builtin / apply.con commit "add -p": work-around an old laziness that does not coalesce hunks (933e44d)
   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 allow_overlap;
  47static int no_add;
  48static const char *fake_ancestor;
  49static int line_termination = '\n';
  50static unsigned int p_context = UINT_MAX;
  51static const char * const apply_usage[] = {
  52        "git apply [options] [<patch>...]",
  53        NULL
  54};
  55
  56static enum ws_error_action {
  57        nowarn_ws_error,
  58        warn_on_ws_error,
  59        die_on_ws_error,
  60        correct_ws_error
  61} ws_error_action = warn_on_ws_error;
  62static int whitespace_error;
  63static int squelch_whitespace_errors = 5;
  64static int applied_after_fixing_ws;
  65
  66static enum ws_ignore {
  67        ignore_ws_none,
  68        ignore_ws_change
  69} ws_ignore_action = ignore_ws_none;
  70
  71
  72static const char *patch_input_file;
  73static const char *root;
  74static int root_len;
  75static int read_stdin = 1;
  76static int options;
  77
  78static void parse_whitespace_option(const char *option)
  79{
  80        if (!option) {
  81                ws_error_action = warn_on_ws_error;
  82                return;
  83        }
  84        if (!strcmp(option, "warn")) {
  85                ws_error_action = warn_on_ws_error;
  86                return;
  87        }
  88        if (!strcmp(option, "nowarn")) {
  89                ws_error_action = nowarn_ws_error;
  90                return;
  91        }
  92        if (!strcmp(option, "error")) {
  93                ws_error_action = die_on_ws_error;
  94                return;
  95        }
  96        if (!strcmp(option, "error-all")) {
  97                ws_error_action = die_on_ws_error;
  98                squelch_whitespace_errors = 0;
  99                return;
 100        }
 101        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
 102                ws_error_action = correct_ws_error;
 103                return;
 104        }
 105        die("unrecognized whitespace option '%s'", option);
 106}
 107
 108static void parse_ignorewhitespace_option(const char *option)
 109{
 110        if (!option || !strcmp(option, "no") ||
 111            !strcmp(option, "false") || !strcmp(option, "never") ||
 112            !strcmp(option, "none")) {
 113                ws_ignore_action = ignore_ws_none;
 114                return;
 115        }
 116        if (!strcmp(option, "change")) {
 117                ws_ignore_action = ignore_ws_change;
 118                return;
 119        }
 120        die("unrecognized whitespace ignore option '%s'", option);
 121}
 122
 123static void set_default_whitespace_mode(const char *whitespace_option)
 124{
 125        if (!whitespace_option && !apply_default_whitespace)
 126                ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
 127}
 128
 129/*
 130 * For "diff-stat" like behaviour, we keep track of the biggest change
 131 * we've seen, and the longest filename. That allows us to do simple
 132 * scaling.
 133 */
 134static int max_change, max_len;
 135
 136/*
 137 * Various "current state", notably line numbers and what
 138 * file (and how) we're patching right now.. The "is_xxxx"
 139 * things are flags, where -1 means "don't know yet".
 140 */
 141static int linenr = 1;
 142
 143/*
 144 * This represents one "hunk" from a patch, starting with
 145 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 146 * patch text is pointed at by patch, and its byte length
 147 * is stored in size.  leading and trailing are the number
 148 * of context lines.
 149 */
 150struct fragment {
 151        unsigned long leading, trailing;
 152        unsigned long oldpos, oldlines;
 153        unsigned long newpos, newlines;
 154        const char *patch;
 155        int size;
 156        int rejected;
 157        int linenr;
 158        struct fragment *next;
 159};
 160
 161/*
 162 * When dealing with a binary patch, we reuse "leading" field
 163 * to store the type of the binary hunk, either deflated "delta"
 164 * or deflated "literal".
 165 */
 166#define binary_patch_method leading
 167#define BINARY_DELTA_DEFLATED   1
 168#define BINARY_LITERAL_DEFLATED 2
 169
 170/*
 171 * This represents a "patch" to a file, both metainfo changes
 172 * such as creation/deletion, filemode and content changes represented
 173 * as a series of fragments.
 174 */
 175struct patch {
 176        char *new_name, *old_name, *def_name;
 177        unsigned int old_mode, new_mode;
 178        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 179        int rejected;
 180        unsigned ws_rule;
 181        unsigned long deflate_origlen;
 182        int lines_added, lines_deleted;
 183        int score;
 184        unsigned int is_toplevel_relative:1;
 185        unsigned int inaccurate_eof:1;
 186        unsigned int is_binary:1;
 187        unsigned int is_copy:1;
 188        unsigned int is_rename:1;
 189        unsigned int recount:1;
 190        struct fragment *fragments;
 191        char *result;
 192        size_t resultsize;
 193        char old_sha1_prefix[41];
 194        char new_sha1_prefix[41];
 195        struct patch *next;
 196};
 197
 198/*
 199 * A line in a file, len-bytes long (includes the terminating LF,
 200 * except for an incomplete line at the end if the file ends with
 201 * one), and its contents hashes to 'hash'.
 202 */
 203struct line {
 204        size_t len;
 205        unsigned hash : 24;
 206        unsigned flag : 8;
 207#define LINE_COMMON     1
 208#define LINE_PATCHED    2
 209};
 210
 211/*
 212 * This represents a "file", which is an array of "lines".
 213 */
 214struct image {
 215        char *buf;
 216        size_t len;
 217        size_t nr;
 218        size_t alloc;
 219        struct line *line_allocated;
 220        struct line *line;
 221};
 222
 223/*
 224 * Records filenames that have been touched, in order to handle
 225 * the case where more than one patches touch the same file.
 226 */
 227
 228static struct string_list fn_table;
 229
 230static uint32_t hash_line(const char *cp, size_t len)
 231{
 232        size_t i;
 233        uint32_t h;
 234        for (i = 0, h = 0; i < len; i++) {
 235                if (!isspace(cp[i])) {
 236                        h = h * 3 + (cp[i] & 0xff);
 237                }
 238        }
 239        return h;
 240}
 241
 242/*
 243 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 244 * whitespace difference. Returns 1 if they match, 0 otherwise
 245 */
 246static int fuzzy_matchlines(const char *s1, size_t n1,
 247                            const char *s2, size_t n2)
 248{
 249        const char *last1 = s1 + n1 - 1;
 250        const char *last2 = s2 + n2 - 1;
 251        int result = 0;
 252
 253        if (n1 < 0 || n2 < 0)
 254                return 0;
 255
 256        /* ignore line endings */
 257        while ((*last1 == '\r') || (*last1 == '\n'))
 258                last1--;
 259        while ((*last2 == '\r') || (*last2 == '\n'))
 260                last2--;
 261
 262        /* skip leading whitespace */
 263        while (isspace(*s1) && (s1 <= last1))
 264                s1++;
 265        while (isspace(*s2) && (s2 <= last2))
 266                s2++;
 267        /* early return if both lines are empty */
 268        if ((s1 > last1) && (s2 > last2))
 269                return 1;
 270        while (!result) {
 271                result = *s1++ - *s2++;
 272                /*
 273                 * Skip whitespace inside. We check for whitespace on
 274                 * both buffers because we don't want "a b" to match
 275                 * "ab"
 276                 */
 277                if (isspace(*s1) && isspace(*s2)) {
 278                        while (isspace(*s1) && s1 <= last1)
 279                                s1++;
 280                        while (isspace(*s2) && s2 <= last2)
 281                                s2++;
 282                }
 283                /*
 284                 * If we reached the end on one side only,
 285                 * lines don't match
 286                 */
 287                if (
 288                    ((s2 > last2) && (s1 <= last1)) ||
 289                    ((s1 > last1) && (s2 <= last2)))
 290                        return 0;
 291                if ((s1 > last1) && (s2 > last2))
 292                        break;
 293        }
 294
 295        return !result;
 296}
 297
 298static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 299{
 300        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 301        img->line_allocated[img->nr].len = len;
 302        img->line_allocated[img->nr].hash = hash_line(bol, len);
 303        img->line_allocated[img->nr].flag = flag;
 304        img->nr++;
 305}
 306
 307static void prepare_image(struct image *image, char *buf, size_t len,
 308                          int prepare_linetable)
 309{
 310        const char *cp, *ep;
 311
 312        memset(image, 0, sizeof(*image));
 313        image->buf = buf;
 314        image->len = len;
 315
 316        if (!prepare_linetable)
 317                return;
 318
 319        ep = image->buf + image->len;
 320        cp = image->buf;
 321        while (cp < ep) {
 322                const char *next;
 323                for (next = cp; next < ep && *next != '\n'; next++)
 324                        ;
 325                if (next < ep)
 326                        next++;
 327                add_line_info(image, cp, next - cp, 0);
 328                cp = next;
 329        }
 330        image->line = image->line_allocated;
 331}
 332
 333static void clear_image(struct image *image)
 334{
 335        free(image->buf);
 336        image->buf = NULL;
 337        image->len = 0;
 338}
 339
 340static void say_patch_name(FILE *output, const char *pre,
 341                           struct patch *patch, const char *post)
 342{
 343        fputs(pre, output);
 344        if (patch->old_name && patch->new_name &&
 345            strcmp(patch->old_name, patch->new_name)) {
 346                quote_c_style(patch->old_name, NULL, output, 0);
 347                fputs(" => ", output);
 348                quote_c_style(patch->new_name, NULL, output, 0);
 349        } else {
 350                const char *n = patch->new_name;
 351                if (!n)
 352                        n = patch->old_name;
 353                quote_c_style(n, NULL, output, 0);
 354        }
 355        fputs(post, output);
 356}
 357
 358#define CHUNKSIZE (8192)
 359#define SLOP (16)
 360
 361static void read_patch_file(struct strbuf *sb, int fd)
 362{
 363        if (strbuf_read(sb, fd, 0) < 0)
 364                die_errno("git apply: failed to read");
 365
 366        /*
 367         * Make sure that we have some slop in the buffer
 368         * so that we can do speculative "memcmp" etc, and
 369         * see to it that it is NUL-filled.
 370         */
 371        strbuf_grow(sb, SLOP);
 372        memset(sb->buf + sb->len, 0, SLOP);
 373}
 374
 375static unsigned long linelen(const char *buffer, unsigned long size)
 376{
 377        unsigned long len = 0;
 378        while (size--) {
 379                len++;
 380                if (*buffer++ == '\n')
 381                        break;
 382        }
 383        return len;
 384}
 385
 386static int is_dev_null(const char *str)
 387{
 388        return !memcmp("/dev/null", str, 9) && isspace(str[9]);
 389}
 390
 391#define TERM_SPACE      1
 392#define TERM_TAB        2
 393
 394static int name_terminate(const char *name, int namelen, int c, int terminate)
 395{
 396        if (c == ' ' && !(terminate & TERM_SPACE))
 397                return 0;
 398        if (c == '\t' && !(terminate & TERM_TAB))
 399                return 0;
 400
 401        return 1;
 402}
 403
 404/* remove double slashes to make --index work with such filenames */
 405static char *squash_slash(char *name)
 406{
 407        int i = 0, j = 0;
 408
 409        if (!name)
 410                return NULL;
 411
 412        while (name[i]) {
 413                if ((name[j++] = name[i++]) == '/')
 414                        while (name[i] == '/')
 415                                i++;
 416        }
 417        name[j] = '\0';
 418        return name;
 419}
 420
 421static char *find_name_gnu(const char *line, char *def, int p_value)
 422{
 423        struct strbuf name = STRBUF_INIT;
 424        char *cp;
 425
 426        /*
 427         * Proposed "new-style" GNU patch/diff format; see
 428         * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
 429         */
 430        if (unquote_c_style(&name, line, NULL)) {
 431                strbuf_release(&name);
 432                return NULL;
 433        }
 434
 435        for (cp = name.buf; p_value; p_value--) {
 436                cp = strchr(cp, '/');
 437                if (!cp) {
 438                        strbuf_release(&name);
 439                        return NULL;
 440                }
 441                cp++;
 442        }
 443
 444        /* name can later be freed, so we need
 445         * to memmove, not just return cp
 446         */
 447        strbuf_remove(&name, 0, cp - name.buf);
 448        free(def);
 449        if (root)
 450                strbuf_insert(&name, 0, root, root_len);
 451        return squash_slash(strbuf_detach(&name, NULL));
 452}
 453
 454static size_t sane_tz_len(const char *line, size_t len)
 455{
 456        const char *tz, *p;
 457
 458        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 459                return 0;
 460        tz = line + len - strlen(" +0500");
 461
 462        if (tz[1] != '+' && tz[1] != '-')
 463                return 0;
 464
 465        for (p = tz + 2; p != line + len; p++)
 466                if (!isdigit(*p))
 467                        return 0;
 468
 469        return line + len - tz;
 470}
 471
 472static size_t tz_with_colon_len(const char *line, size_t len)
 473{
 474        const char *tz, *p;
 475
 476        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 477                return 0;
 478        tz = line + len - strlen(" +08:00");
 479
 480        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 481                return 0;
 482        p = tz + 2;
 483        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 484            !isdigit(*p++) || !isdigit(*p++))
 485                return 0;
 486
 487        return line + len - tz;
 488}
 489
 490static size_t date_len(const char *line, size_t len)
 491{
 492        const char *date, *p;
 493
 494        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 495                return 0;
 496        p = date = line + len - strlen("72-02-05");
 497
 498        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 499            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 500            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 501                return 0;
 502
 503        if (date - line >= strlen("19") &&
 504            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 505                date -= strlen("19");
 506
 507        return line + len - date;
 508}
 509
 510static size_t short_time_len(const char *line, size_t len)
 511{
 512        const char *time, *p;
 513
 514        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 515                return 0;
 516        p = time = line + len - strlen(" 07:01:32");
 517
 518        /* Permit 1-digit hours? */
 519        if (*p++ != ' ' ||
 520            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 521            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 522            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 523                return 0;
 524
 525        return line + len - time;
 526}
 527
 528static size_t fractional_time_len(const char *line, size_t len)
 529{
 530        const char *p;
 531        size_t n;
 532
 533        /* Expected format: 19:41:17.620000023 */
 534        if (!len || !isdigit(line[len - 1]))
 535                return 0;
 536        p = line + len - 1;
 537
 538        /* Fractional seconds. */
 539        while (p > line && isdigit(*p))
 540                p--;
 541        if (*p != '.')
