008}
3009
3010/*
3011 * We optimistically assume that the directories exist,
3012 * which is true 99% of the time anyway. If they don't,
3013 * we create them and try again.
3014 */
3015static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3016{
3017        if (cached)
3018                return;
3019        if (!try_create_file(path, mode, buf, size))
3020                return;
3021
3022        if (errno == ENOENT) {
3023                if (safe_create_leading_directories(path))
3024                        return;
3025                if (!try_create_file(path, mode, buf, size))
3026                        return;
3027        }
3028
3029        if (errno == EEXIST || errno == EACCES) {
3030                /* We may be trying to create a file where a directory
3031                 * used to be.
3032                 */
3033                struct stat st;
3034                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3035                        errno = EEXIST;
3036        }
3037
3038        if (errno == EEXIST) {
3039                unsigned int nr = getpid();
3040
3041                for (;;) {
3042                        char newpath[PATH_MAX];
3043                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3044                        if (!try_create_file(newpath, mode, buf, size)) {
3045                                if (!rename(newpath, path))
3046                                        return;
3047                                unlink_or_warn(newpath);
3048                                break;
3049                        }
3050                        if (errno != EEXIST)
3051                                break;
3052                        ++nr;
3053                }
3054        }
3055        die_errno("unable to write file '%s' mode %o", path, mode);
3056}
3057
3058static void create_file(struct patch *patch)
3059{
3060        char *path = patch->new_name;
3061        unsigned mode = patch->new_mode;
3062        unsigned long size = patch->resultsize;
3063        char *buf = patch->result;
3064
3065        if (!mode)
3066                mode = S_IFREG | 0644;
3067        create_one_file(path, mode, buf, size);
3068        add_index_file(path, mode, buf, size);
3069}
3070
3071/* phase zero is to remove, phase one is to create */
3072static void write_out_one_result(struct patch *patch, int phase)
3073{
3074        if (patch->is_delete > 0) {
3075                if (phase == 0)
3076                        remove_file(patch, 1);
3077                return;
3078        }
3079        if (patch->is_new > 0 || patch->is_copy) {
3080                if (phase == 1)
3081                        create_file(patch);
3082                return;
3083        }
3084        /*
3085         * Rename or modification boils down to the same
3086         * thing: remove the old, write the new
3087         */
3088        if (phase == 0)
3089                remove_file(patch, patch->is_rename);
3090        if (phase == 1)
3091                create_file(patch);
3092}
3093
3094static int write_out_one_reject(struct patch *patch)
3095{
3096        FILE *rej;
3097        char namebuf[PATH_MAX];
3098        struct fragment *frag;
3099        int cnt = 0;
3100
3101        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3102                if (!frag->rejected)
3103                        continue;
3104                cnt++;
3105        }
3106
3107        if (!cnt) {
3108                if (apply_verbosely)
3109                        say_patch_name(stderr,
3110                                       "Applied patch ", patch, " cleanly.\n");
3111                return 0;
3112        }
3113
3114        /* This should not happen, because a removal patch that leaves
3115         * contents are marked "rejected" at the patch level.
3116         */
3117        if (!patch->new_name)
3118                die("internal error");
3119
3120        /* Say this even without --verbose */
3121        say_patch_name(stderr, "Applying patch ", patch, " with");
3122        fprintf(stderr, " %d rejects...\n", cnt);
3123
3124        cnt = strlen(patch->new_name);
3125        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3126                cnt = ARRAY_SIZE(namebuf) - 5;
3127                warning("truncating .rej filename to %.*s.rej",
3128                        cnt - 1, patch->new_name);
3129        }
3130        memcpy(namebuf, patch->new_name, cnt);
3131        memcpy(namebuf + cnt, ".rej", 5);
3132
3133        rej = fopen(namebuf, "w");
3134        if (!rej)
3135                return error("cannot open %s: %s", namebuf, strerror(errno));
3136
3137        /* Normal git tools never deal with .rej, so do not pretend
3138         * this is a git patch by saying --git nor give extended
3139         * headers.  While at it, maybe please "kompare" that wants
3140         * the trailing TAB and some garbage at the end of line ;-).
