builtin / apply.con commit shortlog: fix wrapping lines of wraplen (5b59708)
   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 "diff.h"
  18#include "parse-options.h"
  19
  20/*
  21 *  --check turns on checking that the working tree matches the
  22 *    files that are being modified, but doesn't apply the patch
  23 *  --stat does just a diffstat, and doesn't actually apply
  24 *  --numstat does numeric diffstat, and doesn't actually apply
  25 *  --index-info shows the old and new index info for paths if available.
  26 *  --index updates the cache as well.
  27 *  --cached updates only the cache without ever touching the working tree.
  28 */
  29static const char *prefix;
  30static int prefix_length = -1;
  31static int newfd = -1;
  32
  33static int unidiff_zero;
  34static int p_value = 1;
  35static int p_value_known;
  36static int check_index;
  37static int update_index;
  38static int cached;
  39static int diffstat;
  40static int numstat;
  41static int summary;
  42static int check;
  43static int apply = 1;
  44static int apply_in_reverse;
  45static int apply_with_reject;
  46static int apply_verbosely;
  47static int allow_overlap;
  48static int no_add;
  49static const char *fake_ancestor;
  50static int line_termination = '\n';
  51static unsigned int p_context = UINT_MAX;
  52static const char * const apply_usage[] = {
  53        N_("git apply [options] [<patch>...]"),
  54        NULL
  55};
  56
  57static enum ws_error_action {
  58        nowarn_ws_error,
  59        warn_on_ws_error,
  60        die_on_ws_error,
  61        correct_ws_error
  62} ws_error_action = warn_on_ws_error;
  63static int whitespace_error;
  64static int squelch_whitespace_errors = 5;
  65static int applied_after_fixing_ws;
  66
  67static enum ws_ignore {
  68        ignore_ws_none,
  69        ignore_ws_change
  70} ws_ignore_action = ignore_ws_none;
  71
  72
  73static const char *patch_input_file;
  74static const char *root;
  75static int root_len;
  76static int read_stdin = 1;
  77static int options;
  78
  79static void parse_whitespace_option(const char *option)
  80{
  81        if (!option) {
  82                ws_error_action = warn_on_ws_error;
  83                return;
  84        }
  85        if (!strcmp(option, "warn")) {
  86                ws_error_action = warn_on_ws_error;
  87                return;
  88        }
  89        if (!strcmp(option, "nowarn")) {
  90                ws_error_action = nowarn_ws_error;
  91                return;
  92        }
  93        if (!strcmp(option, "error")) {
  94                ws_error_action = die_on_ws_error;
  95                return;
  96        }
  97        if (!strcmp(option, "error-all")) {
  98                ws_error_action = die_on_ws_error;
  99                squelch_whitespace_errors = 0;
 100                return;
 101        }
 102        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
 103                ws_error_action = correct_ws_error;
 104                return;
 105        }
 106        die(_("unrecognized whitespace option '%s'"), option);
 107}
 108
 109static void parse_ignorewhitespace_option(const char *option)
 110{
 111        if (!option || !strcmp(option, "no") ||
 112            !strcmp(option, "false") || !strcmp(option, "never") ||
 113            !strcmp(option, "none")) {
 114                ws_ignore_action = ignore_ws_none;
 115                return;
 116        }
 117        if (!strcmp(option, "change")) {
 118                ws_ignore_action = ignore_ws_change;
 119                return;
 120        }
 121        die(_("unrecognized whitespace ignore option '%s'"), option);
 122}
 123
 124static void set_default_whitespace_mode(const char *whitespace_option)
 125{
 126        if (!whitespace_option && !apply_default_whitespace)
 127                ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
 128}
 129
 130/*
 131 * For "diff-stat" like behaviour, we keep track of the biggest change
 132 * we've seen, and the longest filename. That allows us to do simple
 133 * scaling.
 134 */
 135static int max_change, max_len;
 136
 137/*
 138 * Various "current state", notably line numbers and what
 139 * file (and how) we're patching right now.. The "is_xxxx"
 140 * things are flags, where -1 means "don't know yet".
 141 */
 142static int linenr = 1;
 143
 144/*
 145 * This represents one "hunk" from a patch, starting with
 146 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 147 * patch text is pointed at by patch, and its byte length
 148 * is stored in size.  leading and trailing are the number
 149 * of context lines.
 150 */
 151struct fragment {
 152        unsigned long leading, trailing;
 153        unsigned long oldpos, oldlines;
 154        unsigned long newpos, newlines;
 155        /*
 156         * 'patch' is usually borrowed from buf in apply_patch(),
 157         * but some codepaths store an allocated buffer.
 158         */
 159        const char *patch;
 160        unsigned free_patch:1,
 161                rejected:1;
 162        int size;
 163        int linenr;
 164        struct fragment *next;
 165};
 166
 167/*
 168 * When dealing with a binary patch, we reuse "leading" field
 169 * to store the type of the binary hunk, either deflated "delta"
 170 * or deflated "literal".
 171 */
 172#define binary_patch_method leading
 173#define BINARY_DELTA_DEFLATED   1
 174#define BINARY_LITERAL_DEFLATED 2
 175
 176/*
 177 * This represents a "patch" to a file, both metainfo changes
 178 * such as creation/deletion, filemode and content changes represented
 179 * as a series of fragments.
 180 */
 181struct patch {
 182        char *new_name, *old_name, *def_name;
 183        unsigned int old_mode, new_mode;
 184        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 185        int rejected;
 186        unsigned ws_rule;
 187        int lines_added, lines_deleted;
 188        int score;
 189        unsigned int is_toplevel_relative:1;
 190        unsigned int inaccurate_eof:1;
 191        unsigned int is_binary:1;
 192        unsigned int is_copy:1;
 193        unsigned int is_rename:1;
 194        unsigned int recount:1;
 195        struct fragment *fragments;
 196        char *result;
 197        size_t resultsize;
 198        char old_sha1_prefix[41];
 199        char new_sha1_prefix[41];
 200        struct patch *next;
 201};
 202
 203static void free_fragment_list(struct fragment *list)
 204{
 205        while (list) {
 206                struct fragment *next = list->next;
 207                if (list->free_patch)
 208                        free((char *)list->patch);
 209                free(list);
 210                list = next;
 211        }
 212}
 213
 214static void free_patch(struct patch *patch)
 215{
 216        free_fragment_list(patch->fragments);
 217        free(patch->def_name);
 218        free(patch->old_name);
 219        free(patch->new_name);
 220        free(patch->result);
 221        free(patch);
 222}
 223
 224static void free_patch_list(struct patch *list)
 225{
 226        while (list) {
 227                struct patch *next = list->next;
 228                free_patch(list);
 229                list = next;
 230        }
 231}
 232
 233/*
 234 * A line in a file, len-bytes long (includes the terminating LF,
 235 * except for an incomplete line at the end if the file ends with
 236 * one), and its contents hashes to 'hash'.
 237 */
 238struct line {
 239        size_t len;
 240        unsigned hash : 24;
 241        unsigned flag : 8;
 242#define LINE_COMMON     1
 243#define LINE_PATCHED    2
 244};
 245
 246/*
 247 * This represents a "file", which is an array of "lines".