 542                return 0;
 543
 544        /* Hours, minutes, and whole seconds. */
 545        n = short_time_len(line, p - line);
 546        if (!n)
 547                return 0;
 548
 549        return line + len - p + n;
 550}
 551
 552static size_t trailing_spaces_len(const char *line, size_t len)
 553{
 554        const char *p;
 555
 556        /* Expected format: ' ' x (1 or more)  */
 557        if (!len || line[len - 1] != ' ')
 558                return 0;
 559
 560        p = line + len;
 561        while (p != line) {
 562                p--;
 563                if (*p != ' ')
 564                        return line + len - (p + 1);
 565        }
 566
 567        /* All spaces! */
 568        return len;
 569}
 570
 571static size_t diff_timestamp_len(const char *line, size_t len)
 572{
 573        const char *end = line + len;
 574        size_t n;
 575
 576        /*
 577         * Posix: 2010-07-05 19:41:17
 578         * GNU: 2010-07-05 19:41:17.620000023 -0500
 579         */
 580
 581        if (!isdigit(end[-1]))
 582                return 0;
 583
 584        n = sane_tz_len(line, end - line);
 585        if (!n)
 586                n = tz_with_colon_len(line, end - line);
 587        end -= n;
 588
 589        n = short_time_len(line, end - line);
 590        if (!n)
 591                n = fractional_time_len(line, end - line);
 592        end -= n;
 593
 594        n = date_len(line, end - line);
 595        if (!n) /* No date.  Too bad. */
 596                return 0;
 597        end -= n;
 598
 599        if (end == line)        /* No space before date. */
 600                return 0;
 601        if (end[-1] == '\t') {  /* Success! */
 602                end--;
 603                return line + len - end;
 604        }
 605        if (end[-1] != ' ')     /* No space before date. */
 606                return 0;
 607
 608        /* Whitespace damage. */
 609        end -= trailing_spaces_len(line, end - line);
 610        return line + len - end;
 611}
 612
 613static char *find_name_common(const char *line, char *def, int p_value,
 614                                const char *end, int terminate)
 615{
 616        int len;
 617        const char *start = NULL;
 618
 619        if (p_value == 0)
 620                start = line;
 621        while (line != end) {
 622                char c = *line;
 623
 624                if (!end && isspace(c)) {
 625                        if (c == '\n')
 626                                break;
 627                        if (name_terminate(start, line-start, c, terminate))
 628                                break;
 629                }
 630                line++;
 631                if (c == '/' && !--p_value)
 632                        start = line;
 633        }
 634        if (!start)
 635                return squash_slash(def);
 636        len = line - start;
 637        if (!len)
 638                return squash_slash(def);
 639
 640        /*
 641         * Generally we prefer the shorter name, especially
 642         * if the other one is just a variation of that with
 643         * something else tacked on to the end (ie "file.orig"
 644         * or "file~").
 645         */
 646        if (def) {
 647                int deflen = strlen(def);
 648                if (deflen < len && !strncmp(start, def, deflen))
 649                        return squash_slash(def);
 650                free(def);
 651        }
 652
 653        if (root) {
 654                char *ret = xmalloc(root_len + len + 1);
 655                strcpy(ret, root);
 656                memcpy(ret + root_len, start, len);
 657                ret[root_len + len] = '\0';
 658                return squash_slash(ret);
 659        }
 660
 661        return squash_slash(xmemdupz(start, len));
 662}
 663
 664static char *find_name(const char *line, char *def, int p_value, int terminate)
 665{
 666        if (*line == '"') {
 667                char *name = find_name_gnu(line, def, p_value);
 668                if (name)
 669                        return name;
 670        }
 671
 672        return find_name_common(line, def, p_value, NULL, terminate);
 673}
 674
 675static char *find_name_traditional(const char *line, char *def, int p_value)
 676{
 677        size_t len = strlen(line);
 678        size_t date_len;
 679
 680        if (*line == '"') {
 681                char *name = find_name_gnu(line, def, p_value);
 682                if (name)
 683                        return name;
 684        }
 685
 686        len = strchrnul(line, '\n') - line;
 687        date_len = diff_timestamp_len(line, len);
 688        if (!date_len)
 689                return find_name_common(line, def, p_value, NULL, TERM_TAB);
 690        len -= date_len;
 691
 692        return find_name_common(line, def, p_value, line + len, 0);
 693}
 694
 695static int count_slashes(const char *cp)
 696{
 697        int cnt = 0;
 698        char ch;
 699
 700        while ((ch = *cp++))
 701                if (ch == '/')
 702                        cnt++;
 703        return cnt;
 704}
 705
 706/*
 707 * Given the string after "--- " or "+++ ", guess the appropriate
 708 * p_value for the given patch.
 709 */
 710static int guess_p_value(const char *nameline)
 711{
 712        char *name, *cp;
 713        int val = -1;
 714
 715        if (is_dev_null(nameline))
 716                return -1;
 717        name = find_name_traditional(nameline, NULL, 0);
 718        if (!name)
 719                return -1;
 720        cp = strchr(name, '/');
 721        if (!cp)
 722                val = 0;
 723        else if (prefix) {
 724                /*
 725                 * Does it begin with "a/$our-prefix" and such?  Then this is
 726                 * very likely to apply to our directory.
 727                 */
 728                if (!strncmp(name, prefix, prefix_length))
 729                        val = count_slashes(prefix);
 730                else {
 731                        cp++;
 732                        if (!strncmp(cp, prefix, prefix_length))
 733                                val = count_slashes(prefix) + 1;
 734                }
 735        }
 736        free(name);
 737        return val;
 738}
 739
 740/*
 741 * Does the ---/+++ line has the POSIX timestamp after the last HT?
 742 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 743 * this such a timestamp?
 744 */
 745static int has_epoch_timestamp(const char *nameline)
 746{
 747        /*
 748         * We are only interested in epoch timestamp; any non-zero
 749         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 750         * For the same reason, the date must be either 1969-12-31 or
 751         * 1970-01-01, and the seconds part must be "00".
 752         */
 753        const char stamp_regexp[] =
 754                "^(1969-12-31|1970-01-01)"
 755                " "
 756                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 757                " "
 758                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 759        const char *timestamp = NULL, *cp, *colon;
 760        static regex_t *stamp;
 761        regmatch_t m[10];
 762        int zoneoffset;
 763        int hourminute;
 764        int status;
 765
 766        for (cp = nameline; *cp != '\n'; cp++) {
 767                if (*cp == '\t')
 768                        timestamp = cp + 1;
 769        }
 770        if (!timestamp)
 771                return 0;
 772        if (!stamp) {
 773                stamp = xmalloc(sizeof(*stamp));
 774                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 775                        warning("Cannot prepare timestamp regexp %s",
 776                                stamp_regexp);
 777                        return 0;
 778                }
 779        }
 780
 781        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 782        if (status) {
 783                if (status != REG_NOMATCH)
 784                        warning("regexec returned %d for input: %s",
 785                                status, timestamp);
 786                return 0;
 787        }
 788
 789        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 790        if (*colon == ':')
 791                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 792        else
 793                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 794        if (timestamp[m[3].rm_so] == '-')
 795                zoneoffset = -zoneoffset;
 796
 797        /*
 798         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 799         * (west of GMT) or 1970-01-01 (east of GMT)
 800         */
 801        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 802            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 803                return 0;
 804
 805        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 806                      strtol(timestamp + 14, NULL, 10) -
 807                      zoneoffset);
 808
 809        return ((zoneoffset < 0 && hourminute == 1440) ||
 810                (0 <= zoneoffset && !hourminute));
 811}
 812
 813/*
 814 * Get the name etc info from the ---/+++ lines of a traditional patch header
 815 *
 816 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 817 * files, we can happily check the index for a match, but for creating a
 818 * new file we should try to match whatever "patch" does. I have no idea.
 819 */
 820static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 821{
 822        char *name;
 823
 824        first += 4;     /* skip "--- " */
 825        second += 4;    /* skip "+++ " */
 826        if (!p_value_known) {
 827                int p, q;
 828                p = guess_p_value(first);
 829                q = guess_p_value(second);
 830                if (p < 0) p = q;
 831                if (0 <= p && p == q) {
 832                        p_value = p;
 833                        p_value_known = 1;
 834                }
 835        }
 836        if (is_dev_null(first)) {
 837                patch->is_new = 1;
 838                patch->is_delete = 0;
 839                name = find_name_traditional(second, NULL, p_value);
 840                patch->new_name = name;
 841        } else if (is_dev_null(second)) {
 842                patch->is_new = 0;
 843                patch->is_delete = 1;
 844                name = find_name_traditional(first, NULL, p_value);
 845                patch->old_name = name;
 846        } else {
 847                name = find_name_traditional(first, NULL, p_value);
 848                name = find_name_traditional(second, name, p_value);
 849                if (has_epoch_timestamp(first)) {
 850                        patch->is_new = 1;
 851                        patch->is_delete = 0;
 852                        patch->new_name = name;
 853                } else if (has_epoch_timestamp(second)) {
 854                        patch->is_new = 0;
 855                        patch->is_delete = 1;
 856                        patch->old_name = name;
 857                } else {
 858                        patch->old_name = patch->new_name = name;
 859                }
 860        }
 861        if (!name)
 862                die("unable to find filename in patch at line %d", linenr);
 863}
 864
 865static int gitdiff_hdrend(const char *line, struct patch *patch)
 866{
 867        return -1;
 868}
 869
 870/*
 871 * We're anal about diff header consistency, to make
 872 * sure that we don't end up having strange ambiguous
 873 * patches floating around.
 874 *
 875 * As a result, gitdiff_{old|new}name() will check
 876 * their names against any previous information, just
 877 * to make sure..
 878 */
 879static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
 880{
 881        if (!orig_name && !isnull)
 882                return find_name(line, NULL, p_value, TERM_TAB);
 883
 884        if (orig_name) {
 885                int len;
 886                const char *name;
 887                char *another;
 888                name = orig_name;
 889                len = strlen(name);
 890                if (isnull)
 891                        die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
 892                another = find_name(line, NULL, p_value, TERM_TAB);
 893                if (!another || memcmp(another, name, len + 1))
 894                        die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
 895                free(another);
 896                return orig_name;
 897        }
 898        else {
 899                /* expect "/dev/null" */
 900                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 901                        die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
 902                return NULL;
 903        }
 904}
 905
 906static int gitdiff_oldname(const char *line, struct patch *patch)
 907{
 908        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
 909        return 0;
 910}
 911
 912static int gitdiff_newname(const char *line, struct patch *patch)
 913{
 914        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
 915        return 0;
 916}
 917
 918static int gitdiff_oldmode(const char *line, struct patch *patch)
 919{
 920        patch->old_mode = strtoul(line, NULL, 8);
 921        return 0;
 922}
 923
 924static int gitdiff_newmode(const char *line, struct patch *patch)
 925{
 926        patch->new_mode = strtoul(line, NULL, 8);
 927        return 0;
 928}
 929
 930static int gitdiff_delete(const char *line, struct patch *patch)
 931{
 932        patch->is_delete = 1;
 933        patch->old_name = patch->def_name;
 934        return gitdiff_oldmode(line, patch);
 935}
 936
 937static int gitdiff_newfile(const char *line, struct patch *patch)
 938{
 939        patch->is_new = 1;
 940        patch->new_name = patch->def_name;
 941        return gitdiff_newmode(line, patch);
 942}
 943
 944static int gitdiff_copysrc(const char *line, struct patch *patch)
 945{
 946        patch->is_copy = 1;
 947        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
 948        return 0;
 949}
 950
 951static int gitdiff_copydst(const char *line, struct patch *patch)
 952{
 953        patch->is_copy = 1;
 954        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
 955        return 0;
 956}
 957
 958static int gitdiff_renamesrc(const char *line, struct patch *patch)
 959{
 960        patch->is_rename = 1;
 961        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
 962        return 0;
 963}
 964
 965static int gitdiff_renamedst(const char *line, struct patch *patch)
 966{
 967        patch->is_rename = 1;
 968        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
 969        return 0;
 970}
 971
 972static int gitdiff_similarity(const char *line, struct patch *patch)
 973{
 974        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 975                patch->score = 0;
 976        return 0;
 977}
 978
 979static int gitdiff_dissimilarity(const char *line, struct patch *patch)
 980{
 981        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 982                patch->score = 0;
 983        return 0;
 984}
 985
 986static int gitdiff_index(const char *line, struct patch *patch)
 987{
 988        /*
 989         * index line is N hexadecimal, "..", N hexadecimal,
 990         * and optional space with octal mode.
 991         */
 992        const char *ptr, *eol;
 993        int len;
 994
 995        ptr = strchr(line, '.');
 996        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
 997                return 0;
 998        len = ptr - line;
 999        memcpy(patch->old_sha1_prefix, line, len);
1000        patch->old_sha1_prefix[len] = 0;
1001
1002        line = ptr + 2;
1003        ptr = strchr(line, ' ');
1004        eol = strchr(line, '\n');
1005
1006        if (!ptr || eol < ptr)
1007                ptr = eol;
1008        len = ptr - line;
1009
1010        if (40 < len)
1011                return 0;
1012        memcpy(patch->new_sha1_prefix, line, len);
1013        patch->new_sha1_prefix[len] = 0;
1014        if (*ptr == ' ')
1015                patch->old_mode = strtoul(ptr+1, NULL, 8);
1016        return 0;
1017}
1018
1019/*
1020 * This is normal for a diff that doesn't change anything: we'll fall through
1021 * into the next diff. Tell the parser to break out.
1022 */
1023static int gitdiff_unrecognized(const char *line, struct patch *patch)
1024{
1025        return -1;
1026}
1027
1028static const char *stop_at_slash(const char *line, int llen)
1029{
1030        int nslash = p_value;
1031        int i;
1032
1033        for (i = 0; i < llen; i++) {
1034                int ch = line[i];
1035                if (ch == '/' && --nslash <= 0)
1036                        return &line[i];
1037        }
1038        return NULL;
1039}
1040
1041/*
1042 * This is to extract the same name that appears on "diff --git"
1043 * line.  We do not find and return anything if it is a rename
1044 * patch, and it is OK because we will find the name elsewhere.
1045 * We need to reliably find name only when it is mode-change only,
1046 * creation or deletion of an empty file.  In any of these cases,
1047 * both sides are the same name under a/ and b/ respectively.
1048 */
1049static char *git_header_name(char *line, int llen)
1050{
1051        const char *name;
1052        const char *second = NULL;
1053        size_t len, line_len;
1054
1055        line += strlen("diff --git ");
1056        llen -= strlen("diff --git ");
1057
1058        if (*line == '"') {
1059                const char *cp;
1060                struct strbuf first = STRBUF_INIT;
1061                struct strbuf sp = STRBUF_INIT;
1062
1063                if (unquote_c_style(&first, line, &second))
1064                        goto free_and_fail1;
1065
1066                /* advance to the first slash */
1067                cp = stop_at_slash(first.buf, first.len);
1068                /* we do not accept absolute paths */
1069                if (!cp || cp == first.buf)
1070                        goto free_and_fail1;
1071                strbuf_remove(&first, 0, cp + 1 - first.buf);
1072
1073                /*
1074                 * second points at one past closing dq of name.
1075                 * find the second name.
1076                 */
1077                while ((second < line + llen) && isspace(*second))
1078                        second++;
1079
1080                if (line + llen <= second)
1081                        goto free_and_fail1;
1082                if (*second == '"') {
1083                        if (unquote_c_style(&sp, second, NULL))
1084                                goto free_and_fail1;
1085                        cp = stop_at_slash(sp.buf, sp.len);
1086                        if (!cp || cp == sp.buf)
1087                                goto free_and_fail1;
1088                        /* They must match, otherwise ignore */
1089                        if (strcmp(cp + 1, first.buf))
1090                                goto free_and_fail1;
1091                        strbuf_release(&sp);
1092                        return strbuf_detach(&first, NULL);
1093                }
1094
1095                /* unquoted second */
1096                cp = stop_at_slash(second, line + llen - second);
1097                if (!cp || cp == second)
1098                        goto free_and_fail1;
1099                cp++;
1100                if (line + llen - cp != first.len + 1 ||
1101                    memcmp(first.buf, cp, first.len))
1102                        goto free_and_fail1;
1103                return strbuf_detach(&first, NULL);
1104
1105        free_and_fail1:
1106                strbuf_release(&first);
1107                strbuf_release(&sp);
1108                return NULL;
1109        }
1110
1111        /* unquoted first name */
1112        name = stop_at_slash(line, llen);
1113        if (!name || name == line)
1114                return NULL;
1115        name++;
1116
1117        /*
1118         * since the first name is unquoted, a dq if exists must be
1119         * the beginning of the second name.