3141         */
3142        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3143                patch->new_name, patch->new_name);
3144        for (cnt = 1, frag = patch->fragments;
3145             frag;
3146             cnt++, frag = frag->next) {
3147                if (!frag->rejected) {
3148                        fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3149                        continue;
3150                }
3151                fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3152                fprintf(rej, "%.*s", frag->size, frag->patch);
3153                if (frag->patch[frag->size-1] != '\n')
3154                        fputc('\n', rej);
3155        }
3156        fclose(rej);
3157        return -1;
3158}
3159
3160static int write_out_results(struct patch *list, int skipped_patch)
3161{
3162        int phase;
3163        int errs = 0;
3164        struct patch *l;
3165
3166        if (!list && !skipped_patch)
3167                return error("No changes");
3168
3169        for (phase = 0; phase < 2; phase++) {
3170                l = list;
3171                while (l) {
3172                        if (l->rejected)
3173                                errs = 1;
3174                        else {
3175                                write_out_one_result(l, phase);
3176                                if (phase == 1 && write_out_one_reject(l))
3177                                        errs = 1;
3178                        }
3179                        l = l->next;
3180                }
3181        }
3182        return errs;
3183}
3184
3185static struct lock_file lock_file;
3186
3187static struct string_list limit_by_name;
3188static int has_include;
3189static void add_name_limit(const char *name, int exclude)
3190{
3191        struct string_list_item *it;
3192
3193        it = string_list_append(name, &limit_by_name);
3194        it->util = exclude ? NULL : (void *) 1;
3195}
3196
3197static int use_patch(struct patch *p)
3198{
3199        const char *pathname = p->new_name ? p->new_name : p->old_name;
3200        int i;
3201
3202        /* Paths outside are not touched regardless of "--include" */
3203        if (0 < prefix_length) {
3204                int pathlen = strlen(pathname);
3205                if (pathlen <= prefix_length ||
3206                    memcmp(prefix, pathname, prefix_length))
3207                        return 0;
3208        }
3209
3210        /* See if it matches any of exclude/include rule */
3211        for (i = 0; i < limit_by_name.nr; i++) {
3212                struct string_list_item *it = &limit_by_name.items[i];
3213                if (!fnmatch(it->string, pathname, 0))
3214                        return (it->util != NULL);
3215        }
3216
3217        /*
3218         * If we had any include, a path that does not match any rule is
3219         * not used.  Otherwise, we saw bunch of exclude rules (or none)
3220         * and such a path is used.
3221         */
3222        return !has_include;
3223}
3224
3225
3226static void prefix_one(char **name)
3227{
3228        char *old_name = *name;
3229        if (!old_name)
3230                return;
3231        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3232        free(old_name);
3233}
3234
3235static void prefix_patches(struct patch *p)
3236{
3237        if (!prefix || p->is_toplevel_relative)
3238                return;
3239        for ( ; p; p = p->next) {
3240                if (p->new_name == p->old_name) {
3241                        char *prefixed = p->new_name;
3242                        prefix_one(&prefixed);
3243                        p->new_name = p->old_name = prefixed;
3244                }
3245                else {
3246                        prefix_one(&p->new_name);
3247                        prefix_one(&p->old_name);
3248                }
3249        }
3250}
3251
3252#define INACCURATE_EOF  (1<<0)
3253#define RECOUNT         (1<<1)
3254
3255static int apply_patch(int fd, const char *filename, int options)
3256{
3257        size_t offset;
3258        struct strbuf buf = STRBUF_INIT;
3259        struct patch *list = NULL, **listp = &list;
3260        int skipped_patch = 0;
3261
3262        /* FIXME - memory leak when using multiple patch files as inputs */
3263        memset(&fn_table, 0, sizeof(struct string_list));
3264        patch_input_file = filename;
3265        read_patch_file(&buf, fd);
3266        offset = 0;
3267        while (offset < buf.len) {
3268                struct patch *patch;
3269                int nr;
3270
3271                patch = xcalloc(1, sizeof(*patch));
3272                patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3273                patch->recount =  !!(options & RECOUNT);
3274                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3275                if (nr < 0)
3276                        break;
3277                if (apply_in_reverse)
3278                        reverse_patches(patch);
3279                if (prefix)
3280                        prefix_patches(patch);
3281                if (use_patch(patch)) {
3282                        patch_stats(patch);
3283                        *listp = patch;
3284                        listp = &patch->next;
3285                }
3286                else {
3287                        /* perhaps free it a bit better? */
3288                        free(patch);
3289                        skipped_patch++;
3290                }
3291                offset += nr;
3292        }
3293
3294        if (whitespace_error && (ws_error_action == die_on_ws_error))
3295                apply = 0;
3296
3297        update_index = check_index && apply;
3298        if (update_index && newfd < 0)
3299                newfd = hold_locked_index(&lock_file, 1);
3300
3301        if (check_index) {
3302                if (read_cache() < 0)