 248 */
 249struct image {
 250        char *buf;
 251        size_t len;
 252        size_t nr;
 253        size_t alloc;
 254        struct line *line_allocated;
 255        struct line *line;
 256};
 257
 258/*
 259 * Records filenames that have been touched, in order to handle
 260 * the case where more than one patches touch the same file.
 261 */
 262
 263static struct string_list fn_table;
 264
 265static uint32_t hash_line(const char *cp, size_t len)
 266{
 267        size_t i;
 268        uint32_t h;
 269        for (i = 0, h = 0; i < len; i++) {
 270                if (!isspace(cp[i])) {
 271                        h = h * 3 + (cp[i] & 0xff);
 272                }
 273        }
 274        return h;
 275}
 276
 277/*
 278 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 279 * whitespace difference. Returns 1 if they match, 0 otherwise
 280 */
 281static int fuzzy_matchlines(const char *s1, size_t n1,
 282                            const char *s2, size_t n2)
 283{
 284        const char *last1 = s1 + n1 - 1;
 285        const char *last2 = s2 + n2 - 1;
 286        int result = 0;
 287
 288        /* ignore line endings */
 289        while ((*last1 == '\r') || (*last1 == '\n'))
 290                last1--;
 291        while ((*last2 == '\r') || (*last2 == '\n'))
 292                last2--;
 293
 294        /* skip leading whitespace */
 295        while (isspace(*s1) && (s1 <= last1))
 296                s1++;
 297        while (isspace(*s2) && (s2 <= last2))
 298                s2++;
 299        /* early return if both lines are empty */
 300        if ((s1 > last1) && (s2 > last2))
 301                return 1;
 302        while (!result) {
 303                result = *s1++ - *s2++;
 304                /*
 305                 * Skip whitespace inside. We check for whitespace on
 306                 * both buffers because we don't want "a b" to match
 307                 * "ab"
 308                 */
 309                if (isspace(*s1) && isspace(*s2)) {
 310                        while (isspace(*s1) && s1 <= last1)
 311                                s1++;
 312                        while (isspace(*s2) && s2 <= last2)
 313                                s2++;
 314                }
 315                /*
 316                 * If we reached the end on one side only,
 317                 * lines don't match
 318                 */
 319                if (
 320                    ((s2 > last2) && (s1 <= last1)) ||
 321                    ((s1 > last1) && (s2 <= last2)))
 322                        return 0;
 323                if ((s1 > last1) && (s2 > last2))
 324                        break;
 325        }
 326
 327        return !result;
 328}
 329
 330static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 331{
 332        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 333        img->line_allocated[img->nr].len = len;
 334        img->line_allocated[img->nr].hash = hash_line(bol, len);
 335        img->line_allocated[img->nr].flag = flag;
 336        img->nr++;
 337}
 338
 339/*
 340 * "buf" has the file contents to be patched (read from various sources).
 341 * attach it to "image" and add line-based index to it.
 342 * "image" now owns the "buf".
 343 */
 344static void prepare_image(struct image *image, char *buf, size_t len,
 345                          int prepare_linetable)
 346{
 347        const char *cp, *ep;
 348
 349        memset(image, 0, sizeof(*image));
 350        image->buf = buf;
 351        image->len = len;
 352
 353        if (!prepare_linetable)
 354                return;
 355
 356        ep = image->buf + image->len;
 357        cp = image->buf;
 358        while (cp < ep) {
 359                const char *next;
 360                for (next = cp; next < ep && *next != '\n'; next++)
 361                        ;
 362                if (next < ep)
 363                        next++;
 364                add_line_info(image, cp, next - cp, 0);
 365                cp = next;
 366        }
 367        image->line = image->line_allocated;
 368}
 369
 370static void clear_image(struct image *image)
 371{
 372        free(image->buf);
 373        image->buf = NULL;
 374        image->len = 0;
 375}
 376
 377/* fmt must contain _one_ %s and no other substitution */
 378static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 379{
 380        struct strbuf sb = STRBUF_INIT;
 381
 382        if (patch->old_name && patch->new_name &&
 383            strcmp(patch->old_name, patch->new_name)) {
 384                quote_c_style(patch->old_name, &sb, NULL, 0);
 385                strbuf_addstr(&sb, " => ");
 386                quote_c_style(patch->new_name, &sb, NULL, 0);
 387        } else {
 388                const char *n = patch->new_name;
 389                if (!n)
 390                        n = patch->old_name;
 391                quote_c_style(n, &sb, NULL, 0);
 392        }
 393        fprintf(output, fmt, sb.buf);
 394        fputc('\n', output);
 395        strbuf_release(&sb);
 396}
 397
 398#define SLOP (16)
 399
 400static void read_patch_file(struct strbuf *sb, int fd)
 401{
 402        if (strbuf_read(sb, fd, 0) < 0)
 403                die_errno("git apply: failed to read");
 404
 405        /*
 406         * Make sure that we have some slop in the buffer
 407         * so that we can do speculative "memcmp" etc, and
 408         * see to it that it is NUL-filled.