1120         */
1121        for (second = name; second < line + llen; second++) {
1122                if (*second == '"') {
1123                        struct strbuf sp = STRBUF_INIT;
1124                        const char *np;
1125
1126                        if (unquote_c_style(&sp, second, NULL))
1127                                goto free_and_fail2;
1128
1129                        np = stop_at_slash(sp.buf, sp.len);
1130                        if (!np || np == sp.buf)
1131                                goto free_and_fail2;
1132                        np++;
1133
1134                        len = sp.buf + sp.len - np;
1135                        if (len < second - name &&
1136                            !strncmp(np, name, len) &&
1137                            isspace(name[len])) {
1138                                /* Good */
1139                                strbuf_remove(&sp, 0, np - sp.buf);
1140                                return strbuf_detach(&sp, NULL);
1141                        }
1142
1143                free_and_fail2:
1144                        strbuf_release(&sp);
1145                        return NULL;
1146                }
1147        }
1148
1149        /*
1150         * Accept a name only if it shows up twice, exactly the same
1151         * form.
1152         */
1153        second = strchr(name, '\n');
1154        if (!second)
1155                return NULL;
1156        line_len = second - name;
1157        for (len = 0 ; ; len++) {
1158                switch (name[len]) {
1159                default:
1160                        continue;
1161                case '\n':
1162                        return NULL;
1163                case '\t': case ' ':
1164                        second = stop_at_slash(name + len, line_len - len);
1165                        if (!second)
1166                                return NULL;
1167                        second++;
1168                        if (second[len] == '\n' && !strncmp(name, second, len)) {
1169                                return xmemdupz(name, len);
1170                        }
1171                }
1172        }
1173}
1174
1175/* Verify that we recognize the lines following a git header */
1176static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
1177{
1178        unsigned long offset;
1179
1180        /* A git diff has explicit new/delete information, so we don't guess */
1181        patch->is_new = 0;
1182        patch->is_delete = 0;
1183
1184        /*
1185         * Some things may not have the old name in the
1186         * rest of the headers anywhere (pure mode changes,
1187         * or removing or adding empty files), so we get
1188         * the default name from the header.
1189         */
1190        patch->def_name = git_header_name(line, len);
1191        if (patch->def_name && root) {
1192                char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1193                strcpy(s, root);
1194                strcpy(s + root_len, patch->def_name);
1195                free(patch->def_name);
1196                patch->def_name = s;
1197        }
1198
1199        line += len;
1200        size -= len;
1201        linenr++;
1202        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1203                static const struct opentry {
1204                        const char *str;
1205                        int (*fn)(const char *, struct patch *);
1206                } optable[] = {
1207                        { "@@ -", gitdiff_hdrend },
1208                        { "--- ", gitdiff_oldname },
1209                        { "+++ ", gitdiff_newname },
1210                        { "old mode ", gitdiff_oldmode },
1211                        { "new mode ", gitdiff_newmode },
1212                        { "deleted file mode ", gitdiff_delete },
1213                        { "new file mode ", gitdiff_newfile },
1214                        { "copy from ", gitdiff_copysrc },
1215                        { "copy to ", gitdiff_copydst },
1216                        { "rename old ", gitdiff_renamesrc },
1217                        { "rename new ", gitdiff_renamedst },
1218                        { "rename from ", gitdiff_renamesrc },
1219                        { "rename to ", gitdiff_renamedst },
1220                        { "similarity index ", gitdiff_similarity },
1221                        { "dissimilarity index ", gitdiff_dissimilarity },
1222                        { "index ", gitdiff_index },
1223                        { "", gitdiff_unrecognized },
1224                };
1225                int i;
1226
1227                len = linelen(line, size);
1228                if (!len || line[len-1] != '\n')
1229                        break;
1230                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1231                        const struct opentry *p = optable + i;
1232                        int oplen = strlen(p->str);
1233                        if (len < oplen || memcmp(p->str, line, oplen))
1234                                continue;
1235                        if (p->fn(line + oplen, patch) < 0)
1236                                return offset;
1237                        break;
1238                }
1239        }
1240
1241        return offset;
1242}
1243
1244static int parse_num(const char *line, unsigned long *p)
1245{
1246        char *ptr;
1247
1248        if (!isdigit(*line))
1249                return 0;
1250        *p = strtoul(line, &ptr, 10);
1251        return ptr - line;
1252}
1253
1254static int parse_range(const char *line, int len, int offset, const char *expect,
1255                       unsigned long *p1, unsigned long *p2)
1256{
1257        int digits, ex;
1258
1259        if (offset < 0 || offset >= len)
1260                return -1;
1261        line += offset;
1262        len -= offset;
1263
1264        digits = parse_num(line, p1);
1265        if (!digits)
1266                return -1;
1267
1268        offset += digits;
1269        line += digits;
1270        len -= digits;
1271
1272        *p2 = 1;
1273        if (*line == ',') {
1274                digits = parse_num(line+1, p2);
1275                if (!digits)
1276                        return -1;
1277
1278                offset += digits+1;
1279                line += digits+1;
1280                len -= digits+1;
1281        }
1282
1283        ex = strlen(expect);
1284        if (ex > len)
1285                return -1;
1286        if (memcmp(line, expect, ex))
1287                return -1;
1288
1289        return offset + ex;
1290}
1291
1292static void recount_diff(char *line, int size, struct fragment *fragment)
1293{
1294        int oldlines = 0, newlines = 0, ret = 0;
1295
1296        if (size < 1) {
1297                warning("recount: ignore empty hunk");
1298                return;
1299        }
1300
1301        for (;;) {
1302                int len = linelen(line, size);
1303                size -= len;
1304                line += len;
1305
1306                if (size < 1)
1307                        break;
1308
1309                switch (*line) {
1310                case ' ': case '\n':
1311                        newlines++;
1312                        /* fall through */
1313                case '-':
1314                        oldlines++;
1315                        continue;
1316                case '+':
1317                        newlines++;
1318                        continue;
1319                case '\\':
1320                        continue;
1321                case '@':
1322                        ret = size < 3 || prefixcmp(line, "@@ ");
1323                        break;
1324                case 'd':
1325                        ret = size < 5 || prefixcmp(line, "diff ");
1326                        break;
1327                default:
1328                        ret = -1;
1329                        break;
1330                }
1331                if (ret) {
1332                        warning("recount: unexpected line: %.*s",
1333                                (int)linelen(line, size), line);
1334                        return;
1335                }
1336                break;
1337        }
1338        fragment->oldlines = oldlines;
1339        fragment->newlines = newlines;
1340}
1341
1342/*
1343 * Parse a unified diff fragment header of the
1344 * form "@@ -a,b +c,d @@"
1345 */
1346static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1347{
1348        int offset;
1349
1350        if (!len || line[len-1] != '\n')
1351                return -1;
1352
1353        /* Figure out the number of lines in a fragment */
1354        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1355        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1356
1357        return offset;
1358}
1359
1360static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1361{
1362        unsigned long offset, len;
1363
1364        patch->is_toplevel_relative = 0;
1365        patch->is_rename = patch->is_copy = 0;
1366        patch->is_new = patch->is_delete = -1;
1367        patch->old_mode = patch->new_mode = 0;
1368        patch->old_name = patch->new_name = NULL;
1369        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1370                unsigned long nextlen;
1371
1372                len = linelen(line, size);
1373                if (!len)
1374                        break;
1375
1376                /* Testing this early allows us to take a few shortcuts.. */
1377                if (len < 6)
1378                        continue;
1379
1380                /*
1381                 * Make sure we don't find any unconnected patch fragments.
1382                 * That's a sign that we didn't find a header, and that a
1383                 * patch has become corrupted/broken up.
1384                 */
1385                if (!memcmp("@@ -", line, 4)) {
1386                        struct fragment dummy;
1387                        if (parse_fragment_header(line, len, &dummy) < 0)
1388                                continue;
1389                        die("patch fragment without header at line %d: %.*s",
1390                            linenr, (int)len-1, line);
1391                }
1392
1393                if (size < len + 6)
1394                        break;
1395
1396                /*
1397                 * Git patch? It might not have a real patch, just a rename
1398                 * or mode change, so we handle that specially
1399                 */
1400                if (!memcmp("diff --git ", line, 11)) {
1401                        int git_hdr_len = parse_git_header(line, len, size, patch);
1402                        if (git_hdr_len <= len)
1403                                continue;
1404                        if (!patch->old_name && !patch->new_name) {
1405                                if (!patch->def_name)
1406                                        die("git diff header lacks filename information when removing "
1407                                            "%d leading pathname components (line %d)" , p_value, linenr);
1408                                patch->old_name = patch->new_name = patch->def_name;
1409                        }
1410                        patch->is_toplevel_relative = 1;
1411                        *hdrsize = git_hdr_len;
1412                        return offset;
1413                }
1414
1415                /* --- followed by +++ ? */
1416                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1417                        continue;
1418
1419                /*
1420                 * We only accept unified patches, so we want it to
1421                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1422                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1423                 */
1424                nextlen = linelen(line + len, size - len);
1425                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1426                        continue;
1427
1428                /* Ok, we'll consider it a patch */
1429                parse_traditional_patch(line, line+len, patch);
1430                *hdrsize = len + nextlen;
1431                linenr += 2;
1432                return offset;
1433        }
1434        return -1;
1435}
1436
1437static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1438{
1439        char *err;
1440
1441        if (!result)
1442                return;
1443
1444        whitespace_error++;
1445        if (squelch_whitespace_errors &&
1446            squelch_whitespace_errors < whitespace_error)
1447                return;
1448
1449        err = whitespace_error_string(result);
1450        fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1451                patch_input_file, linenr, err, len, line);
1452        free(err);
1453}
1454
1455static void check_whitespace(const char *line, int len, unsigned ws_rule)
1456{
1457        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1458
1459        record_ws_error(result, line + 1, len - 2, linenr);
1460}
1461
1462/*
1463 * Parse a unified diff. Note that this really needs to parse each
1464 * fragment separately, since the only way to know the difference
1465 * between a "---" that is part of a patch, and a "---" that starts
1466 * the next patch is to look at the line counts..
1467 */
1468static int parse_fragment(char *line, unsigned long size,
1469                          struct patch *patch, struct fragment *fragment)
1470{
1471        int added, deleted;
1472        int len = linelen(line, size), offset;
1473        unsigned long oldlines, newlines;
1474        unsigned long leading, trailing;
1475
1476        offset = parse_fragment_header(line, len, fragment);
1477        if (offset < 0)
1478                return -1;
1479        if (offset > 0 && patch->recount)
1480                recount_diff(line + offset, size - offset, fragment);
1481        oldlines = fragment->oldlines;
1482        newlines = fragment->newlines;
1483        leading = 0;
1484        trailing = 0;
1485
1486        /* Parse the thing.. */
1487        line += len;
1488        size -= len;
1489        linenr++;
1490        added = deleted = 0;
1491        for (offset = len;
1492             0 < size;
1493             offset += len, size -= len, line += len, linenr++) {
1494                if (!oldlines && !newlines)
1495                        break;
1496                len = linelen(line, size);
1497                if (!len || line[len-1] != '\n')
1498                        return -1;
1499                switch (*line) {
1500                default:
1501                        return -1;
1502                case '\n': /* newer GNU diff, an empty context line */
1503                case ' ':
1504                        oldlines--;
1505                        newlines--;
1506                        if (!deleted && !added)
1507                                leading++;
1508                        trailing++;
1509                        break;
1510                case '-':
1511                        if (apply_in_reverse &&
1512                            ws_error_action != nowarn_ws_error)
1513                                check_whitespace(line, len, patch->ws_rule);
1514                        deleted++;
1515                        oldlines--;
1516                        trailing = 0;
1517                        break;
1518                case '+':
1519                        if (!apply_in_reverse &&
1520                            ws_error_action != nowarn_ws_error)
1521                                check_whitespace(line, len, patch->ws_rule);
1522                        added++;
1523                        newlines--;
1524                        trailing = 0;
1525                        break;
1526
1527                /*
1528                 * We allow "\ No newline at end of file". Depending
1529                 * on locale settings when the patch was produced we
1530                 * don't know what this line looks like. The only
1531                 * thing we do know is that it begins with "\ ".
1532                 * Checking for 12 is just for sanity check -- any
1533                 * l10n of "\ No newline..." is at least that long.
1534                 */
1535                case '\\':
1536                        if (len < 12 || memcmp(line, "\\ ", 2))
1537                                return -1;
1538                        break;
1539                }
1540        }
1541        if (oldlines || newlines)
1542                return -1;
1543        fragment->leading = leading;
1544        fragment->trailing = trailing;
1545
1546        /*
1547         * If a fragment ends with an incomplete line, we failed to include
1548         * it in the above loop because we hit oldlines == newlines == 0
1549         * before seeing it.
1550         */
1551        if (12 < size && !memcmp(line, "\\ ", 2))
1552                offset += linelen(line, size);
1553
1554        patch->lines_added += added;
1555        patch->lines_deleted += deleted;
1556
1557        if (0 < patch->is_new && oldlines)
1558                return error("new file depends on old contents");
1559        if (0 < patch->is_delete && newlines)
1560                return error("deleted file still has contents");
1561        return offset;
1562}
1563
1564static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1565{
1566        unsigned long offset = 0;
1567        unsigned long oldlines = 0, newlines = 0, context = 0;
1568        struct fragment **fragp = &patch->fragments;
1569
1570        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1571                struct fragment *fragment;
1572                int len;
1573
1574                fragment = xcalloc(1, sizeof(*fragment));
1575                fragment->linenr = linenr;
1576                len = parse_fragment(line, size, patch, fragment);
1577                if (len <= 0)
1578                        die("corrupt patch at line %d", linenr);
1579                fragment->patch = line;
1580                fragment->size = len;
1581                oldlines += fragment->oldlines;
1582                newlines += fragment->newlines;
1583                context += fragment->leading + fragment->trailing;
1584
1585                *fragp = fragment;
1586                fragp = &fragment->next;
1587
1588                offset += len;
1589                line += len;
1590                size -= len;
1591        }
1592
1593        /*
1594         * If something was removed (i.e. we have old-lines) it cannot
1595         * be creation, and if something was added it cannot be
1596         * deletion.  However, the reverse is not true; --unified=0
1597         * patches that only add are not necessarily creation even
1598         * though they do not have any old lines, and ones that only
1599         * delete are not necessarily deletion.
1600         *
1601         * Unfortunately, a real creation/deletion patch do _not_ have
1602         * any context line by definition, so we cannot safely tell it
1603         * apart with --unified=0 insanity.  At least if the patch has
1604         * more than one hunk it is not creation or deletion.