3303                        die("unable to read index file");
3304        }
3305
3306        if ((check || apply) &&
3307            check_patch_list(list) < 0 &&
3308            !apply_with_reject)
3309                exit(1);
3310
3311        if (apply && write_out_results(list, skipped_patch))
3312                exit(1);
3313
3314        if (fake_ancestor)
3315                build_fake_ancestor(list, fake_ancestor);
3316
3317        if (diffstat)
3318                stat_patch_list(list);
3319
3320        if (numstat)
3321                numstat_patch_list(list);
3322
3323        if (summary)
3324                summary_patch_list(list);
3325
3326        strbuf_release(&buf);
3327        return 0;
3328}
3329
3330static int git_apply_config(const char *var, const char *value, void *cb)
3331{
3332        if (!strcmp(var, "apply.whitespace"))
3333                return git_config_string(&apply_default_whitespace, var, value);
3334        else if (!strcmp(var, "apply.ignorewhitespace"))
3335                return git_config_string(&apply_default_ignorewhitespace, var, value);
3336        return git_default_config(var, value, cb);
3337}
3338
3339static int option_parse_exclude(const struct option *opt,
3340                                const char *arg, int unset)
3341{
3342        add_name_limit(arg, 1);
3343        return 0;
3344}
3345
3346static int option_parse_include(const struct option *opt,
3347                                const char *arg, int unset)
3348{
3349        add_name_limit(arg, 0);
3350        has_include = 1;
3351        return 0;
3352}
3353
3354static int option_parse_p(const struct option *opt,
3355                          const char *arg, int unset)
3356{
3357        p_value = atoi(arg);
3358        p_value_known = 1;
3359        return 0;
3360}
3361
3362static int option_parse_z(const struct option *opt,
3363                          const char *arg, int unset)
3364{
3365        if (unset)
3366                line_termination = '\n';
3367        else
3368                line_termination = 0;
3369        return 0;
3370}
3371
3372static int option_parse_space_change(const struct option *opt,
3373                          const char *arg, int unset)
3374{
3375        if (unset)
3376                ws_ignore_action = ignore_ws_none;
3377        else
3378                ws_ignore_action = ignore_ws_change;
3379        return 0;
3380}
3381
3382static int option_parse_whitespace(const struct option *opt,
3383                                   const char *arg, int unset)
3384{
3385        const char **whitespace_option = opt->value;
3386
3387        *whitespace_option = arg;
3388        parse_whitespace_option(arg);
3389        return 0;
3390}
3391
3392static int option_parse_directory(const struct option *opt,
3393                                  const char *arg, int unset)
3394{
3395        root_len = strlen(arg);
3396        if (root_len && arg[root_len - 1] != '/') {
3397                char *new_root;
3398                root = new_root = xmalloc(root_len + 2);
3399                strcpy(new_root, arg);
3400                strcpy(new_root + root_len++, "/");
3401        } else
3402                root = arg;
3403        return 0;
3404}
3405
3406int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3407{
3408        int i;
3409        int errs = 0;
3410        int is_not_gitdir;
3411        int binary;
3412        int force_apply = 0;
3413
3414        const char *whitespace_option = NULL;
3415
3416        struct option builtin_apply_options[] = {
3417                { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3418                        "don't apply changes matching the given path",
3419                        0, option_parse_exclude },
3420                { OPTION_CALLBACK, 0, "include", NULL, "path",
3421                        "apply changes matching the given path",
3422                        0, option_parse_include },
3423                { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3424                        "remove <num> leading slashes from traditional diff paths",
3425                        0, option_parse_p },
3426                OPT_BOOLEAN(0, "no-add", &no_add,
3427                        "ignore additions made by the patch"),
3428                OPT_BOOLEAN(0, "stat", &diffstat,
3429                        "instead of applying the patch, output diffstat for the input"),
3430                { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3431                  NULL, "old option, now no-op",
3432                  PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3433                { OPTION_BOOLEAN, 0, "binary", &binary,
3434                  NULL, "old option, now no-op",
3435                  PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3436                OPT_BOOLEAN(0, "numstat", &numstat,
3437                        "shows number of added and deleted lines in decimal notation"),
3438                OPT_BOOLEAN(0, "summary", &summary,
3439                        "instead of applying the patch, output a summary for the input"),
3440                OPT_BOOLEAN(0, "check", &check,
3441                        "instead of applying the patch, see if the patch is applicable"),
3442                OPT_BOOLEAN(0, "index", &check_index,
3443                        "make sure the patch is applicable to the current index"),
3444                OPT_BOOLEAN(0, "cached", &cached,
3445                        "apply a patch without touching the working tree"),
3446                OPT_BOOLEAN(0, "apply", &force_apply,
3447                        "also apply the patch (use with --stat/--summary/--check)"),
3448                OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3449                        "build a temporary index based on embedded index information"),