 409         */
 410        strbuf_grow(sb, SLOP);
 411        memset(sb->buf + sb->len, 0, SLOP);
 412}
 413
 414static unsigned long linelen(const char *buffer, unsigned long size)
 415{
 416        unsigned long len = 0;
 417        while (size--) {
 418                len++;
 419                if (*buffer++ == '\n')
 420                        break;
 421        }
 422        return len;
 423}
 424
 425static int is_dev_null(const char *str)
 426{
 427        return !memcmp("/dev/null", str, 9) && isspace(str[9]);
 428}
 429
 430#define TERM_SPACE      1
 431#define TERM_TAB        2
 432
 433static int name_terminate(const char *name, int namelen, int c, int terminate)
 434{
 435        if (c == ' ' && !(terminate & TERM_SPACE))
 436                return 0;
 437        if (c == '\t' && !(terminate & TERM_TAB))
 438                return 0;
 439
 440        return 1;
 441}
 442
 443/* remove double slashes to make --index work with such filenames */
 444static char *squash_slash(char *name)
 445{
 446        int i = 0, j = 0;
 447
 448        if (!name)
 449                return NULL;
 450
 451        while (name[i]) {
 452                if ((name[j++] = name[i++]) == '/')
 453                        while (name[i] == '/')
 454                                i++;
 455        }
 456        name[j] = '\0';
 457        return name;
 458}
 459
 460static char *find_name_gnu(const char *line, const char *def, int p_value)
 461{
 462        struct strbuf name = STRBUF_INIT;
 463        char *cp;
 464
 465        /*
 466         * Proposed "new-style" GNU patch/diff format; see
 467         * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
 468         */
 469        if (unquote_c_style(&name, line, NULL)) {
 470                strbuf_release(&name);
 471                return NULL;
 472        }
 473
 474        for (cp = name.buf; p_value; p_value--) {
 475                cp = strchr(cp, '/');
 476                if (!cp) {
 477                        strbuf_release(&name);
 478                        return NULL;
 479                }
 480                cp++;
 481        }
 482
 483        strbuf_remove(&name, 0, cp - name.buf);
 484        if (root)
 485                strbuf_insert(&name, 0, root, root_len);
 486        return squash_slash(strbuf_detach(&name, NULL));
 487}
 488
 489static size_t sane_tz_len(const char *line, size_t len)
 490{
 491        const char *tz, *p;
 492
 493        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 494                return 0;
 495        tz = line + len - strlen(" +0500");
 496
 497        if (tz[1] != '+' && tz[1] != '-')
 498                return 0;
 499
 500        for (p = tz + 2; p != line + len; p++)
 501                if (!isdigit(*p))
 502                        return 0;
 503
 504        return line + len - tz;
 505}
 506
 507static size_t tz_with_colon_len(const char *line, size_t len)
 508{
 509        const char *tz, *p;
 510
 511        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 512                return 0;
 513        tz = line + len - strlen(" +08:00");
 514
 515        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 516                return 0;
 517        p = tz + 2;
 518        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 519            !isdigit(*p++) || !isdigit(*p++))
 520                return 0;
 521
 522        return line + len - tz;
 523}
 524
 525static size_t date_len(const char *line, size_t len)
 526{
 527        const char *date, *p;
 528
 529        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 530                return 0;
 531        p = date = line + len - strlen("72-02-05");
 532
 533        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 534            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 535            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 536                return 0;
 537
 538        if (date - line >= strlen("19") &&
 539            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 540                date -= strlen("19");
 541
 542        return line + len - date;
 543}
 544
 545static size_t short_time_len(const char *line, size_t len)
 546{
 547        const char *time, *p;
 548
 549        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 550                return 0;
 551        p = time = line + len - strlen(" 07:01:32");
 552
 553        /* Permit 1-digit hours? */
 554        if (*p++ != ' ' ||
 555            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 556            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 557            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 558                return 0;
 559
 560        return line + len - time;
 561}
 562
 563static size_t fractional_time_len(const char *line, size_t len)
 564{
 565        const char *p;
 566        size_t n;
 567
 568        /* Expected format: 19:41:17.620000023 */
 569        if (!len || !isdigit(line[len - 1]))
 570                return 0;
 571        p = line + len - 1;
 572
 573        /* Fractional seconds. */
 574        while (p > line && isdigit(*p))
 575                p--;
 576        if (*p != '.')
 577                return 0;
 578
 579        /* Hours, minutes, and whole seconds. */
 580        n = short_time_len(line, p - line);
 581        if (!n)
 582                return 0;
 583
 584        return line + len - p + n;
 585}
 586
 587static size_t trailing_spaces_len(const char *line, size_t len)
 588{
 589        const char *p;
 590
 591        /* Expected format: ' ' x (1 or more)  */
 592        if (!len || line[len - 1] != ' ')
 593                return 0;
 594
 595        p = line + len;
 596        while (p != line) {
 597                p--;
 598                if (*p != ' ')
 599                        return line + len - (p + 1);
 600        }
 601
 602        /* All spaces! */
 603        return len;
 604}
 605
 606static size_t diff_timestamp_len(const char *line, size_t len)
 607{
 608        const char *end = line + len;
 609        size_t n;
 610
 611        /*
 612         * Posix: 2010-07-05 19:41:17
 613         * GNU: 2010-07-05 19:41:17.620000023 -0500
 614         */
 615
 616        if (!isdigit(end[-1]))
 617                return 0;
 618
 619        n = sane_tz_len(line, end - line);
 620        if (!n)
 621                n = tz_with_colon_len(line, end - line);
 622        end -= n;
 623
 624        n = short_time_len(line, end - line);
 625        if (!n)
 626                n = fractional_time_len(line, end - line);
 627        end -= n;
 628
 629        n = date_len(line, end - line);
 630        if (!n) /* No date.  Too bad. */
 631                return 0;
 632        end -= n;
 633
 634        if (end == line)        /* No space before date. */
 635                return 0;
 636        if (end[-1] == '\t') {  /* Success! */
 637                end--;
 638                return line + len - end;
 639        }
 640        if (end[-1] != ' ')     /* No space before date. */
 641                return 0;
 642
 643        /* Whitespace damage. */
 644        end -= trailing_spaces_len(line, end - line);
 645        return line + len - end;
 646}
 647
 648static char *null_strdup(const char *s)
 649{
 650        return s ? xstrdup(s) : NULL;
 651}
 652
 653static char *find_name_common(const char *line, const char *def,
 654                              int p_value, const char *end, int terminate)
 655{
 656        int len;
 657        const char *start = NULL;
 658
 659        if (p_value == 0)
 660                start = line;
 661        while (line != end) {
 662                char c = *line;
 663
 664                if (!end && isspace(c)) {
 665                        if (c == '\n')
 666                                break;
 667                        if (name_terminate(start, line-start, c, terminate))
 668                                break;
 669                }
 670                line++;
 671                if (c == '/' && !--p_value)
 672                        start = line;
 673        }
 674        if (!start)
 675                return squash_slash(null_strdup(def));
 676        len = line - start;
 677        if (!len)
 678                return squash_slash(null_strdup(def));
 679
 680        /*
 681         * Generally we prefer the shorter name, especially
 682         * if the other one is just a variation of that with
 683         * something else tacked on to the end (ie "file.orig"
 684         * or "file~").
 685         */
 686        if (def) {
 687                int deflen = strlen(def);
 688                if (deflen < len && !strncmp(start, def, deflen))
 689                        return squash_slash(xstrdup(def));
 690        }
 691
 692        if (root) {
 693                char *ret = xmalloc(root_len + len + 1);
 694                strcpy(ret, root);
 695                memcpy(ret + root_len, start, len);
 696                ret[root_len + len] = '\0';
 697                return squash_slash(ret);
 698        }
 699
 700        return squash_slash(xmemdupz(start, len));
 701}
 702
 703static char *find_name(const char *line, char *def, int p_value, int terminate)
 704{
 705        if (*line == '"') {
 706                char *name = find_name_gnu(line, def, p_value);
 707                if (name)
 708                        return name;
 709        }
 710
 711        return find_name_common(line, def, p_value, NULL, terminate);
 712}
 713
 714static char *find_name_traditional(const char *line, char *def, int p_value)
 715{
 716        size_t len = strlen(line);
 717        size_t date_len;
 718
 719        if (*line == '"') {
 720                char *name = find_name_gnu(line, def, p_value);
 721                if (name)
 722                        return name;
 723        }
 724
 725        len = strchrnul(line, '\n') - line;
 726        date_len = diff_timestamp_len(line, len);
 727        if (!date_len)
 728                return find_name_common(line, def, p_value, NULL, TERM_TAB);
 729        len -= date_len;
 730
 731        return find_name_common(line, def, p_value, line + len, 0);
 732}
 733
 734static int count_slashes(const char *cp)
 735{
 736        int cnt = 0;
 737        char ch;
 738
 739        while ((ch = *cp++))
 740                if (ch == '/')
 741                        cnt++;
 742        return cnt;
 743}
 744
 745/*
 746 * Given the string after "--- " or "+++ ", guess the appropriate
 747 * p_value for the given patch.