1605         */
1606        if (patch->is_new < 0 &&
1607            (oldlines || (patch->fragments && patch->fragments->next)))
1608                patch->is_new = 0;
1609        if (patch->is_delete < 0 &&
1610            (newlines || (patch->fragments && patch->fragments->next)))
1611                patch->is_delete = 0;
1612
1613        if (0 < patch->is_new && oldlines)
1614                die("new file %s depends on old contents", patch->new_name);
1615        if (0 < patch->is_delete && newlines)
1616                die("deleted file %s still has contents", patch->old_name);
1617        if (!patch->is_delete && !newlines && context)
1618                fprintf(stderr, "** warning: file %s becomes empty but "
1619                        "is not deleted\n", patch->new_name);
1620
1621        return offset;
1622}
1623
1624static inline int metadata_changes(struct patch *patch)
1625{
1626        return  patch->is_rename > 0 ||
1627                patch->is_copy > 0 ||
1628                patch->is_new > 0 ||
1629                patch->is_delete ||
1630                (patch->old_mode && patch->new_mode &&
1631                 patch->old_mode != patch->new_mode);
1632}
1633
1634static char *inflate_it(const void *data, unsigned long size,
1635                        unsigned long inflated_size)
1636{
1637        z_stream stream;
1638        void *out;
1639        int st;
1640
1641        memset(&stream, 0, sizeof(stream));
1642
1643        stream.next_in = (unsigned char *)data;
1644        stream.avail_in = size;
1645        stream.next_out = out = xmalloc(inflated_size);
1646        stream.avail_out = inflated_size;
1647        git_inflate_init(&stream);
1648        st = git_inflate(&stream, Z_FINISH);
1649        git_inflate_end(&stream);
1650        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1651                free(out);
1652                return NULL;
1653        }
1654        return out;
1655}
1656
1657static struct fragment *parse_binary_hunk(char **buf_p,
1658                                          unsigned long *sz_p,
1659                                          int *status_p,
1660                                          int *used_p)
1661{
1662        /*
1663         * Expect a line that begins with binary patch method ("literal"
1664         * or "delta"), followed by the length of data before deflating.
1665         * a sequence of 'length-byte' followed by base-85 encoded data
1666         * should follow, terminated by a newline.
1667         *
1668         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1669         * and we would limit the patch line to 66 characters,
1670         * so one line can fit up to 13 groups that would decode
1671         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1672         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1673         */
1674        int llen, used;
1675        unsigned long size = *sz_p;
1676        char *buffer = *buf_p;
1677        int patch_method;
1678        unsigned long origlen;
1679        char *data = NULL;
1680        int hunk_size = 0;
1681        struct fragment *frag;
1682
1683        llen = linelen(buffer, size);
1684        used = llen;
1685
1686        *status_p = 0;
1687
1688        if (!prefixcmp(buffer, "delta ")) {
1689                patch_method = BINARY_DELTA_DEFLATED;
1690                origlen = strtoul(buffer + 6, NULL, 10);
1691        }
1692        else if (!prefixcmp(buffer, "literal ")) {
1693                patch_method = BINARY_LITERAL_DEFLATED;
1694                origlen = strtoul(buffer + 8, NULL, 10);
1695        }
1696        else
1697                return NULL;
1698
1699        linenr++;
1700        buffer += llen;
1701        while (1) {
1702                int byte_length, max_byte_length, newsize;
1703                llen = linelen(buffer, size);
1704                used += llen;
1705                linenr++;
1706                if (llen == 1) {
1707                        /* consume the blank line */
1708                        buffer++;
1709                        size--;
1710                        break;
1711                }
1712                /*
1713                 * Minimum line is "A00000\n" which is 7-byte long,
1714                 * and the line length must be multiple of 5 plus 2.
1715                 */
1716                if ((llen < 7) || (llen-2) % 5)
1717                        goto corrupt;
1718                max_byte_length = (llen - 2) / 5 * 4;
1719                byte_length = *buffer;
1720                if ('A' <= byte_length && byte_length <= 'Z')
1721                        byte_length = byte_length - 'A' + 1;
1722                else if ('a' <= byte_length && byte_length <= 'z')
1723                        byte_length = byte_length - 'a' + 27;
1724                else
1725                        goto corrupt;
1726                /* if the input length was not multiple of 4, we would
1727                 * have filler at the end but the filler should never
1728                 * exceed 3 bytes
1729                 */
1730                if (max_byte_length < byte_length ||
1731                    byte_length <= max_byte_length - 4)
1732                        goto corrupt;
1733                newsize = hunk_size + byte_length;
1734                data = xrealloc(data, newsize);
1735                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1736                        goto corrupt;
1737                hunk_size = newsize;
1738                buffer += llen;
1739                size -= llen;
1740        }
1741
1742        frag = xcalloc(1, sizeof(*frag));
1743        frag->patch = inflate_it(data, hunk_size, origlen);
1744        if (!frag->patch)
1745                goto corrupt;
1746        free(data);
1747        frag->size = origlen;
1748        *buf_p = buffer;
1749        *sz_p = size;
1750        *used_p = used;
1751        frag->binary_patch_method = patch_method;
1752        return frag;
1753
1754 corrupt:
1755        free(data);
1756        *status_p = -1;
1757        error("corrupt binary patch at line %d: %.*s",
1758              linenr-1, llen-1, buffer);
1759        return NULL;
1760}
1761
1762static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1763{
1764        /*
1765         * We have read "GIT binary patch\n"; what follows is a line
1766         * that says the patch method (currently, either "literal" or
1767         * "delta") and the length of data before deflating; a
1768         * sequence of 'length-byte' followed by base-85 encoded data
1769         * follows.
1770         *
1771         * When a binary patch is reversible, there is another binary
1772         * hunk in the same format, starting with patch method (either
1773         * "literal" or "delta") with the length of data, and a sequence
1774         * of length-byte + base-85 encoded data, terminated with another
1775         * empty line.  This data, when applied to the postimage, produces
1776         * the preimage.
1777         */
1778        struct fragment *forward;
1779        struct fragment *reverse;
1780        int status;
1781        int used, used_1;
1782
1783        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1784        if (!forward && !status)
1785                /* there has to be one hunk (forward hunk) */
1786                return error("unrecognized binary patch at line %d", linenr-1);
1787        if (status)
1788                /* otherwise we already gave an error message */
1789                return status;
1790
1791        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1792        if (reverse)
1793                used += used_1;
1794        else if (status) {
1795                /*
1796                 * Not having reverse hunk is not an error, but having
1797                 * a corrupt reverse hunk is.
1798                 */
1799                free((void*) forward->patch);
1800                free(forward);
1801                return status;
1802        }
1803        forward->next = reverse;
1804        patch->fragments = forward;
1805        patch->is_binary = 1;
1806        return used;
1807}
1808
1809static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1810{
1811        int hdrsize, patchsize;
1812        int offset = find_header(buffer, size, &hdrsize, patch);
1813
1814        if (offset < 0)
1815                return offset;
1816
1817        patch->ws_rule = whitespace_rule(patch->new_name
1818                                         ? patch->new_name
1819                                         : patch->old_name);
1820
1821        patchsize = parse_single_patch(buffer + offset + hdrsize,
1822                                       size - offset - hdrsize, patch);
1823
1824        if (!patchsize) {
1825                static const char *binhdr[] = {
1826                        "Binary files ",
1827                        "Files ",
1828                        NULL,
1829                };
1830                static const char git_binary[] = "GIT binary patch\n";
1831                int i;
1832                int hd = hdrsize + offset;
1833                unsigned long llen = linelen(buffer + hd, size - hd);
1834
1835                if (llen == sizeof(git_binary) - 1 &&
1836                    !memcmp(git_binary, buffer + hd, llen)) {
1837                        int used;
1838                        linenr++;
1839                        used = parse_binary(buffer + hd + llen,
1840                                            size - hd - llen, patch);
1841                        if (used)
1842                                patchsize = used + llen;
1843                        else
1844                                patchsize = 0;
1845                }
1846                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1847                        for (i = 0; binhdr[i]; i++) {
1848                                int len = strlen(binhdr[i]);
1849                                if (len < size - hd &&
1850                                    !memcmp(binhdr[i], buffer + hd, len)) {
1851                                        linenr++;
1852                                        patch->is_binary = 1;
1853                                        patchsize = llen;
1854                                        break;
1855                                }
1856                        }
1857                }
1858
1859                /* Empty patch cannot be applied if it is a text patch
1860                 * without metadata change.  A binary patch appears
1861                 * empty to us here.
1862                 */
1863                if ((apply || check) &&
1864                    (!patch->is_binary && !metadata_changes(patch)))
1865                        die("patch with only garbage at line %d", linenr);
1866        }
1867
1868        return offset + hdrsize + patchsize;
1869}
1870
1871#define swap(a,b) myswap((a),(b),sizeof(a))
1872
1873#define myswap(a, b, size) do {         \
1874        unsigned char mytmp[size];      \
1875        memcpy(mytmp, &a, size);                \
1876        memcpy(&a, &b, size);           \
1877        memcpy(&b, mytmp, size);                \
1878} while (0)
1879
1880static void reverse_patches(struct patch *p)
1881{
1882        for (; p; p = p->next) {
1883                struct fragment *frag = p->fragments;
1884
1885                swap(p->new_name, p->old_name);
1886                swap(p->new_mode, p->old_mode);
1887                swap(p->is_new, p->is_delete);
1888                swap(p->lines_added, p->lines_deleted);
1889                swap(p->old_sha1_prefix, p->new_sha1_prefix);
1890
1891                for (; frag; frag = frag->next) {
1892                        swap(frag->newpos, frag->oldpos);
1893                        swap(frag->newlines, frag->oldlines);
1894                }
1895        }
1896}
1897
1898static const char pluses[] =
1899"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1900static const char minuses[]=
1901"----------------------------------------------------------------------";
1902
1903static void show_stats(struct patch *patch)
1904{
1905        struct strbuf qname = STRBUF_INIT;
1906        char *cp = patch->new_name ? patch->new_name : patch->old_name;
1907        int max, add, del;
1908
1909        quote_c_style(cp, &qname, NULL, 0);
1910
1911        /*
1912         * "scale" the filename
1913         */
1914        max = max_len;
1915        if (max > 50)
1916                max = 50;
1917
1918        if (qname.len > max) {
1919                cp = strchr(qname.buf + qname.len + 3 - max, '/');
1920                if (!cp)
1921                        cp = qname.buf + qname.len + 3 - max;
1922                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1923        }
1924
1925        if (patch->is_binary) {
1926                printf(" %-*s |  Bin\n", max, qname.buf);
1927                strbuf_release(&qname);
1928                return;
1929        }
1930
1931        printf(" %-*s |", max, qname.buf);
1932        strbuf_release(&qname);
1933
1934        /*
1935         * scale the add/delete
1936         */
1937        max = max + max_change > 70 ? 70 - max : max_change;
1938        add = patch->lines_added;
1939        del = patch->lines_deleted;
1940
1941        if (max_change > 0) {
1942                int total = ((add + del) * max + max_change / 2) / max_change;
1943                add = (add * max + max_change / 2) / max_change;
1944                del = total - add;
1945        }
1946        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1947                add, pluses, del, minuses);
1948}
1949
1950static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1951{
1952        switch (st->st_mode & S_IFMT) {
1953        case S_IFLNK:
1954                if (strbuf_readlink(buf, path, st->st_size) < 0)
1955                        return error("unable to read symlink %s", path);
1956                return 0;
1957        case S_IFREG:
1958                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1959                        return error("unable to open or read %s", path);
1960                convert_to_git(path, buf->buf, buf->len, buf, 0);
1961                return 0;
1962        default:
1963                return -1;
1964        }
1965}
1966
1967/*
1968 * Update the preimage, and the common lines in postimage,
1969 * from buffer buf of length len. If postlen is 0 the postimage
1970 * is updated in place, otherwise it's updated on a new buffer
1971 * of length postlen
1972 */
1973
1974static void update_pre_post_images(struct image *preimage,
1975                                   struct image *postimage,
1976                                   char *buf,
1977                                   size_t len, size_t postlen)
1978{
1979        int i, ctx;
1980        char *new, *old, *fixed;
1981        struct image fixed_preimage;
1982
1983        /*
1984         * Update the preimage with whitespace fixes.  Note that we
1985         * are not losing preimage->buf -- apply_one_fragment() will
1986         * free "oldlines".
1987         */
1988        prepare_image(&fixed_preimage, buf, len, 1);
1989        assert(fixed_preimage.nr == preimage->nr);
1990        for (i = 0; i < preimage->nr; i++)
1991                fixed_preimage.line[i].flag = preimage->line[i].flag;
1992        free(preimage->line_allocated);
1993        *preimage = fixed_preimage;
1994
1995        /*
1996         * Adjust the common context lines in postimage. This can be
1997         * done in-place when we are just doing whitespace fixing,
1998         * which does not make the string grow, but needs a new buffer
1999         * when ignoring whitespace causes the update, since in this case
2000         * we could have e.g. tabs converted to multiple spaces.
2001         * We trust the caller to tell us if the update can be done
2002         * in place (postlen==0) or not.
2003         */
2004        old = postimage->buf;
2005        if (postlen)
2006                new = postimage->buf = xmalloc(postlen);
2007        else
2008                new = old;
2009        fixed = preimage->buf;
2010        for (i = ctx = 0; i < postimage->nr; i++) {
2011                size_t len = postimage->line[i].len;
2012                if (!(postimage->line[i].flag & LINE_COMMON)) {
2013                        /* an added line -- no counterparts in preimage */
2014                        memmove(new, old, len);
2015                        old += len;
2016                        new += len;
2017                        continue;
2018                }
2019
2020                /* a common context -- skip it in the original postimage */
2021                old += len;
2022
2023                /* and find the corresponding one in the fixed preimage */
2024                while (ctx < preimage->nr &&
2025                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2026                        fixed += preimage->line[ctx].len;
2027                        ctx++;
2028                }
2029                if (preimage->nr <= ctx)
2030                        die("oops");
2031
2032                /* and copy it in, while fixing the line length */
2033                len = preimage->line[ctx].len;
2034                memcpy(new, fixed, len);
2035                new += len;
2036                fixed += len;
2037                postimage->line[i].len = len;
2038                ctx++;
2039        }
2040
2041        /* Fix the length of the whole thing */
2042        postimage->len = new - postimage->buf;
2043}
2044
2045static int match_fragment(struct image *img,
2046                          struct image *preimage,
2047                          struct image *postimage,
2048                          unsigned long try,
2049                          int try_lno,
2050                          unsigned ws_rule,
2051                          int match_beginning, int match_end)
2052{
2053        int i;
2054        char *fixed_buf, *buf, *orig, *target;
2055        struct strbuf fixed;
2056        size_t fixed_len;
2057        int preimage_limit;
2058
2059        if (preimage->nr + try_lno <= img->nr) {
2060                /*
2061                 * The hunk falls within the boundaries of img.
2062                 */
2063                preimage_limit = preimage->nr;
2064                if (match_end && (preimage->nr + try_lno != img->nr))
2065                        return 0;
2066        } else if (ws_error_action == correct_ws_error &&
2067                   (ws_rule & WS_BLANK_AT_EOF)) {
2068                /*
2069                 * This hunk extends beyond the end of img, and we are
2070                 * removing blank lines at the end of the file.  This
2071                 * many lines from the beginning of the preimage must
2072                 * match with img, and the remainder of the preimage
2073                 * must be blank.