3450                { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3451                        "paths are separated with NUL character",
3452                        PARSE_OPT_NOARG, option_parse_z },
3453                OPT_INTEGER('C', NULL, &p_context,
3454                                "ensure at least <n> lines of context match"),
3455                { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3456                        "detect new or modified lines that have whitespace errors",
3457                        0, option_parse_whitespace },
3458                { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3459                        "ignore changes in whitespace when finding context",
3460                        PARSE_OPT_NOARG, option_parse_space_change },
3461                { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3462                        "ignore changes in whitespace when finding context",
3463                        PARSE_OPT_NOARG, option_parse_space_change },
3464                OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3465                        "apply the patch in reverse"),
3466                OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3467                        "don't expect at least one line of context"),
3468                OPT_BOOLEAN(0, "reject", &apply_with_reject,
3469                        "leave the rejected hunks in corresponding *.rej files"),
3470                OPT__VERBOSE(&apply_verbosely),
3471                OPT_BIT(0, "inaccurate-eof", &options,
3472                        "tolerate incorrectly detected missing new-line at the end of file",
3473                        INACCURATE_EOF),
3474                OPT_BIT(0, "recount", &options,
3475                        "do not trust the line counts in the hunk headers",
3476                        RECOUNT),
3477                { OPTION_CALLBACK, 0, "directory", NULL, "root",
3478                        "prepend <root> to all filenames",
3479                        0, option_parse_directory },
3480                OPT_END()
3481        };
3482
3483        prefix = setup_git_directory_gently(&is_not_gitdir);
3484        prefix_length = prefix ? strlen(prefix) : 0;
3485        git_config(git_apply_config, NULL);
3486        if (apply_default_whitespace)
3487                parse_whitespace_option(apply_default_whitespace);
3488        if (apply_default_ignorewhitespace)
3489                parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3490
3491        argc = parse_options(argc, argv, prefix, builtin_apply_options,
3492                        apply_usage, 0);
3493
3494        if (apply_with_reject)
3495                apply = apply_verbosely = 1;
3496        if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3497                apply = 0;
3498        if (check_index && is_not_gitdir)
3499                die("--index outside a repository");
3500        if (cached) {
3501                if (is_not_gitdir)
3502                        die("--cached outside a repository");
3503                check_index = 1;
3504        }
3505        for (i = 0; i < argc; i++) {
3506                const char *arg = argv[i];
3507                int fd;
3508
3509                if (!strcmp(arg, "-")) {
3510                        errs |= apply_patch(0, "<stdin>", options);
3511                        read_stdin = 0;
3512                        continue;
3513                } else if (0 < prefix_length)
3514                        arg = prefix_filename(prefix, prefix_length, arg);
3515
3516                fd = open(arg, O_RDONLY);
3517                if (fd < 0)
3518                        die_errno("can't open patch '%s'", arg);
3519                read_stdin = 0;
3520                set_default_whitespace_mode(whitespace_option);
3521                errs |= apply_patch(fd, arg, options);
3522                close(fd);
3523        }
3524        set_default_whitespace_mode(whitespace_option);
3525        if (read_stdin)
3526                errs |= apply_patch(0, "<stdin>", options);
3527        if (whitespace_error) {
3528                if (squelch_whitespace_errors &&
3529                    squelch_whitespace_errors < whitespace_error) {
3530                        int squelched =
3531                                whitespace_error - squelch_whitespace_errors;
3532                        warning("squelched %d "
3533                                "whitespace error%s",
3534                                squelched,
3535                                squelched == 1 ? "" : "s");
3536                }
3537                if (ws_error_action == die_on_ws_error)
3538                        die("%d line%s add%s whitespace errors.",
3539                            whitespace_error,
3540                            whitespace_error == 1 ? "" : "s",
3541                            whitespace_error == 1 ? "s" : "");
3542                if (applied_after_fixing_ws && apply)
3543                        warning("%d line%s applied after"
3544                                " fixing whitespace errors.",
3545                                applied_after_fixing_ws,
3546                                applied_after_fixing_ws == 1 ? "" : "s");
3547                else if (whitespace_error)
3548                        warning("%d line%s add%s whitespace errors.",
3549                                whitespace_error,
3550                                whitespace_error == 1 ? "" : "s",
3551                                whitespace_error == 1 ? "s" : "");
3552        }
3553
3554        if (update_index) {
3555                if (write_cache(newfd, active_cache, active_nr) ||
3556                    commit_locked_index(&lock_file))
3557                        die("Unable to write new index file");
3558        }
3559
3560        return !!errs;
3561}