 748 */
 749static int guess_p_value(const char *nameline)
 750{
 751        char *name, *cp;
 752        int val = -1;
 753
 754        if (is_dev_null(nameline))
 755                return -1;
 756        name = find_name_traditional(nameline, NULL, 0);
 757        if (!name)
 758                return -1;
 759        cp = strchr(name, '/');
 760        if (!cp)
 761                val = 0;
 762        else if (prefix) {
 763                /*
 764                 * Does it begin with "a/$our-prefix" and such?  Then this is
 765                 * very likely to apply to our directory.
 766                 */
 767                if (!strncmp(name, prefix, prefix_length))
 768                        val = count_slashes(prefix);
 769                else {
 770                        cp++;
 771                        if (!strncmp(cp, prefix, prefix_length))
 772                                val = count_slashes(prefix) + 1;
 773                }
 774        }
 775        free(name);
 776        return val;
 777}
 778
 779/*
 780 * Does the ---/+++ line has the POSIX timestamp after the last HT?
 781 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 782 * this such a timestamp?
 783 */
 784static int has_epoch_timestamp(const char *nameline)
 785{
 786        /*
 787         * We are only interested in epoch timestamp; any non-zero
 788         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 789         * For the same reason, the date must be either 1969-12-31 or
 790         * 1970-01-01, and the seconds part must be "00".
 791         */
 792        const char stamp_regexp[] =
 793                "^(1969-12-31|1970-01-01)"
 794                " "
 795                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 796                " "
 797                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 798        const char *timestamp = NULL, *cp, *colon;
 799        static regex_t *stamp;
 800        regmatch_t m[10];
 801        int zoneoffset;
 802        int hourminute;
 803        int status;
 804
 805        for (cp = nameline; *cp != '\n'; cp++) {
 806                if (*cp == '\t')
 807                        timestamp = cp + 1;
 808        }
 809        if (!timestamp)
 810                return 0;
 811        if (!stamp) {
 812                stamp = xmalloc(sizeof(*stamp));
 813                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 814                        warning(_("Cannot prepare timestamp regexp %s"),
 815                                stamp_regexp);
 816                        return 0;
 817                }
 818        }
 819
 820        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 821        if (status) {
 822                if (status != REG_NOMATCH)
 823                        warning(_("regexec returned %d for input: %s"),
 824                                status, timestamp);
 825                return 0;
 826        }
 827
 828        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 829        if (*colon == ':')
 830                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 831        else
 832                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 833        if (timestamp[m[3].rm_so] == '-')
 834                zoneoffset = -zoneoffset;
 835
 836        /*
 837         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 838         * (west of GMT) or 1970-01-01 (east of GMT)
 839         */
 840        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 841            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 842                return 0;
 843
 844        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 845                      strtol(timestamp + 14, NULL, 10) -
 846                      zoneoffset);
 847
 848        return ((zoneoffset < 0 && hourminute == 1440) ||
 849                (0 <= zoneoffset && !hourminute));
 850}
 851
 852/*
 853 * Get the name etc info from the ---/+++ lines of a traditional patch header
 854 *
 855 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 856 * files, we can happily check the index for a match, but for creating a
 857 * new file we should try to match whatever "patch" does. I have no idea.
 858 */
 859static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 860{
 861        char *name;
 862
 863        first += 4;     /* skip "--- " */
 864        second += 4;    /* skip "+++ " */
 865        if (!p_value_known) {
 866                int p, q;
 867                p = guess_p_value(first);
 868                q = guess_p_value(second);
 869                if (p < 0) p = q;
 870                if (0 <= p && p == q) {
 871                        p_value = p;
 872                        p_value_known = 1;
 873                }
 874        }
 875        if (is_dev_null(first)) {
 876                patch->is_new = 1;
 877                patch->is_delete = 0;
 878                name = find_name_traditional(second, NULL, p_value);
 879                patch->new_name = name;
 880        } else if (is_dev_null(second)) {
 881                patch->is_new = 0;
 882                patch->is_delete = 1;
 883                name = find_name_traditional(first, NULL, p_value);
 884                patch->old_name = name;
 885        } else {
 886                char *first_name;
 887                first_name = find_name_traditional(first, NULL, p_value);
 888                name = find_name_traditional(second, first_name, p_value);
 889                free(first_name);
 890                if (has_epoch_timestamp(first)) {
 891                        patch->is_new = 1;
 892                        patch->is_delete = 0;
 893                        patch->new_name = name;
 894                } else if (has_epoch_timestamp(second)) {
 895                        patch->is_new = 0;
 896                        patch->is_delete = 1;
 897                        patch->old_name = name;
 898                } else {
 899                        patch->old_name = name;
 900                        patch->new_name = xstrdup(name);
 901                }
 902        }
 903        if (!name)
 904                die(_("unable to find filename in patch at line %d"), linenr);
 905}
 906
 907static int gitdiff_hdrend(const char *line, struct patch *patch)
 908{
 909        return -1;
 910}
 911
 912/*
 913 * We're anal about diff header consistency, to make
 914 * sure that we don't end up having strange ambiguous
 915 * patches floating around.
 916 *
 917 * As a result, gitdiff_{old|new}name() will check
 918 * their names against any previous information, just
 919 * to make sure..
 920 */
 921#define DIFF_OLD_NAME 0
 922#define DIFF_NEW_NAME 1
 923
 924static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
 925{
 926        if (!orig_name && !isnull)
 927                return find_name(line, NULL, p_value, TERM_TAB);
 928
 929        if (orig_name) {
 930                int len;
 931                const char *name;
 932                char *another;
 933                name = orig_name;
 934                len = strlen(name);
 935                if (isnull)
 936                        die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
 937                another = find_name(line, NULL, p_value, TERM_TAB);
 938                if (!another || memcmp(another, name, len + 1))
 939                        die((side == DIFF_NEW_NAME) ?