2074                 */
2075                preimage_limit = img->nr - try_lno;
2076        } else {
2077                /*
2078                 * The hunk extends beyond the end of the img and
2079                 * we are not removing blanks at the end, so we
2080                 * should reject the hunk at this position.
2081                 */
2082                return 0;
2083        }
2084
2085        if (match_beginning && try_lno)
2086                return 0;
2087
2088        /* Quick hash check */
2089        for (i = 0; i < preimage_limit; i++)
2090                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2091                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2092                        return 0;
2093
2094        if (preimage_limit == preimage->nr) {
2095                /*
2096                 * Do we have an exact match?  If we were told to match
2097                 * at the end, size must be exactly at try+fragsize,
2098                 * otherwise try+fragsize must be still within the preimage,
2099                 * and either case, the old piece should match the preimage
2100                 * exactly.
2101                 */
2102                if ((match_end
2103                     ? (try + preimage->len == img->len)
2104                     : (try + preimage->len <= img->len)) &&
2105                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2106                        return 1;
2107        } else {
2108                /*
2109                 * The preimage extends beyond the end of img, so
2110                 * there cannot be an exact match.
2111                 *
2112                 * There must be one non-blank context line that match
2113                 * a line before the end of img.
2114                 */
2115                char *buf_end;
2116
2117                buf = preimage->buf;
2118                buf_end = buf;
2119                for (i = 0; i < preimage_limit; i++)
2120                        buf_end += preimage->line[i].len;
2121
2122                for ( ; buf < buf_end; buf++)
2123                        if (!isspace(*buf))
2124                                break;
2125                if (buf == buf_end)
2126                        return 0;
2127        }
2128
2129        /*
2130         * No exact match. If we are ignoring whitespace, run a line-by-line
2131         * fuzzy matching. We collect all the line length information because
2132         * we need it to adjust whitespace if we match.
2133         */
2134        if (ws_ignore_action == ignore_ws_change) {
2135                size_t imgoff = 0;
2136                size_t preoff = 0;
2137                size_t postlen = postimage->len;
2138                size_t extra_chars;
2139                char *preimage_eof;
2140                char *preimage_end;
2141                for (i = 0; i < preimage_limit; i++) {
2142                        size_t prelen = preimage->line[i].len;
2143                        size_t imglen = img->line[try_lno+i].len;
2144
2145                        if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2146                                              preimage->buf + preoff, prelen))
2147                                return 0;
2148                        if (preimage->line[i].flag & LINE_COMMON)
2149                                postlen += imglen - prelen;
2150                        imgoff += imglen;
2151                        preoff += prelen;
2152                }
2153
2154                /*
2155                 * Ok, the preimage matches with whitespace fuzz.
2156                 *
2157                 * imgoff now holds the true length of the target that
2158                 * matches the preimage before the end of the file.
2159                 *
2160                 * Count the number of characters in the preimage that fall
2161                 * beyond the end of the file and make sure that all of them
2162                 * are whitespace characters. (This can only happen if
2163                 * we are removing blank lines at the end of the file.)
2164                 */
2165                buf = preimage_eof = preimage->buf + preoff;
2166                for ( ; i < preimage->nr; i++)
2167                        preoff += preimage->line[i].len;
2168                preimage_end = preimage->buf + preoff;
2169                for ( ; buf < preimage_end; buf++)
2170                        if (!isspace(*buf))
2171                                return 0;
2172
2173                /*
2174                 * Update the preimage and the common postimage context
2175                 * lines to use the same whitespace as the target.
2176                 * If whitespace is missing in the target (i.e.
2177                 * if the preimage extends beyond the end of the file),
2178                 * use the whitespace from the preimage.
2179                 */
2180                extra_chars = preimage_end - preimage_eof;
2181                strbuf_init(&fixed, imgoff + extra_chars);
2182                strbuf_add(&fixed, img->buf + try, imgoff);
2183                strbuf_add(&fixed, preimage_eof, extra_chars);
2184                fixed_buf = strbuf_detach(&fixed, &fixed_len);
2185                update_pre_post_images(preimage, postimage,
2186                                fixed_buf, fixed_len, postlen);
2187                return 1;
2188        }
2189
2190        if (ws_error_action != correct_ws_error)
2191                return 0;
2192
2193        /*
2194         * The hunk does not apply byte-by-byte, but the hash says
2195         * it might with whitespace fuzz. We haven't been asked to
2196         * ignore whitespace, we were asked to correct whitespace
2197         * errors, so let's try matching after whitespace correction.
2198         *
2199         * The preimage may extend beyond the end of the file,
2200         * but in this loop we will only handle the part of the
2201         * preimage that falls within the file.
2202         */
2203        strbuf_init(&fixed, preimage->len + 1);
2204        orig = preimage->buf;
2205        target = img->buf + try;
2206        for (i = 0; i < preimage_limit; i++) {
2207                size_t oldlen = preimage->line[i].len;
2208                size_t tgtlen = img->line[try_lno + i].len;
2209                size_t fixstart = fixed.len;
2210                struct strbuf tgtfix;
2211                int match;
2212
2213                /* Try fixing the line in the preimage */
2214                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2215
2216                /* Try fixing the line in the target */
2217                strbuf_init(&tgtfix, tgtlen);
2218                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2219
2220                /*
2221                 * If they match, either the preimage was based on
2222                 * a version before our tree fixed whitespace breakage,
2223                 * or we are lacking a whitespace-fix patch the tree
2224                 * the preimage was based on already had (i.e. target
2225                 * has whitespace breakage, the preimage doesn't).
2226                 * In either case, we are fixing the whitespace breakages
2227                 * so we might as well take the fix together with their
2228                 * real change.
2229                 */
2230                match = (tgtfix.len == fixed.len - fixstart &&
2231                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2232                                             fixed.len - fixstart));
2233
2234                strbuf_release(&tgtfix);
2235                if (!match)
2236                        goto unmatch_exit;
2237
2238                orig += oldlen;
2239                target += tgtlen;
2240        }
2241
2242
2243        /*
2244         * Now handle the lines in the preimage that falls beyond the
2245         * end of the file (if any). They will only match if they are
2246         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2247         * false).
2248         */
2249        for ( ; i < preimage->nr; i++) {
2250                size_t fixstart = fixed.len; /* start of the fixed preimage */
2251                size_t oldlen = preimage->line[i].len;
2252                int j;
2253
2254                /* Try fixing the line in the preimage */
2255                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2256
2257                for (j = fixstart; j < fixed.len; j++)
2258                        if (!isspace(fixed.buf[j]))
2259                                goto unmatch_exit;
2260
2261                orig += oldlen;
2262        }
2263
2264        /*
2265         * Yes, the preimage is based on an older version that still
2266         * has whitespace breakages unfixed, and fixing them makes the
2267         * hunk match.  Update the context lines in the postimage.
2268         */
2269        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2270        update_pre_post_images(preimage, postimage,
2271                               fixed_buf, fixed_len, 0);
2272        return 1;
2273
2274 unmatch_exit:
2275        strbuf_release(&fixed);
2276        return 0;
2277}
2278
2279static int find_pos(struct image *img,
2280                    struct image *preimage,
2281                    struct image *postimage,
2282                    int line,
2283                    unsigned ws_rule,
2284                    int match_beginning, int match_end)
2285{
2286        int i;
2287        unsigned long backwards, forwards, try;
2288        int backwards_lno, forwards_lno, try_lno;
2289
2290        /*
2291         * If match_beginning or match_end is specified, there is no
2292         * point starting from a wrong line that will never match and
2293         * wander around and wait for a match at the specified end.
2294         */
2295        if (match_beginning)
2296                line = 0;
2297        else if (match_end)
2298                line = img->nr - preimage->nr;
2299
2300        /*
2301         * Because the comparison is unsigned, the following test
2302         * will also take care of a negative line number that can
2303         * result when match_end and preimage is larger than the target.
2304         */
2305        if ((size_t) line > img->nr)
2306                line = img->nr;
2307
2308        try = 0;
2309        for (i = 0; i < line; i++)
2310                try += img->line[i].len;
2311
2312        /*
2313         * There's probably some smart way to do this, but I'll leave
2314         * that to the smart and beautiful people. I'm simple and stupid.
2315         */
2316        backwards = try;
2317        backwards_lno = line;
2318        forwards = try;
2319        forwards_lno = line;
2320        try_lno = line;
2321
2322        for (i = 0; ; i++) {
2323                if (match_fragment(img, preimage, postimage,
2324                                   try, try_lno, ws_rule,
2325                                   match_beginning, match_end))
2326                        return try_lno;
2327
2328        again:
2329                if (backwards_lno == 0 && forwards_lno == img->nr)
2330                        break;
2331
2332                if (i & 1) {
2333                        if (backwards_lno == 0) {
2334                                i++;
2335                                goto again;
2336                        }
2337                        backwards_lno--;
2338                        backwards -= img->line[backwards_lno].len;
2339                        try = backwards;
2340                        try_lno = backwards_lno;
2341                } else {
2342                        if (forwards_lno == img->nr) {
2343                                i++;
2344                                goto again;
2345                        }
2346                        forwards += img->line[forwards_lno].len;
2347                        forwards_lno++;
2348                        try = forwards;
2349                        try_lno = forwards_lno;
2350                }
2351
2352        }
2353        return -1;
2354}
2355
2356static void remove_first_line(struct image *img)
2357{
2358        img->buf += img->line[0].len;
2359        img->len -= img->line[0].len;
2360        img->line++;
2361        img->nr--;
2362}
2363
2364static void remove_last_line(struct image *img)
2365{
2366        img->len -= img->line[--img->nr].len;
2367}
2368
2369static void update_image(struct image *img,
2370                         int applied_pos,
2371                         struct image *preimage,
2372                         struct image *postimage)
2373{
2374        /*
2375         * remove the copy of preimage at offset in img
2376         * and replace it with postimage
2377         */
2378        int i, nr;
2379        size_t remove_count, insert_count, applied_at = 0;
2380        char *result;
2381        int preimage_limit;
2382
2383        /*
2384         * If we are removing blank lines at the end of img,
2385         * the preimage may extend beyond the end.
2386         * If that is the case, we must be careful only to
2387         * remove the part of the preimage that falls within
2388         * the boundaries of img. Initialize preimage_limit
2389         * to the number of lines in the preimage that falls
2390         * within the boundaries.
2391         */
2392        preimage_limit = preimage->nr;
2393        if (preimage_limit > img->nr - applied_pos)
2394                preimage_limit = img->nr - applied_pos;
2395
2396        for (i = 0; i < applied_pos; i++)
2397                applied_at += img->line[i].len;
2398
2399        remove_count = 0;
2400        for (i = 0; i < preimage_limit; i++)
2401                remove_count += img->line[applied_pos + i].len;
2402        insert_count = postimage->len;
2403
2404        /* Adjust the contents */
2405        result = xmalloc(img->len + insert_count - remove_count + 1);
2406        memcpy(result, img->buf, applied_at);
2407        memcpy(result + applied_at, postimage->buf, postimage->len);
2408        memcpy(result + applied_at + postimage->len,
2409               img->buf + (applied_at + remove_count),
2410               img->len - (applied_at + remove_count));
2411        free(img->buf);
2412        img->buf = result;
2413        img->len += insert_count - remove_count;
2414        result[img->len] = '\0';
2415
2416        /* Adjust the line table */
2417        nr = img->nr + postimage->nr - preimage_limit;
2418        if (preimage_limit < postimage->nr) {
2419                /*
2420                 * NOTE: this knows that we never call remove_first_line()
2421                 * on anything other than pre/post image.
2422                 */
2423                img->line = xrealloc(img->line, nr * sizeof(*img->line));
2424                img->line_allocated = img->line;
2425        }
2426        if (preimage_limit != postimage->nr)
2427                memmove(img->line + applied_pos + postimage->nr,
2428                        img->line + applied_pos + preimage_limit,
2429                        (img->nr - (applied_pos + preimage_limit)) *
2430                        sizeof(*img->line));
2431        memcpy(img->line + applied_pos,
2432               postimage->line,
2433               postimage->nr * sizeof(*img->line));
2434        if (!allow_overlap)
2435                for (i = 0; i < postimage->nr; i++)
2436                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2437        img->nr = nr;
2438}
2439
2440static int apply_one_fragment(struct image *img, struct fragment *frag,
2441                              int inaccurate_eof, unsigned ws_rule)
2442{
2443        int match_beginning, match_end;
2444        const char *patch = frag->patch;
2445        int size = frag->size;
2446        char *old, *oldlines;
2447        struct strbuf newlines;
2448        int new_blank_lines_at_end = 0;
2449        unsigned long leading, trailing;
2450        int pos, applied_pos;
2451        struct image preimage;
2452        struct image postimage;
2453
2454        memset(&preimage, 0, sizeof(preimage));
2455        memset(&postimage, 0, sizeof(postimage));
2456        oldlines = xmalloc(size);
2457        strbuf_init(&newlines, size);
2458
2459        old = oldlines;
2460        while (size > 0) {
2461                char first;
2462                int len = linelen(patch, size);
2463                int plen;
2464                int added_blank_line = 0;
2465                int is_blank_context = 0;
2466                size_t start;
2467
2468                if (!len)
2469                        break;
2470
2471                /*
2472                 * "plen" is how much of the line we should use for
2473                 * the actual patch data. Normally we just remove the
2474                 * first character on the line, but if the line is
2475                 * followed by "\ No newline", then we also remove the
2476                 * last one (which is the newline, of course).
2477                 */
2478                plen = len - 1;
2479                if (len < size && patch[len] == '\\')
2480                        plen--;
2481                first = *patch;
2482                if (apply_in_reverse) {
2483                        if (first == '-')
2484                                first = '+';
2485                        else if (first == '+')
2486                                first = '-';
2487                }
2488
2489                switch (first) {
2490                case '\n':
2491                        /* Newer GNU diff, empty context line */
2492                        if (plen < 0)
2493                                /* ... followed by '\No newline'; nothing */
2494                                break;
2495                        *old++ = '\n';
2496                        strbuf_addch(&newlines, '\n');
2497                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2498                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2499                        is_blank_context = 1;
2500                        break;
2501                case ' ':
2502                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2503                            ws_blank_line(patch + 1, plen, ws_rule))
2504                                is_blank_context = 1;
2505                case '-':
2506                        memcpy(old, patch + 1, plen);
2507                        add_line_info(&preimage, old, plen,
2508                                      (first == ' ' ? LINE_COMMON : 0));
2509                        old += plen;
2510                        if (first == '-')
2511                                break;
2512                /* Fall-through for ' ' */
2513                case '+':
2514                        /* --no-add does not add new lines */
2515                        if (first == '+' && no_add)
2516                                break;
2517
2518                        start = newlines.len;
2519                        if (first != '+' ||
2520                            !whitespace_error ||
2521                            ws_error_action != correct_ws_error) {
2522                                strbuf_add(&newlines, patch + 1, plen);
2523                        }
2524                        else {
2525                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2526                        }
2527                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2528                                      (first == '+' ? 0 : LINE_COMMON));
2529                        if (first == '+' &&
2530                            (ws_rule & WS_BLANK_AT_EOF) &&
2531                            ws_blank_line(patch + 1, plen, ws_rule))
2532                                added_blank_line = 1;
2533                        break;
2534                case '@': case '\\':
2535                        /* Ignore it, we already handled it */
2536                        break;
2537                default:
2538                        if (apply_verbosely)
2539                                error("invalid start of line: '%c'", first);
2540                        return -1;
2541                }
2542                if (added_blank_line)
2543                        new_blank_lines_at_end++;
2544                else if (is_blank_context)
2545                        ;
2546                else
2547                        new_blank_lines_at_end = 0;
2548                patch += len;
2549                size -= len;
2550        }
2551        if (inaccurate_eof &&
2552            old > oldlines && old[-1] == '\n' &&
2553            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2554                old--;
2555                strbuf_setlen(&newlines, newlines.len - 1);
2556        }
2557
2558        leading = frag->leading;
2559        trailing = frag->trailing;
2560
2561        /*
2562         * A hunk to change lines at the beginning would begin with
2563         * @@ -1,L +N,M @@
2564         * but we need to be careful.  -U0 that inserts before the second
2565         * line also has this pattern.