 940                            _("git apply: bad git-diff - inconsistent new filename on line %d") :
 941                            _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
 942                free(another);
 943                return orig_name;
 944        }
 945        else {
 946                /* expect "/dev/null" */
 947                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 948                        die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
 949                return NULL;
 950        }
 951}
 952
 953static int gitdiff_oldname(const char *line, struct patch *patch)
 954{
 955        char *orig = patch->old_name;
 956        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
 957                                              DIFF_OLD_NAME);
 958        if (orig != patch->old_name)
 959                free(orig);
 960        return 0;
 961}
 962
 963static int gitdiff_newname(const char *line, struct patch *patch)
 964{
 965        char *orig = patch->new_name;
 966        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
 967                                              DIFF_NEW_NAME);
 968        if (orig != patch->new_name)
 969                free(orig);
 970        return 0;
 971}
 972
 973static int gitdiff_oldmode(const char *line, struct patch *patch)
 974{
 975        patch->old_mode = strtoul(line, NULL, 8);
 976        return 0;
 977}
 978
 979static int gitdiff_newmode(const char *line, struct patch *patch)
 980{
 981        patch->new_mode = strtoul(line, NULL, 8);
 982        return 0;
 983}
 984
 985static int gitdiff_delete(const char *line, struct patch *patch)
 986{
 987        patch->is_delete = 1;
 988        free(patch->old_name);
 989        patch->old_name = null_strdup(patch->def_name);
 990        return gitdiff_oldmode(line, patch);
 991}
 992
 993static int gitdiff_newfile(const char *line, struct patch *patch)
 994{
 995        patch->is_new = 1;
 996        free(patch->new_name);
 997        patch->new_name = null_strdup(patch->def_name);
 998        return gitdiff_newmode(line, patch);
 999}
1000
1001static int gitdiff_copysrc(const char *line, struct patch *patch)
1002{
1003        patch->is_copy = 1;
1004        free(patch->old_name);
1005        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1006        return 0;
1007}
1008
1009static int gitdiff_copydst(const char *line, struct patch *patch)
1010{
1011        patch->is_copy = 1;
1012        free(patch->new_name);
1013        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1014        return 0;
1015}
1016
1017static int gitdiff_renamesrc(const char *line, struct patch *patch)
1018{
1019        patch->is_rename = 1;
1020        free(patch->old_name);
1021        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1022        return 0;
1023}
1024
1025static int gitdiff_renamedst(const char *line, struct patch *patch)
1026{
1027        patch->is_rename = 1;
1028        free(patch->new_name);
1029        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1030        return 0;
1031}
1032
1033static int gitdiff_similarity(const char *line, struct patch *patch)
1034{
1035        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1036                patch->score = 0;
1037        return 0;
1038}
1039
1040static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1041{
1042        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1043                patch->score = 0;
1044        return 0;
1045}
1046
1047static int gitdiff_index(const char *line, struct patch *patch)
1048{
1049        /*
1050         * index line is N hexadecimal, "..", N hexadecimal,
1051         * and optional space with octal mode.
1052         */
1053        const char *ptr, *eol;
1054        int len;
1055
1056        ptr = strchr(line, '.');
1057        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1058                return 0;
1059        len = ptr - line;
1060        memcpy(patch->old_sha1_prefix, line, len);
1061        patch->old_sha1_prefix[len] = 0;
1062
1063        line = ptr + 2;
1064        ptr = strchr(line, ' ');
1065        eol = strchr(line, '\n');
1066
1067        if (!ptr || eol < ptr)
1068                ptr = eol;
1069        len = ptr - line;
1070
1071        if (40 < len)
1072                return 0;
1073        memcpy(patch->new_sha1_prefix, line, len);
1074        patch->new_sha1_prefix[len] = 0;
1075        if (*ptr == ' ')
1076                patch->old_mode = strtoul(ptr+1, NULL, 8);
1077        return 0;
1078}
1079
1080/*
1081 * This is normal for a diff that doesn't change anything: we'll fall through
1082 * into the next diff. Tell the parser to break out.
1083 */
1084static int gitdiff_unrecognized(const char *line, struct patch *patch)
1085{
1086        return -1;
1087}
1088
1089/*
1090 * Skip p_value leading components from "line"; as we do not accept
1091 * absolute paths, return NULL in that case.
1092 */
1093static const char *skip_tree_prefix(const char *line, int llen)
1094{
1095        int nslash;
1096        int i;
1097
1098        if (!p_value)
1099                return (llen && line[0] == '/') ? NULL : line;
1100
1101        nslash = p_value;
1102        for (i = 0; i < llen; i++) {
1103                int ch = line[i];
1104                if (ch == '/' && --nslash <= 0)
1105                        return (i == 0) ? NULL : &line[i + 1];
1106        }
1107        return NULL;
1108}
1109
1110/*
1111 * This is to extract the same name that appears on "diff --git"
1112 * line.  We do not find and return anything if it is a rename
1113 * patch, and it is OK because we will find the name elsewhere.
1114 * We need to reliably find name only when it is mode-change only,
1115 * creation or deletion of an empty file.  In any of these cases,
1116 * both sides are the same name under a/ and b/ respectively.
1117 */
1118static char *git_header_name(const char *line, int llen)
1119{
1120        const char *name;
1121        const char *second = NULL;
1122        size_t len, line_len;
1123
1124        line += strlen("diff --git ");
1125        llen -= strlen("diff --git ");
1126
1127        if (*line == '"') {
1128                const char *cp;
1129                struct strbuf first = STRBUF_INIT;
1130                struct strbuf sp = STRBUF_INIT;
1131
1132                if (unquote_c_style(&first, line, &second))
1133                        goto free_and_fail1;
1134
1135                /* strip the a/b prefix including trailing slash */
1136                cp = skip_tree_prefix(first.buf, first.len);
1137                if (!cp)
1138                        goto free_and_fail1;
1139                strbuf_remove(&first, 0, cp - first.buf);
1140
1141                /*
1142                 * second points at one past closing dq of name.
1143                 * find the second name.
1144                 */
1145                while ((second < line + llen) && isspace(*second))
1146                        second++;
1147
1148                if (line + llen <= second)
1149                        goto free_and_fail1;
1150                if (*second == '"') {
1151                        if (unquote_c_style(&sp, second, NULL))
1152                                goto free_and_fail1;
1153                        cp = skip_tree_prefix(sp.buf, sp.len);
1154                        if (!cp)
1155                                goto free_and_fail1;
1156                        /* They must match, otherwise ignore */
1157                        if (strcmp(cp, first.buf))
1158                                goto free_and_fail1;
1159                        strbuf_release(&sp);
1160                        return strbuf_detach(&first, NULL);
1161                }
1162
1163                /* unquoted second */
1164                cp = skip_tree_prefix(second, line + llen - second);
1165                if (!cp)
1166                        goto free_and_fail1;
1167                if (line + llen - cp != first.len ||
1168                    memcmp(first.buf, cp, first.len))
1169                        goto free_and_fail1;
1170                return strbuf_detach(&first, NULL);
1171
1172        free_and_fail1:
1173                strbuf_release(&first);
1174                strbuf_release(&sp);
1175                return NULL;
1176        }
1177
1178        /* unquoted first name */
1179        name = skip_tree_prefix(line, llen);
1180        if (!name)
1181                return NULL;
1182
1183        /*
1184         * since the first name is unquoted, a dq if exists must be
1185         * the beginning of the second name.
1186         */
1187        for (second = name; second < line + llen; second++) {
1188                if (*second == '"') {
1189                        struct strbuf sp = STRBUF_INIT;
1190                        const char *np;
1191
1192                        if (unquote_c_style(&sp, second, NULL))
1193                                goto free_and_fail2;
1194
1195                        np = skip_tree_prefix(sp.buf, sp.len);
1196                        if (!np)
1197                                goto free_and_fail2;
1198
1199                        len = sp.buf + sp.len - np;
1200                        if (len < second - name &&
1201                            !strncmp(np, name, len) &&
1202                            isspace(name[len])) {
1203                                /* Good */
1204                                strbuf_remove(&sp, 0, np - sp.buf);
1205                                return strbuf_detach(&sp, NULL);
1206                        }
1207
1208                free_and_fail2:
1209                        strbuf_release(&sp);
1210                        return NULL;
1211                }
1212        }
1213
1214        /*
1215         * Accept a name only if it shows up twice, exactly the same
1216         * form.
1217         */
1218        second = strchr(name, '\n');
1219        if (!second)
1220                return NULL;
1221        line_len = second - name;
1222        for (len = 0 ; ; len++) {
1223                switch (name[len]) {
1224                default:
1225                        continue;
1226                case '\n':
1227                        return NULL;
1228                case '\t': case ' ':
1229                        /*
1230                         * Is this the separator between the preimage
1231                         * and the postimage pathname?  Again, we are
1232                         * only interested in the case where there is
1233                         * no rename, as this is only to set def_name
1234                         * and a rename patch has the names elsewhere
1235                         * in an unambiguous form.
1236                         */
1237                        if (!name[len + 1])
1238                                return NULL; /* no postimage name */
1239                        second = skip_tree_prefix(name + len + 1,
1240                                                  line_len - (len + 1));
1241                        if (!second)
1242                                return NULL;
1243                        /*
1244                         * Does len bytes starting at "name" and "second"
1245                         * (that are separated by one HT or SP we just
1246                         * found) exactly match?
1247                         */
1248                        if (second[len] == '\n' && !strncmp(name, second, len))
1249                                return xmemdupz(name, len);
1250                }
1251        }
1252}
1253
1254/* Verify that we recognize the lines following a git header */
1255static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1256{
1257        unsigned long offset;
1258
1259        /* A git diff has explicit new/delete information, so we don't guess */
1260        patch->is_new = 0;
1261        patch->is_delete = 0;
1262
1263        /*
1264         * Some things may not have the old name in the
1265         * rest of the headers anywhere (pure mode changes,
1266         * or removing or adding empty files), so we get
1267         * the default name from the header.
1268         */
1269        patch->def_name = git_header_name(line, len);
1270        if (patch->def_name && root) {
1271                char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1272                strcpy(s, root);
1273                strcpy(s + root_len, patch->def_name);
1274                free(patch->def_name);
1275                patch->def_name = s;
1276        }
1277
1278        line += len;
1279        size -= len;
1280        linenr++;
1281        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1282                static const struct opentry {
1283                        const char *str;
1284                        int (*fn)(const char *, struct patch *);
1285                } optable[] = {
1286                        { "@@ -", gitdiff_hdrend },
1287                        { "--- ", gitdiff_oldname },
1288                        { "+++ ", gitdiff_newname },
1289                        { "old mode ", gitdiff_oldmode },
1290                        { "new mode ", gitdiff_newmode },
1291                        { "deleted file mode ", gitdiff_delete },
1292                        { "new file mode ", gitdiff_newfile },
1293                        { "copy from ", gitdiff_copysrc },
1294                        { "copy to ", gitdiff_copydst },
1295                        { "rename old ", gitdiff_renamesrc },
1296                        { "rename new ", gitdiff_renamedst },
1297                        { "rename from ", gitdiff_renamesrc },
1298                        { "rename to ", gitdiff_renamedst },
1299                        { "similarity index ", gitdiff_similarity },
1300                        { "dissimilarity index ", gitdiff_dissimilarity },
1301                        { "index ", gitdiff_index },
1302                        { "", gitdiff_unrecognized },
1303                };
1304                int i;
1305
1306                len = linelen(line, size);
1307                if (!len || line[len-1] != '\n')
1308                        break;
1309                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1310                        const struct opentry *p = optable + i;
1311                        int oplen = strlen(p->str);
1312                        if (len < oplen || memcmp(p->str, line, oplen))
1313                                continue;
1314                        if (p->fn(line + oplen, patch) < 0)
1315                                return offset;
1316                        break;
1317                }
1318        }
1319
1320        return offset;
1321}
1322
1323static int parse_num(const char *line, unsigned long *p)
1324{
1325        char *ptr;
1326
1327        if (!isdigit(*line))
1328                return 0;
1329        *p = strtoul(line, &ptr, 10);
1330        return ptr - line;
1331}
1332
1333static int parse_range(const char *line, int len, int offset, const char *expect,
1334                       unsigned long *p1, unsigned long *p2)
1335{
1336        int digits, ex;
1337
1338        if (offset < 0 || offset >= len)
1339                return -1;
1340        line += offset;
1341        len -= offset;
1342
1343        digits = parse_num(line, p1);
1344        if (!digits)
1345                return -1;
1346
1347        offset += digits;
1348        line += digits;
1349        len -= digits;
1350
1351        *p2 = 1;
1352        if (*line == ',') {
1353                digits = parse_num(line+1, p2);
1354                if (!digits)
1355                        return -1;
1356
1357                offset += digits+1;
1358                line += digits+1;
1359                len -= digits+1;
1360        }
1361
1362        ex = strlen(expect);
1363        if (ex > len)
1364                return -1;
1365        if (memcmp(line, expect, ex))
1366                return -1;
1367
1368        return offset + ex;
1369}
1370
1371static void recount_diff(const char *line, int size, struct fragment *fragment)
1372{
1373        int oldlines = 0, newlines = 0, ret = 0;
1374
1375        if (size < 1) {
1376                warning("recount: ignore empty hunk");
1377                return;
1378        }
1379
1380        for (;;) {
1381                int len = linelen(line, size);
1382                size -= len;
1383                line += len;
1384
1385                if (size < 1)
1386                        break;
1387
1388                switch (*line) {
1389                case ' ': case '\n':
1390                        newlines++;
1391                        /* fall through */
1392                case '-':
1393                        oldlines++;
1394                        continue;
1395                case '+':
1396                        newlines++;
1397                        continue;
1398                case '\\':
1399                        continue;
1400                case '@':
1401                        ret = size < 3 || prefixcmp(line, "@@ ");
1402                        break;
1403                case 'd':
1404                        ret = size < 5 || prefixcmp(line, "diff ");
1405                        break;
1406                default:
1407                        ret = -1;
1408                        break;
1409                }
1410                if (ret) {
1411                        warning(_("recount: unexpected line: %.*s"),
1412                                (int)linelen(line, size), line);
1413                        return;
1414                }
1415                break;
1416        }
1417        fragment->oldlines = oldlines;
1418        fragment->newlines = newlines;
1419}
1420
1421/*
1422 * Parse a unified diff fragment header of the
1423 * form "@@ -a,b +c,d @@"
1424 */
1425static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1426{
1427        int offset;
1428
1429        if (!len || line[len-1] != '\n')
1430                return -1;
1431
1432        /* Figure out the number of lines in a fragment */
1433        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1434        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1435
1436        return offset;
1437}
1438
1439static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1440{
1441        unsigned long offset, len;
1442
1443        patch->is_toplevel_relative = 0;
1444        patch->is_rename = patch->is_copy = 0;
1445        patch->is_new = patch->is_delete = -1;
1446        patch->old_mode = patch->new_mode = 0;
1447        patch->old_name = patch->new_name = NULL;
1448        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1449                unsigned long nextlen;
1450
1451                len = linelen(line, size);
1452                if (!len)
1453                        break;
1454
1455                /* Testing this early allows us to take a few shortcuts.. */
1456                if (len < 6)
1457                        continue;
1458
1459                /*
1460                 * Make sure we don't find any unconnected patch fragments.
1461                 * That's a sign that we didn't find a header, and that a
1462                 * patch has become corrupted/broken up.