2566         *
2567         * And a hunk to add to an empty file would begin with
2568         * @@ -0,0 +N,M @@
2569         *
2570         * In other words, a hunk that is (frag->oldpos <= 1) with or
2571         * without leading context must match at the beginning.
2572         */
2573        match_beginning = (!frag->oldpos ||
2574                           (frag->oldpos == 1 && !unidiff_zero));
2575
2576        /*
2577         * A hunk without trailing lines must match at the end.
2578         * However, we simply cannot tell if a hunk must match end
2579         * from the lack of trailing lines if the patch was generated
2580         * with unidiff without any context.
2581         */
2582        match_end = !unidiff_zero && !trailing;
2583
2584        pos = frag->newpos ? (frag->newpos - 1) : 0;
2585        preimage.buf = oldlines;
2586        preimage.len = old - oldlines;
2587        postimage.buf = newlines.buf;
2588        postimage.len = newlines.len;
2589        preimage.line = preimage.line_allocated;
2590        postimage.line = postimage.line_allocated;
2591
2592        for (;;) {
2593
2594                applied_pos = find_pos(img, &preimage, &postimage, pos,
2595                                       ws_rule, match_beginning, match_end);
2596
2597                if (applied_pos >= 0)
2598                        break;
2599
2600                /* Am I at my context limits? */
2601                if ((leading <= p_context) && (trailing <= p_context))
2602                        break;
2603                if (match_beginning || match_end) {
2604                        match_beginning = match_end = 0;
2605                        continue;
2606                }
2607
2608                /*
2609                 * Reduce the number of context lines; reduce both
2610                 * leading and trailing if they are equal otherwise
2611                 * just reduce the larger context.
2612                 */
2613                if (leading >= trailing) {
2614                        remove_first_line(&preimage);
2615                        remove_first_line(&postimage);
2616                        pos--;
2617                        leading--;
2618                }
2619                if (trailing > leading) {
2620                        remove_last_line(&preimage);
2621                        remove_last_line(&postimage);
2622                        trailing--;
2623                }
2624        }
2625
2626        if (applied_pos >= 0) {
2627                if (new_blank_lines_at_end &&
2628                    preimage.nr + applied_pos >= img->nr &&
2629                    (ws_rule & WS_BLANK_AT_EOF) &&
2630                    ws_error_action != nowarn_ws_error) {
2631                        record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);
2632                        if (ws_error_action == correct_ws_error) {
2633                                while (new_blank_lines_at_end--)
2634                                        remove_last_line(&postimage);
2635                        }
2636                        /*
2637                         * We would want to prevent write_out_results()
2638                         * from taking place in apply_patch() that follows
2639                         * the callchain led us here, which is:
2640                         * apply_patch->check_patch_list->check_patch->
2641                         * apply_data->apply_fragments->apply_one_fragment
2642                         */
2643                        if (ws_error_action == die_on_ws_error)
2644                                apply = 0;
2645                }
2646
2647                /*
2648                 * Warn if it was necessary to reduce the number
2649                 * of context lines.
2650                 */
2651                if ((leading != frag->leading) ||
2652                    (trailing != frag->trailing))
2653                        fprintf(stderr, "Context reduced to (%ld/%ld)"
2654                                " to apply fragment at %d\n",
2655                                leading, trailing, applied_pos+1);
2656                update_image(img, applied_pos, &preimage, &postimage);
2657        } else {
2658                if (apply_verbosely)
2659                        error("while searching for:\n%.*s",
2660                              (int)(old - oldlines), oldlines);
2661        }
2662
2663        free(oldlines);
2664        strbuf_release(&newlines);
2665        free(preimage.line_allocated);
2666        free(postimage.line_allocated);
2667
2668        return (applied_pos < 0);
2669}
2670
2671static int apply_binary_fragment(struct image *img, struct patch *patch)
2672{
2673        struct fragment *fragment = patch->fragments;
2674        unsigned long len;
2675        void *dst;
2676
2677        if (!fragment)
2678                return error("missing binary patch data for '%s'",
2679                             patch->new_name ?
2680                             patch->new_name :
2681                             patch->old_name);
2682
2683        /* Binary patch is irreversible without the optional second hunk */
2684        if (apply_in_reverse) {
2685                if (!fragment->next)
2686                        return error("cannot reverse-apply a binary patch "
2687                                     "without the reverse hunk to '%s'",
2688                                     patch->new_name
2689                                     ? patch->new_name : patch->old_name);
2690                fragment = fragment->next;
2691        }
2692        switch (fragment->binary_patch_method) {
2693        case BINARY_DELTA_DEFLATED:
2694                dst = patch_delta(img->buf, img->len, fragment->patch,
2695                                  fragment->size, &len);
2696                if (!dst)
2697                        return -1;
2698                clear_image(img);
2699                img->buf = dst;
2700                img->len = len;
2701                return 0;
2702        case BINARY_LITERAL_DEFLATED:
2703                clear_image(img);
2704                img->len = fragment->size;
2705                img->buf = xmalloc(img->len+1);
2706                memcpy(img->buf, fragment->patch, img->len);
2707                img->buf[img->len] = '\0';
2708                return 0;
2709        }
2710        return -1;
2711}
2712
2713static int apply_binary(struct image *img, struct patch *patch)
2714{
2715        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2716        unsigned char sha1[20];
2717
2718        /*
2719         * For safety, we require patch index line to contain
2720         * full 40-byte textual SHA1 for old and new, at least for now.
2721         */
2722        if (strlen(patch->old_sha1_prefix) != 40 ||
2723            strlen(patch->new_sha1_prefix) != 40 ||
2724            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2725            get_sha1_hex(patch->new_sha1_prefix, sha1))
2726                return error("cannot apply binary patch to '%s' "
2727                             "without full index line", name);
2728
2729        if (patch->old_name) {
2730                /*
2731                 * See if the old one matches what the patch
2732                 * applies to.
2733                 */
2734                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2735                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2736                        return error("the patch applies to '%s' (%s), "
2737                                     "which does not match the "
2738                                     "current contents.",
2739                                     name, sha1_to_hex(sha1));
2740        }
2741        else {
2742                /* Otherwise, the old one must be empty. */
2743                if (img->len)
2744                        return error("the patch applies to an empty "
2745                                     "'%s' but it is not empty", name);
2746        }
2747
2748        get_sha1_hex(patch->new_sha1_prefix, sha1);
2749        if (is_null_sha1(sha1)) {
2750                clear_image(img);
2751                return 0; /* deletion patch */
2752        }
2753
2754        if (has_sha1_file(sha1)) {
2755                /* We already have the postimage */
2756                enum object_type type;
2757                unsigned long size;
2758                char *result;
2759
2760                result = read_sha1_file(sha1, &type, &size);
2761                if (!result)
2762                        return error("the necessary postimage %s for "
2763                                     "'%s' cannot be read",
2764                                     patch->new_sha1_prefix, name);
2765                clear_image(img);
2766                img->buf = result;
2767                img->len = size;
2768        } else {
2769                /*
2770                 * We have verified buf matches the preimage;
2771                 * apply the patch data to it, which is stored
2772                 * in the patch->fragments->{patch,size}.
2773                 */
2774                if (apply_binary_fragment(img, patch))
2775                        return error("binary patch does not apply to '%s'",
2776                                     name);
2777
2778                /* verify that the result matches */
2779                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2780                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2781                        return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2782                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2783        }
2784
2785        return 0;
2786}
2787
2788static int apply_fragments(struct image *img, struct patch *patch)
2789{
2790        struct fragment *frag = patch->fragments;
2791        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2792        unsigned ws_rule = patch->ws_rule;
2793        unsigned inaccurate_eof = patch->inaccurate_eof;
2794
2795        if (patch->is_binary)
2796                return apply_binary(img, patch);
2797
2798        while (frag) {
2799                if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2800                        error("patch failed: %s:%ld", name, frag->oldpos);
2801                        if (!apply_with_reject)
2802                                return -1;
2803                        frag->rejected = 1;
2804                }
2805                frag = frag->next;
2806        }
2807        return 0;
2808}
2809
2810static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2811{
2812        if (!ce)
2813                return 0;
2814
2815        if (S_ISGITLINK(ce->ce_mode)) {
2816                strbuf_grow(buf, 100);
2817                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2818        } else {
2819                enum object_type type;
2820                unsigned long sz;
2821                char *result;
2822
2823                result = read_sha1_file(ce->sha1, &type, &sz);
2824                if (!result)
2825                        return -1;
2826                /* XXX read_sha1_file NUL-terminates */
2827                strbuf_attach(buf, result, sz, sz + 1);
2828        }
2829        return 0;
2830}
2831
2832static struct patch *in_fn_table(const char *name)
2833{
2834        struct string_list_item *item;
2835
2836        if (name == NULL)
2837                return NULL;
2838
2839        item = string_list_lookup(&fn_table, name);
2840        if (item != NULL)
2841                return (struct patch *)item->util;
2842
2843        return NULL;
2844}
2845
2846/*
2847 * item->util in the filename table records the status of the path.
2848 * Usually it points at a patch (whose result records the contents
2849 * of it after applying it), but it could be PATH_WAS_DELETED for a
2850 * path that a previously applied patch has already removed.
2851 */
2852 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2853#define PATH_WAS_DELETED ((struct patch *) -1)
2854
2855static int to_be_deleted(struct patch *patch)
2856{
2857        return patch == PATH_TO_BE_DELETED;
2858}
2859
2860static int was_deleted(struct patch *patch)
2861{
2862        return patch == PATH_WAS_DELETED;
2863}
2864
2865static void add_to_fn_table(struct patch *patch)
2866{
2867        struct string_list_item *item;
2868
2869        /*
2870         * Always add new_name unless patch is a deletion
2871         * This should cover the cases for normal diffs,
2872         * file creations and copies
2873         */
2874        if (patch->new_name != NULL) {
2875                item = string_list_insert(&fn_table, patch->new_name);
2876                item->util = patch;
2877        }
2878
2879        /*
2880         * store a failure on rename/deletion cases because
2881         * later chunks shouldn't patch old names
2882         */
2883        if ((patch->new_name == NULL) || (patch->is_rename)) {
2884                item = string_list_insert(&fn_table, patch->old_name);
2885                item->util = PATH_WAS_DELETED;
2886        }
2887}
2888
2889static void prepare_fn_table(struct patch *patch)
2890{
2891        /*
2892         * store information about incoming file deletion
2893         */
2894        while (patch) {
2895                if ((patch->new_name == NULL) || (patch->is_rename)) {
2896                        struct string_list_item *item;
2897                        item = string_list_insert(&fn_table, patch->old_name);
2898                        item->util = PATH_TO_BE_DELETED;
2899                }
2900                patch = patch->next;
2901        }
2902}
2903
2904static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2905{
2906        struct strbuf buf = STRBUF_INIT;
2907        struct image image;
2908        size_t len;
2909        char *img;
2910        struct patch *tpatch;
2911
2912        if (!(patch->is_copy || patch->is_rename) &&
2913            (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2914                if (was_deleted(tpatch)) {
2915                        return error("patch %s has been renamed/deleted",
2916                                patch->old_name);
2917                }
2918                /* We have a patched copy in memory use that */
2919                strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2920        } else if (cached) {
2921                if (read_file_or_gitlink(ce, &buf))
2922                        return error("read of %s failed", patch->old_name);
2923        } else if (patch->old_name) {
2924                if (S_ISGITLINK(patch->old_mode)) {
2925                        if (ce) {
2926                                read_file_or_gitlink(ce, &buf);
2927                        } else {
2928                                /*
2929                                 * There is no way to apply subproject
2930                                 * patch without looking at the index.
2931                                 */
2932                                patch->fragments = NULL;
2933                        }
2934                } else {
2935                        if (read_old_data(st, patch->old_name, &buf))
2936                                return error("read of %s failed", patch->old_name);
2937                }
2938        }
2939
2940        img = strbuf_detach(&buf, &len);
2941        prepare_image(&image, img, len, !patch->is_binary);
2942
2943        if (apply_fragments(&image, patch) < 0)
2944                return -1; /* note with --reject this succeeds. */
2945        patch->result = image.buf;
2946        patch->resultsize = image.len;
2947        add_to_fn_table(patch);
2948        free(image.line_allocated);
2949
2950        if (0 < patch->is_delete && patch->resultsize)
2951                return error("removal patch leaves file contents");
2952
2953        return 0;
2954}
2955
2956static int check_to_create_blob(const char *new_name, int ok_if_exists)
2957{
2958        struct stat nst;
2959        if (!lstat(new_name, &nst)) {
2960                if (S_ISDIR(nst.st_mode) || ok_if_exists)
2961                        return 0;
2962                /*
2963                 * A leading component of new_name might be a symlink
2964                 * that is going to be removed with this patch, but
2965                 * still pointing at somewhere that has the path.
2966                 * In such a case, path "new_name" does not exist as
2967                 * far as git is concerned.
2968                 */
2969                if (has_symlink_leading_path(new_name, strlen(new_name)))
2970                        return 0;
2971
2972                return error("%s: already exists in working directory", new_name);
2973        }
2974        else if ((errno != ENOENT) && (errno != ENOTDIR))
2975                return error("%s: %s", new_name, strerror(errno));
2976        return 0;
2977}
2978
2979static int verify_index_match(struct cache_entry *ce, struct stat *st)
2980{
2981        if (S_ISGITLINK(ce->ce_mode)) {
2982                if (!S_ISDIR(st->st_mode))
2983                        return -1;
2984                return 0;
2985        }
2986        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2987}
2988
2989static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2990{
2991        const char *old_name = patch->old_name;
2992        struct patch *tpatch = NULL;
2993        int stat_ret = 0;
2994        unsigned st_mode = 0;
2995
2996        /*
2997         * Make sure that we do not have local modifications from the
2998         * index when we are looking at the index.  Also make sure
2999         * we have the preimage file to be patched in the work tree,
3000         * unless --cached, which tells git to apply only in the index.
3001         */
3002        if (!old_name)
3003                return 0;
3004
3005        assert(patch->is_new <= 0);
3006
3007        if (!(patch->is_copy || patch->is_rename) &&
3008            (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
3009                if (was_deleted(tpatch))
3010                        return error("%s: has been deleted/renamed", old_name);
3011                st_mode = tpatch->new_mode;
3012        } else if (!cached) {
3013                stat_ret = lstat(old_name, st);
3014                if (stat_ret && errno != ENOENT)
3015                        return error("%s: %s", old_name, strerror(errno));
3016        }
3017
3018        if (to_be_deleted(tpatch))
3019                tpatch = NULL;
3020
3021        if (check_index && !tpatch) {
3022                int pos = cache_name_pos(old_name, strlen(old_name));
3023                if (pos < 0) {
3024                        if (patch->is_new < 0)
3025                                goto is_new;
3026                        return error("%s: does not exist in index", old_name);
3027                }
3028                *ce = active_cache[pos];
3029                if (stat_ret < 0) {
3030                        struct checkout costate;
3031                        /* checkout */
3032                        memset(&costate, 0, sizeof(costate));
3033                        costate.base_dir = "";
3034                        costate.refresh_cache = 1;
3035                        if (checkout_entry(*ce, &costate, NULL) ||
3036                            lstat(old_name, st))
3037                                return -1;
3038                }
3039                if (!cached && verify_index_match(*ce, st))
3040                        return error("%s: does not match index", old_name);
3041                if (cached)
3042                        st_mode = (*ce)->ce_mode;
3043        } else if (stat_ret < 0) {
3044                if (patch->is_new < 0)
3045                        goto is_new;
3046                return error("%s: %s", old_name, strerror(errno));
3047        }
3048
3049        if (!cached && !tpatch)
3050                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3051
3052        if (patch->is_new < 0)
3053                patch->is_new = 0;
3054        if (!patch->old_mode)
3055                patch->old_mode = st_mode;
3056        if ((st_mode ^ patch->old_mode) & S_IFMT)
3057                return error("%s: wrong type", old_name);
3058        if (st_mode != patch->old_mode)
3059                warning("%s has type %o, expected %o",
3060                        old_name, st_mode, patch->old_mode);
3061        if (!patch->new_mode && !patch->is_delete)
3062                patch->new_mode = st_mode;
3063        return 0;
3064
3065 is_new:
3066        patch->is_new = 1;
3067        patch->is_delete = 0;
3068        patch->old_name = NULL;
3069        return 0;
3070}
3071
3072static int check_patch(struct patch *patch)
3073{
3074        struct stat st;
3075        const char *old_name = patch->old_name;
3076        const char *new_name = patch->new_name;
3077        const char *name = old_name ? old_name : new_name;
3078        struct cache_entry *ce = NULL;
3079        struct patch *tpatch;
3080        int ok_if_exists;
3081        int status;
3082
3083        patch->rejected = 1; /* we will drop this after we succeed */
3084
3085        status = check_preimage(patch, &ce, &st);
3086        if (status)
3087                return status;
3088        old_name = patch->old_name;
3089
3090        if ((tpatch = in_fn_table(new_name)) &&
3091                        (was_deleted(tpatch) || to_be_deleted(tpatch)))
3092                /*
3093                 * A type-change diff is always split into a patch to
3094                 * delete old, immediately followed by a patch to
3095                 * create new (see diff.c::run_diff()); in such a case
3096                 * it is Ok that the entry to be deleted by the
3097                 * previous patch is still in the working tree and in
3098                 * the index.
3099                 */
3100                ok_if_exists = 1;
3101        else
3102                ok_if_exists = 0;
3103
3104        if (new_name &&
3105            ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3106                if (check_index &&
3107                    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3108                    !ok_if_exists)
3109                        return error("%s: already exists in index", new_name);
3110                if (!cached) {
3111                        int err = check_to_create_blob(new_name, ok_if_exists);
3112                        if (err)
3113                                return err;
3114                }
3115                if (!patch->new_mode) {
3116                        if (0 < patch->is_new)
3117                                patch->new_mode = S_IFREG | 0644;
3118                        else
3119                                patch->new_mode = patch->old_mode;
3120                }
3121        }
3122
3123        if (new_name && old_name) {
3124                int same = !strcmp(old_name, new_name);
3125                if (!patch->new_mode)
3126                        patch->new_mode = patch->old_mode;
3127                if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
3128                        return error("new mode (%o) of %s does not match old mode (%o)%s%s",
3129                                patch->new_mode, new_name, patch->old_mode,
3130                                same ? "" : " of ", same ? "" : old_name);
3131        }
3132
3133        if (apply_data(patch, &st, ce) < 0)
3134                return error("%s: patch does not apply", name);
3135        patch->rejected = 0;
3136        return 0;
3137}
3138
3139static int check_patch_list(struct patch *patch)
3140{
3141        int err = 0;
3142
3143        prepare_fn_table(patch);
3144        while (patch) {
3145                if (apply_verbosely)
3146                        say_patch_name(stderr,
3147                                       "Checking patch ", patch, "...\n");
3148                err |= check_patch(patch);
3149                patch = patch->next;
3150        }
3151        return err;
3152}
3153
3154/* This function tries to read the sha1 from the current index */
3155static int get_current_sha1(const char *path, unsigned char *sha1)
3156{
3157        int pos;
3158
3159        if (read_cache() < 0)
3160                return -1;
3161        pos = cache_name_pos(path, strlen(path));
3162        if (pos < 0)
3163                return -1;
3164        hashcpy(sha1, active_cache[pos]->sha1);
3165        return 0;
3166}
3167
3168/* Build an index that contains the just the files needed for a 3way merge */
3169static void build_fake_ancestor(struct patch *list, const char *filename)
3170{
3171        struct patch *patch;
3172        struct index_state result = { NULL };
3173        int fd;
3174
3175        /* Once we start supporting the reverse patch, it may be
3176         * worth showing the new sha1 prefix, but until then...
3177         */
3178        for (patch = list; patch; patch = patch->next) {
3179                const unsigned char *sha1_ptr;
3180                unsigned char sha1[20];
3181                struct cache_entry *ce;
3182                const char *name;
3183
3184                name = patch->old_name ? patch->old_name : patch->new_name;
3185                if (0 < patch->is_new)
3186                        continue;
3187                else if (get_sha1(patch->old_sha1_prefix, sha1))
3188                        /* git diff has no index line for mode/type changes */
3189                        if (!patch->lines_added && !patch->lines_deleted) {
3190                                if (get_current_sha1(patch->old_name, sha1))
3191                                        die("mode change for %s, which is not "
3192                                                "in current HEAD", name);
3193                                sha1_ptr = sha1;
3194                        } else
3195                                die("sha1 information is lacking or useless "
3196                                        "(%s).", name);
3197                else
3198                        sha1_ptr = sha1;
3199
3200                ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3201                if (!ce)
3202                        die("make_cache_entry failed for path '%s'", name);
3203                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3204                        die ("Could not add %s to temporary index", name);
3205        }
3206
3207        fd = open(filename, O_WRONLY | O_CREAT, 0666);
3208        if (fd < 0 || write_index(&result, fd) || close(fd))
3209                die ("Could not write temporary index to %s", filename);
3210
3211        discard_index(&result);
3212}
3213
3214static void stat_patch_list(struct patch *patch)
3215{
3216        int files, adds, dels;
3217
3218        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3219                files++;
3220                adds += patch->lines_added;
3221                dels += patch->lines_deleted;
3222                show_stats(patch);
3223        }
3224
3225        printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
3226}
3227
3228static void numstat_patch_list(struct patch *patch)
3229{
3230        for ( ; patch; patch = patch->next) {
3231                const char *name;
3232                name = patch->new_name ? patch->new_name : patch->old_name;
3233                if (patch->is_binary)
3234                        printf("-\t-\t");
3235                else
3236                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3237                write_name_quoted(name, stdout, line_termination);
3238        }
3239}
3240
3241static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3242{
3243        if (mode)
3244                printf(" %s mode %06o %s\n", newdelete, mode, name);
3245        else
3246                printf(" %s %s\n", newdelete, name);
3247}
3248
3249static void show_mode_change(struct patch *p, int show_name)
3250{
3251        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3252                if (show_name)
3253                        printf(" mode change %06o => %06o %s\n",
3254                               p->old_mode, p->new_mode, p->new_name);
3255                else
3256                        printf(" mode change %06o => %06o\n",
3257                               p->old_mode, p->new_mode);
3258        }
3259}
3260
3261static void show_rename_copy(struct patch *p)
3262{
3263        const char *renamecopy = p->is_rename ? "rename" : "copy";
3264        const char *old, *new;
3265
3266        /* Find common prefix */
3267        old = p->old_name;
3268        new = p->new_name;
3269        while (1) {
3270                const char *slash_old, *slash_new;
3271                slash_old = strchr(old, '/');
3272                slash_new = strchr(new, '/');
3273                if (!slash_old ||
3274                    !slash_new ||
3275                    slash_old - old != slash_new - new ||
3276                    memcmp(old, new, slash_new - new))
3277                        break;
3278                old = slash_old + 1;
3279                new = slash_new + 1;
3280        }
3281        /* p->old_name thru old is the common prefix, and old and new
3282         * through the end of names are renames
3283         */
3284        if (old != p->old_name)
3285                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3286                       (int)(old - p->old_name), p->old_name,
3287                       old, new, p->score);
3288        else
3289                printf(" %s %s => %s (%d%%)\n", renamecopy,
3290                       p->old_name, p->new_name, p->score);
3291        show_mode_change(p, 0);
3292}
3293
3294static void summary_patch_list(struct patch *patch)
3295{
3296        struct patch *p;
3297
3298        for (p = patch; p; p = p->next) {
3299                if (p->is_new)
3300                        show_file_mode_name("create", p->new_mode, p->new_name);
3301                else if (p->is_delete)
3302                        show_file_mode_name("delete", p->old_mode, p->old_name);
3303                else {
3304                        if (p->is_rename || p->is_copy)
3305                                show_rename_copy(p);
3306                        else {
3307                                if (p->score) {
3308                                        printf(" rewrite %s (%d%%)\n",
3309                                               p->new_name, p->score);
3310                                        show_mode_change(p, 0);
3311                                }
3312                                else
3313                                        show_mode_change(p, 1);
3314                        }
3315                }
3316        }
3317}
3318
3319static void patch_stats(struct patch *patch)
3320{
3321        int lines = patch->lines_added + patch->lines_deleted;
3322
3323        if (lines > max_change)
3324                max_change = lines;
3325        if (patch->old_name) {
3326                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3327                if (!len)
3328                        len = strlen(patch->old_name);
3329                if (len > max_len)
3330                        max_len = len;
3331        }
3332        if (patch->new_name) {
3333                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3334                if (!len)
3335                        len = strlen(patch->new_name);
3336                if (len > max_len)
3337                        max_len = len;
3338        }
3339}
3340
3341static void remove_file(struct patch *patch, int rmdir_empty)
3342{
3343        if (update_index) {
3344                if (remove_file_from_cache(patch->old_name) < 0)
3345                        die("unable to remove %s from index", patch->old_name);
3346        }
3347        if (!cached) {
3348                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3349                        remove_path(patch->old_name);
3350                }
3351        }
3352}
3353
3354static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3355{
3356        struct stat st;
3357        struct cache_entry *ce;
3358        int namelen = strlen(path);
3359        unsigned ce_size = cache_entry_size(namelen);
3360
3361        if (!update_index)
3362                return;
3363
3364        ce = xcalloc(1, ce_size);
3365        memcpy(ce->name, path, namelen);
3366        ce->ce_mode = create_ce_mode(mode);
3367        ce->ce_flags = namelen;
3368        if (S_ISGITLINK(mode)) {
3369                const char *s = buf;
3370
3371                if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3372                        die("corrupt patch for subproject %s", path);
3373        } else {
3374                if (!cached) {
3375                        if (lstat(path, &st) < 0)
3376                                die_errno("unable to stat newly created file '%s'",
3377                                          path);
3378                        fill_stat_cache_info(ce, &st);
3379                }
3380                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3381                        die("unable to create backing store for newly created file %s", path);
3382        }
3383        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3384                die("unable to add cache entry for %s", path);
3385}
3386
3387static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3388{
3389        int fd;
3390        struct strbuf nbuf = STRBUF_INIT;
3391
3392        if (S_ISGITLINK(mode)) {
3393                struct stat st;
3394                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3395                        return 0;
3396                return mkdir(path, 0777);
3397        }
3398
3399        if (has_symlinks && S_ISLNK(mode))
3400                /* Although buf:size is counted string, it also is NUL
3401                 * terminated.
3402                 */
3403                return symlink(buf, path);
3404
3405        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3406        if (fd < 0)
3407                return -1;
3408
3409        if (convert_to_working_tree(path, buf, size, &nbuf)) {
3410                size = nbuf.len;
3411                buf  = nbuf.buf;
3412        }
3413        write_or_die(fd, buf, size);
3414        strbuf_release(&nbuf);
3415
3416        if (close(fd) < 0)
3417                die_errno("closing file '%s'", path);
3418        return 0;
3419}
3420
3421/*
3422 * We optimistically assume that the directories exist,
3423 * which is true 99% of the time anyway. If they don't,
3424 * we create them and try again.
3425 */
3426static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3427{
3428        if (cached)
3429                return;
3430        if (!try_create_file(path, mode, buf, size))
3431                return;
3432
3433        if (errno == ENOENT) {
3434                if (safe_create_leading_directories(path))
3435                        return;
3436                if (!try_create_file(path, mode, buf, size))
3437                        return;
3438        }
3439
3440        if (errno == EEXIST || errno == EACCES) {
3441                /* We may be trying to create a file where a directory
3442                 * used to be.
3443                 */
3444                struct stat st;
3445                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3446                        errno = EEXIST;
3447        }
3448
3449        if (errno == EEXIST) {
3450                unsigned int nr = getpid();
3451
3452                for (;;) {
3453                        char newpath[PATH_MAX];
3454                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3455                        if (!try_create_file(newpath, mode, buf, size)) {
3456                                if (!rename(newpath, path))
3457                                        return;
3458                                unlink_or_warn(newpath);
3459                                break;
3460                        }
3461                        if (errno != EEXIST)
3462                                break;
3463                        ++nr;
3464                }
3465        }
3466        die_errno("unable to write file '%s' mode %o", path, mode);
3467}
3468
3469static void create_file(struct patch *patch)
3470{
3471        char *path = patch->new_name;
3472        unsigned mode = patch->new_mode;
3473        unsigned long size = patch->resultsize;
3474        char *buf = patch->result;
3475
3476        if (!mode)
3477                mode = S_IFREG | 0644;
3478        create_one_file(path, mode, buf, size);
3479        add_index_file(path, mode, buf, size);
3480}
3481
3482/* phase zero is to remove, phase one is to create */
3483static void write_out_one_result(struct patch *patch, int phase)
3484{
3485        if (patch->is_delete > 0) {
3486                if (phase == 0)
3487                        remove_file(patch, 1);
3488                return;
3489        }
3490        if (patch->is_new > 0 || patch->is_copy) {
3491                if (phase == 1)
3492                        create_file(patch);
3493                return;
3494        }
3495        /*
3496         * Rename or modification boils down to the same
3497         * thing: remove the old, write the new
3498         */
3499        if (phase == 0)
3500                remove_file(patch, patch->is_rename);
3501        if (phase == 1)
3502                create_file(patch);
3503}
3504
3505static int write_out_one_reject(struct patch *patch)
3506{
3507        FILE *rej;
3508        char namebuf[PATH_MAX];
3509        struct fragment *frag;
3510        int cnt = 0;
3511
3512        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3513                if (!frag->rejected)
3514                        continue;
3515                cnt++;
3516        }
3517
3518        if (!cnt) {
3519                if (apply_verbosely)
3520                        say_patch_name(stderr,
3521                                       "Applied patch ", patch, " cleanly.\n");
3522                return 0;
3523        }
3524
3525        /* This should not happen, because a removal patch that leaves
3526         * contents are marked "rejected" at the patch level.