1463                 */
1464                if (!memcmp("@@ -", line, 4)) {
1465                        struct fragment dummy;
1466                        if (parse_fragment_header(line, len, &dummy) < 0)
1467                                continue;
1468                        die(_("patch fragment without header at line %d: %.*s"),
1469                            linenr, (int)len-1, line);
1470                }
1471
1472                if (size < len + 6)
1473                        break;
1474
1475                /*
1476                 * Git patch? It might not have a real patch, just a rename
1477                 * or mode change, so we handle that specially
1478                 */
1479                if (!memcmp("diff --git ", line, 11)) {
1480                        int git_hdr_len = parse_git_header(line, len, size, patch);
1481                        if (git_hdr_len <= len)
1482                                continue;
1483                        if (!patch->old_name && !patch->new_name) {
1484                                if (!patch->def_name)
1485                                        die(Q_("git diff header lacks filename information when removing "
1486                                               "%d leading pathname component (line %d)",
1487                                               "git diff header lacks filename information when removing "
1488                                               "%d leading pathname components (line %d)",
1489                                               p_value),
1490                                            p_value, linenr);
1491                                patch->old_name = xstrdup(patch->def_name);
1492                                patch->new_name = xstrdup(patch->def_name);
1493                        }
1494                        if (!patch->is_delete && !patch->new_name)
1495                                die("git diff header lacks filename information "
1496                                    "(line %d)", linenr);
1497                        patch->is_toplevel_relative = 1;
1498                        *hdrsize = git_hdr_len;
1499                        return offset;
1500                }
1501
1502                /* --- followed by +++ ? */
1503                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1504                        continue;
1505
1506                /*
1507                 * We only accept unified patches, so we want it to
1508                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1509                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1510                 */
1511                nextlen = linelen(line + len, size - len);
1512                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1513                        continue;
1514
1515                /* Ok, we'll consider it a patch */
1516                parse_traditional_patch(line, line+len, patch);
1517                *hdrsize = len + nextlen;
1518                linenr += 2;
1519                return offset;
1520        }
1521        return -1;
1522}
1523
1524static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1525{
1526        char *err;
1527
1528        if (!result)
1529                return;
1530
1531        whitespace_error++;
1532        if (squelch_whitespace_errors &&
1533            squelch_whitespace_errors < whitespace_error)
1534                return;
1535
1536        err = whitespace_error_string(result);
1537        fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1538                patch_input_file, linenr, err, len, line);
1539        free(err);
1540}
1541
1542static void check_whitespace(const char *line, int len, unsigned ws_rule)
1543{
1544        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1545
1546        record_ws_error(result, line + 1, len - 2, linenr);
1547}
1548
1549/*
1550 * Parse a unified diff. Note that this really needs to parse each
1551 * fragment separately, since the only way to know the difference
1552 * between a "---" that is part of a patch, and a "---" that starts
1553 * the next patch is to look at the line counts..
1554 */
1555static int parse_fragment(const char *line, unsigned long size,
1556                          struct patch *patch, struct fragment *fragment)
1557{
1558        int added, deleted;
1559        int len = linelen(line, size), offset;
1560        unsigned long oldlines, newlines;
1561        unsigned long leading, trailing;
1562
1563        offset = parse_fragment_header(line, len, fragment);
1564        if (offset < 0)
1565                return -1;
1566        if (offset > 0 && patch->recount)
1567                recount_diff(line + offset, size - offset, fragment);
1568        oldlines = fragment->oldlines;
1569        newlines = fragment->newlines;
1570        leading = 0;
1571        trailing = 0;
1572
1573        /* Parse the thing.. */
1574        line += len;
1575        size -= len;
1576        linenr++;
1577        added = deleted = 0;
1578        for (offset = len;
1579             0 < size;
1580             offset += len, size -= len, line += len, linenr++) {
1581                if (!oldlines && !newlines)
1582                        break;
1583                len = linelen(line, size);
1584                if (!len || line[len-1] != '\n')
1585                        return -1;
1586                switch (*line) {
1587                default:
1588                        return -1;
1589                case '\n': /* newer GNU diff, an empty context line */
1590                case ' ':
1591                        oldlines--;
1592                        newlines--;
1593                        if (!deleted && !added)
1594                                leading++;
1595                        trailing++;
1596                        break;
1597                case '-':
1598                        if (apply_in_reverse &&
1599                            ws_error_action != nowarn_ws_error)
1600                                check_whitespace(line, len, patch->ws_rule);
1601                        deleted++;
1602                        oldlines--;
1603                        trailing = 0;
1604                        break;
1605                case '+':
1606                        if (!apply_in_reverse &&
1607                            ws_error_action != nowarn_ws_error)
1608                                check_whitespace(line, len, patch->ws_rule);
1609                        added++;
1610                        newlines--;
1611                        trailing = 0;
1612                        break;
1613
1614                /*
1615                 * We allow "\ No newline at end of file". Depending
1616                 * on locale settings when the patch was produced we
1617                 * don't know what this line looks like. The only
1618                 * thing we do know is that it begins with "\ ".
1619                 * Checking for 12 is just for sanity check -- any
1620                 * l10n of "\ No newline..." is at least that long.
1621                 */
1622                case '\\':
1623                        if (len < 12 || memcmp(line, "\\ ", 2))
1624                                return -1;
1625                        break;
1626                }
1627        }
1628        if (oldlines || newlines)
1629                return -1;
1630        fragment->leading = leading;
1631        fragment->trailing = trailing;
1632
1633        /*
1634         * If a fragment ends with an incomplete line, we failed to include
1635         * it in the above loop because we hit oldlines == newlines == 0
1636         * before seeing it.
1637         */
1638        if (12 < size && !memcmp(line, "\\ ", 2))
1639                offset += linelen(line, size);
1640
1641        patch->lines_added += added;
1642        patch->lines_deleted += deleted;
1643
1644        if (0 < patch->is_new && oldlines)
1645                return error(_("new file depends on old contents"));
1646        if (0 < patch->is_delete && newlines)
1647                return error(_("deleted file still has contents"));
1648        return offset;
1649}
1650
1651/*
1652 * We have seen "diff --git a/... b/..." header (or a traditional patch
1653 * header).  Read hunks that belong to this patch into fragments and hang
1654 * them to the given patch structure.
1655 *
1656 * The (fragment->patch, fragment->size) pair points into the memory given
1657 * by the caller, not a copy, when we return.
1658 */
1659static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1660{
1661        unsigned long offset = 0;
1662        unsigned long oldlines = 0, newlines = 0, context = 0;
1663        struct fragment **fragp = &patch->fragments;
1664
1665        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1666                struct fragment *fragment;
1667                int len;
1668
1669                fragment = xcalloc(1, sizeof(*fragment));
1670                fragment->linenr = linenr;
1671                len = parse_fragment(line, size, patch, fragment);
1672                if (len <= 0)
1673                        die(_("corrupt patch at line %d"), linenr);
1674                fragment->patch = line;
1675                fragment->size = len;
1676                oldlines += fragment->oldlines;
1677                newlines += fragment->newlines;
1678                context += fragment->leading + fragment->trailing;
1679
1680                *fragp = fragment;
1681                fragp = &fragment->next;
1682
1683                offset += len;
1684                line += len;
1685                size -= len;
1686        }
1687
1688        /*
1689         * If something was removed (i.e. we have old-lines) it cannot
1690         * be creation, and if something was added it cannot be
1691         * deletion.  However, the reverse is not true; --unified=0
1692         * patches that only add are not necessarily creation even
1693         * though they do not have any old lines, and ones that only
1694         * delete are not necessarily deletion.