3527         */
3528        if (!patch->new_name)
3529                die("internal error");
3530
3531        /* Say this even without --verbose */
3532        say_patch_name(stderr, "Applying patch ", patch, " with");
3533        fprintf(stderr, " %d rejects...\n", cnt);
3534
3535        cnt = strlen(patch->new_name);
3536        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3537                cnt = ARRAY_SIZE(namebuf) - 5;
3538                warning("truncating .rej filename to %.*s.rej",
3539                        cnt - 1, patch->new_name);
3540        }
3541        memcpy(namebuf, patch->new_name, cnt);
3542        memcpy(namebuf + cnt, ".rej", 5);
3543
3544        rej = fopen(namebuf, "w");
3545        if (!rej)
3546                return error("cannot open %s: %s", namebuf, strerror(errno));
3547
3548        /* Normal git tools never deal with .rej, so do not pretend
3549         * this is a git patch by saying --git nor give extended
3550         * headers.  While at it, maybe please "kompare" that wants
3551         * the trailing TAB and some garbage at the end of line ;-).
3552         */
3553        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3554                patch->new_name, patch->new_name);
3555        for (cnt = 1, frag = patch->fragments;
3556             frag;
3557             cnt++, frag = frag->next) {
3558                if (!frag->rejected) {
3559                        fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3560                        continue;
3561                }
3562                fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3563                fprintf(rej, "%.*s", frag->size, frag->patch);
3564                if (frag->patch[frag->size-1] != '\n')
3565                        fputc('\n', rej);
3566        }
3567        fclose(rej);
3568        return -1;
3569}
3570
3571static int write_out_results(struct patch *list, int skipped_patch)
3572{
3573        int phase;
3574        int errs = 0;
3575        struct patch *l;
3576
3577        if (!list && !skipped_patch)
3578                return error("No changes");
3579
3580        for (phase = 0; phase < 2; phase++) {
3581                l = list;
3582                while (l) {
3583                        if (l->rejected)
3584                                errs = 1;
3585                        else {
3586                                write_out_one_result(l, phase);
3587                                if (phase == 1 && write_out_one_reject(l))
3588                                        errs = 1;
3589                        }
3590                        l = l->next;
3591                }
3592        }
3593        return errs;
3594}
3595
3596static struct lock_file lock_file;
3597
3598static struct string_list limit_by_name;
3599static int has_include;
3600static void add_name_limit(const char *name, int exclude)
3601{
3602        struct string_list_item *it;
3603
3604        it = string_list_append(&limit_by_name, name);
3605        it->util = exclude ? NULL : (void *) 1;
3606}
3607
3608static int use_patch(struct patch *p)
3609{
3610        const char *pathname = p->new_name ? p->new_name : p->old_name;
3611        int i;
3612
3613        /* Paths outside are not touched regardless of "--include" */
3614        if (0 < prefix_length) {
3615                int pathlen = strlen(pathname);
3616                if (pathlen <= prefix_length ||
3617                    memcmp(prefix, pathname, prefix_length))
3618                        return 0;
3619        }
3620
3621        /* See if it matches any of exclude/include rule */
3622        for (i = 0; i < limit_by_name.nr; i++) {
3623                struct string_list_item *it = &limit_by_name.items[i];
3624                if (!fnmatch(it->string, pathname, 0))
3625                        return (it->util != NULL);
3626        }
3627
3628        /*
3629         * If we had any include, a path that does not match any rule is
3630         * not used.  Otherwise, we saw bunch of exclude rules (or none)
3631         * and such a path is used.
3632         */
3633        return !has_include;
3634}
3635
3636
3637static void prefix_one(char **name)
3638{
3639        char *old_name = *name;
3640        if (!old_name)
3641                return;
3642        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3643        free(old_name);
3644}
3645
3646static void prefix_patches(struct patch *p)
3647{
3648        if (!prefix || p->is_toplevel_relative)
3649                return;
3650        for ( ; p; p = p->next) {
3651                if (p->new_name == p->old_name) {
3652                        char *prefixed = p->new_name;
3653                        prefix_one(&prefixed);
3654                        p->new_name = p->old_name = prefixed;
3655                }
3656                else {
3657                        prefix_one(&p->new_name);
3658                        prefix_one(&p->old_name);
3659                }
3660        }
3661}
3662
3663#define INACCURATE_EOF  (1<<0)
3664#define RECOUNT         (1<<1)
3665
3666static int apply_patch(int fd, const char *filename, int options)
3667{
3668        size_t offset;
3669        struct strbuf buf = STRBUF_INIT;
3670        struct patch *list = NULL, **listp = &list;
3671        int skipped_patch = 0;
3672
3673        /* FIXME - memory leak when using multiple patch files as inputs */
3674        memset(&fn_table, 0, sizeof(struct string_list));
3675        patch_input_file = filename;
3676        read_patch_file(&buf, fd);
3677        offset = 0;
3678        while (offset < buf.len) {
3679                struct patch *patch;
3680                int nr;
3681
3682                patch = xcalloc(1, sizeof(*patch));
3683                patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3684                patch->recount =  !!(options & RECOUNT);
3685                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3686                if (nr < 0)
3687                        break;
3688                if (apply_in_reverse)
3689                        reverse_patches(patch);
3690                if (prefix)
3691                        prefix_patches(patch);
3692                if (use_patch(patch)) {
3693                        patch_stats(patch);
3694                        *listp = patch;
3695                        listp = &patch->next;
3696                }
3697                else {
3698                        /* perhaps free it a bit better? */
3699                        free(patch);
3700                        skipped_patch++;
3701                }
3702                offset += nr;
3703        }
3704
3705        if (whitespace_error && (ws_error_action == die_on_ws_error))
3706                apply = 0;
3707
3708        update_index = check_index && apply;
3709        if (update_index && newfd < 0)
3710                newfd = hold_locked_index(&lock_file, 1);
3711
3712        if (check_index) {
3713                if (read_cache() < 0)
3714                        die("unable to read index file");
3715        }
3716
3717        if ((check || apply) &&
3718            check_patch_list(list) < 0 &&
3719            !apply_with_reject)
3720                exit(1);
3721
3722        if (apply && write_out_results(list, skipped_patch))
3723                exit(1);
3724
3725        if (fake_ancestor)
3726                build_fake_ancestor(list, fake_ancestor);
3727
3728        if (diffstat)
3729                stat_patch_list(list);
3730
3731        if (numstat)
3732                numstat_patch_list(list);
3733
3734        if (summary)
3735                summary_patch_list(list);
3736
3737        strbuf_release(&buf);
3738        return 0;
3739}
3740
3741static int git_apply_config(const char *var, const char *value, void *cb)
3742{
3743        if (!strcmp(var, "apply.whitespace"))
3744                return git_config_string(&apply_default_whitespace, var, value);
3745        else if (!strcmp(var, "apply.ignorewhitespace"))
3746                return git_config_string(&apply_default_ignorewhitespace, var, value);
3747        return git_default_config(var, value, cb);
3748}
3749
3750static int option_parse_exclude(const struct option *opt,
3751                                const char *arg, int unset)
3752{
3753        add_name_limit(arg, 1);
3754        return 0;
3755}
3756
3757static int option_parse_include(const struct option *opt,
3758                                const char *arg, int unset)
3759{
3760        add_name_limit(arg, 0);
3761        has_include = 1;
3762        return 0;
3763}
3764
3765static int option_parse_p(const struct option *opt,
3766                          const char *arg, int unset)
3767{
3768        p_value = atoi(arg);
3769        p_value_known = 1;
3770        return 0;
3771}
3772
3773static int option_parse_z(const struct option *opt,
3774                          const char *arg, int unset)
3775{
3776        if (unset)
3777                line_termination = '\n';
3778        else
3779                line_termination = 0;
3780        return 0;
3781}
3782
3783static int option_parse_space_change(const struct option *opt,
3784                          const char *arg, int unset)
3785{
3786        if (unset)
3787                ws_ignore_action = ignore_ws_none;
3788        else
3789                ws_ignore_action = ignore_ws_change;
3790        return 0;
3791}
3792
3793static int option_parse_whitespace(const struct option *opt,
3794                                   const char *arg, int unset)
3795{
3796        const char **whitespace_option = opt->value;
3797
3798        *whitespace_option = arg;
3799        parse_whitespace_option(arg);
3800        return 0;
3801}
3802
3803static int option_parse_directory(const struct option *opt,
3804                                  const char *arg, int unset)
3805{
3806        root_len = strlen(arg);
3807        if (root_len && arg[root_len - 1] != '/') {
3808                char *new_root;
3809                root = new_root = xmalloc(root_len + 2);
3810                strcpy(new_root, arg);
3811                strcpy(new_root + root_len++, "/");
3812        } else
3813                root = arg;
3814        return 0;
3815}
3816
3817int cmd_apply(int argc, const char **argv, const char *prefix_)
3818{
3819        int i;
3820        int errs = 0;
3821        int is_not_gitdir = !startup_info->have_repository;
3822        int binary;
3823        int force_apply = 0;
3824
3825        const char *whitespace_option = NULL;
3826
3827        struct option builtin_apply_options[] = {
3828                { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3829                        "don't apply changes matching the given path",
3830                        0, option_parse_exclude },
3831                { OPTION_CALLBACK, 0, "include", NULL, "path",
3832                        "apply changes matching the given path",
3833                        0, option_parse_include },
3834                { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3835                        "remove <num> leading slashes from traditional diff paths",
3836                        0, option_parse_p },
3837                OPT_BOOLEAN(0, "no-add", &no_add,
3838                        "ignore additions made by the patch"),
3839                OPT_BOOLEAN(0, "stat", &diffstat,
3840                        "instead of applying the patch, output diffstat for the input"),
3841                { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3842                  NULL, "old option, now no-op",
3843                  PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3844                { OPTION_BOOLEAN, 0, "binary", &binary,
3845                  NULL, "old option, now no-op",
3846                  PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3847                OPT_BOOLEAN(0, "numstat", &numstat,
3848                        "shows number of added and deleted lines in decimal notation"),
3849                OPT_BOOLEAN(0, "summary", &summary,
3850                        "instead of applying the patch, output a summary for the input"),
3851                OPT_BOOLEAN(0, "check", &check,
3852                        "instead of applying the patch, see if the patch is applicable"),
3853                OPT_BOOLEAN(0, "index", &check_index,
3854                        "make sure the patch is applicable to the current index"),
3855                OPT_BOOLEAN(0, "cached", &cached,
3856                        "apply a patch without touching the working tree"),
3857                OPT_BOOLEAN(0, "apply", &force_apply,
3858                        "also apply the patch (use with --stat/--summary/--check)"),
3859                OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3860                        "build a temporary index based on embedded index information"),
3861                { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3862                        "paths are separated with NUL character",
3863                        PARSE_OPT_NOARG, option_parse_z },
3864                OPT_INTEGER('C', NULL, &p_context,
3865                                "ensure at least <n> lines of context match"),
3866                { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3867                        "detect new or modified lines that have whitespace errors",
3868                        0, option_parse_whitespace },
3869                { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3870                        "ignore changes in whitespace when finding context",
3871                        PARSE_OPT_NOARG, option_parse_space_change },
3872                { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3873                        "ignore changes in whitespace when finding context",
3874                        PARSE_OPT_NOARG, option_parse_space_change },
3875                OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3876                        "apply the patch in reverse"),
3877                OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3878                        "don't expect at least one line of context"),
3879                OPT_BOOLEAN(0, "reject", &apply_with_reject,
3880                        "leave the rejected hunks in corresponding *.rej files"),
3881                OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
3882                        "allow overlapping hunks"),
3883                OPT__VERBOSE(&apply_verbosely, "be verbose"),
3884                OPT_BIT(0, "inaccurate-eof", &options,
3885                        "tolerate incorrectly detected missing new-line at the end of file",
3886                        INACCURATE_EOF),
3887                OPT_BIT(0, "recount", &options,
3888                        "do not trust the line counts in the hunk headers",
3889                        RECOUNT),
3890                { OPTION_CALLBACK, 0, "directory", NULL, "root",
3891                        "prepend <root> to all filenames",
3892                        0, option_parse_directory },
3893                OPT_END()
3894        };
3895
3896        prefix = prefix_;
3897        prefix_length = prefix ? strlen(prefix) : 0;
3898        git_config(git_apply_config, NULL);
3899        if (apply_default_whitespace)
3900                parse_whitespace_option(apply_default_whitespace);
3901        if (apply_default_ignorewhitespace)
3902                parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3903
3904        argc = parse_options(argc, argv, prefix, builtin_apply_options,
3905                        apply_usage, 0);
3906
3907        if (apply_with_reject)
3908                apply = apply_verbosely = 1;
3909        if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3910                apply = 0;
3911        if (check_index && is_not_gitdir)
3912                die("--index outside a repository");
3913        if (cached) {
3914                if (is_not_gitdir)
3915                        die("--cached outside a repository");
3916                check_index = 1;
3917        }
3918        for (i = 0; i < argc; i++) {
3919                const char *arg = argv[i];
3920                int fd;
3921
3922                if (!strcmp(arg, "-")) {
3923                        errs |= apply_patch(0, "<stdin>", options);
3924                        read_stdin = 0;
3925                        continue;
3926                } else if (0 < prefix_length)
3927                        arg = prefix_filename(prefix, prefix_length, arg);
3928
3929                fd = open(arg, O_RDONLY);
3930                if (fd < 0)
3931                        die_errno("can't open patch '%s'", arg);
3932                read_stdin = 0;
3933                set_default_whitespace_mode(whitespace_option);
3934                errs |= apply_patch(fd, arg, options);
3935                close(fd);
3936        }
3937        set_default_whitespace_mode(whitespace_option);
3938        if (read_stdin)
3939                errs |= apply_patch(0, "<stdin>", options);
3940        if (whitespace_error) {
3941                if (squelch_whitespace_errors &&
3942                    squelch_whitespace_errors < whitespace_error) {
3943                        int squelched =
3944                                whitespace_error - squelch_whitespace_errors;
3945                        warning("squelched %d "
3946                                "whitespace error%s",
3947                                squelched,
3948                                squelched == 1 ? "" : "s");
3949                }
3950                if (ws_error_action == die_on_ws_error)
3951                        die("%d line%s add%s whitespace errors.",
3952                            whitespace_error,
3953                            whitespace_error == 1 ? "" : "s",
3954                            whitespace_error == 1 ? "s" : "");
3955                if (applied_after_fixing_ws && apply)
3956                        warning("%d line%s applied after"
3957                                " fixing whitespace errors.",
3958                                applied_after_fixing_ws,
3959                                applied_after_fixing_ws == 1 ? "" : "s");
3960                else if (whitespace_error)
3961                        warning("%d line%s add%s whitespace errors.",
3962                                whitespace_error,
3963                                whitespace_error == 1 ? "" : "s",
3964                                whitespace_error == 1 ? "s" : "");
3965        }
3966
3967        if (update_index) {
3968                if (write_cache(newfd, active_cache, active_nr) ||
3969                    commit_locked_index(&lock_file))
3970                        die("Unable to write new index file");
3971        }
3972
3973        return !!errs;
3974}