1695         *
1696         * Unfortunately, a real creation/deletion patch do _not_ have
1697         * any context line by definition, so we cannot safely tell it
1698         * apart with --unified=0 insanity.  At least if the patch has
1699         * more than one hunk it is not creation or deletion.
1700         */
1701        if (patch->is_new < 0 &&
1702            (oldlines || (patch->fragments && patch->fragments->next)))
1703                patch->is_new = 0;
1704        if (patch->is_delete < 0 &&
1705            (newlines || (patch->fragments && patch->fragments->next)))
1706                patch->is_delete = 0;
1707
1708        if (0 < patch->is_new && oldlines)
1709                die(_("new file %s depends on old contents"), patch->new_name);
1710        if (0 < patch->is_delete && newlines)
1711                die(_("deleted file %s still has contents"), patch->old_name);
1712        if (!patch->is_delete && !newlines && context)
1713                fprintf_ln(stderr,
1714                           _("** warning: "
1715                             "file %s becomes empty but is not deleted"),
1716                           patch->new_name);
1717
1718        return offset;
1719}
1720
1721static inline int metadata_changes(struct patch *patch)
1722{
1723        return  patch->is_rename > 0 ||
1724                patch->is_copy > 0 ||
1725                patch->is_new > 0 ||
1726                patch->is_delete ||
1727                (patch->old_mode && patch->new_mode &&
1728                 patch->old_mode != patch->new_mode);
1729}
1730
1731static char *inflate_it(const void *data, unsigned long size,
1732                        unsigned long inflated_size)
1733{
1734        git_zstream stream;
1735        void *out;
1736        int st;
1737
1738        memset(&stream, 0, sizeof(stream));
1739
1740        stream.next_in = (unsigned char *)data;
1741        stream.avail_in = size;
1742        stream.next_out = out = xmalloc(inflated_size);
1743        stream.avail_out = inflated_size;
1744        git_inflate_init(&stream);
1745        st = git_inflate(&stream, Z_FINISH);
1746        git_inflate_end(&stream);
1747        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1748                free(out);
1749                return NULL;
1750        }
1751        return out;
1752}
1753
1754/*
1755 * Read a binary hunk and return a new fragment; fragment->patch
1756 * points at an allocated memory that the caller must free, so
1757 * it is marked as "->free_patch = 1".
1758 */
1759static struct fragment *parse_binary_hunk(char **buf_p,
1760                                          unsigned long *sz_p,
1761                                          int *status_p,
1762                                          int *used_p)
1763{
1764        /*
1765         * Expect a line that begins with binary patch method ("literal"
1766         * or "delta"), followed by the length of data before deflating.
1767         * a sequence of 'length-byte' followed by base-85 encoded data
1768         * should follow, terminated by a newline.
1769         *
1770         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1771         * and we would limit the patch line to 66 characters,
1772         * so one line can fit up to 13 groups that would decode
1773         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1774         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1775         */
1776        int llen, used;
1777        unsigned long size = *sz_p;
1778        char *buffer = *buf_p;
1779        int patch_method;
1780        unsigned long origlen;
1781        char *data = NULL;
1782        int hunk_size = 0;
1783        struct fragment *frag;
1784
1785        llen = linelen(buffer, size);
1786        used = llen;
1787
1788        *status_p = 0;
1789
1790        if (!prefixcmp(buffer, "delta ")) {
1791                patch_method = BINARY_DELTA_DEFLATED;
1792                origlen = strtoul(buffer + 6, NULL, 10);
1793        }
1794        else if (!prefixcmp(buffer, "literal ")) {
1795                patch_method = BINARY_LITERAL_DEFLATED;
1796                origlen = strtoul(buffer + 8, NULL, 10);
1797        }
1798        else
1799                return NULL;
1800
1801        linenr++;
1802        buffer += llen;
1803        while (1) {
1804                int byte_length, max_byte_length, newsize;
1805                llen = linelen(buffer, size);
1806                used += llen;
1807                linenr++;
1808                if (llen == 1) {
1809                        /* consume the blank line */
1810                        buffer++;
1811                        size--;
1812                        break;
1813                }
1814                /*
1815                 * Minimum line is "A00000\n" which is 7-byte long,
1816                 * and the line length must be multiple of 5 plus 2.
1817                 */
1818                if ((llen < 7) || (llen-2) % 5)
1819                        goto corrupt;
1820                max_byte_length = (llen - 2) / 5 * 4;
1821                byte_length = *buffer;
1822                if ('A' <= byte_length && byte_length <= 'Z')
1823                        byte_length = byte_length - 'A' + 1;
1824                else if ('a' <= byte_length && byte_length <= 'z')
1825                        byte_length = byte_length - 'a' + 27;
1826                else
1827                        goto corrupt;
1828                /* if the input length was not multiple of 4, we would
1829                 * have filler at the end but the filler should never
1830                 * exceed 3 bytes
1831                 */
1832                if (max_byte_length < byte_length ||
1833                    byte_length <= max_byte_length - 4)
1834                        goto corrupt;
1835                newsize = hunk_size + byte_length;
1836                data = xrealloc(data, newsize);
1837                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1838                        goto corrupt;
1839                hunk_size = newsize;
1840                buffer += llen;
1841                size -= llen;
1842        }
1843
1844        frag = xcalloc(1, sizeof(*frag));
1845        frag->patch = inflate_it(data, hunk_size, origlen);
1846        frag->free_patch = 1;
1847        if (!frag->patch)
1848                goto corrupt;
1849        free(data);
1850        frag->size = origlen;
1851        *buf_p = buffer;
1852        *sz_p = size;
1853        *used_p = used;
1854        frag->binary_patch_method = patch_method;
1855        return frag;
1856
1857 corrupt:
1858        free(data);
1859        *status_p = -1;
1860        error(_("corrupt binary patch at line %d: %.*s"),
1861              linenr-1, llen-1, buffer);
1862        return NULL;
1863}
1864
1865static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1866{
1867        /*
1868         * We have read "GIT binary patch\n"; what follows is a line
1869         * that says the patch method (currently, either "literal" or
1870         * "delta") and the length of data before deflating; a
1871         * sequence of 'length-byte' followed by base-85 encoded data
1872         * follows.
1873         *
1874         * When a binary patch is reversible, there is another binary
1875         * hunk in the same format, starting with patch method (either
1876         * "literal" or "delta") with the length of data, and a sequence
1877         * of length-byte + base-85 encoded data, terminated with another
1878         * empty line.  This data, when applied to the postimage, produces
1879         * the preimage.
1880         */
1881        struct fragment *forward;
1882        struct fragment *reverse;
1883        int status;
1884        int used, used_1;
1885
1886        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1887        if (!forward && !status)
1888                /* there has to be one hunk (forward hunk) */
1889                return error(_("unrecognized binary patch at line %d"), linenr-1);
1890        if (status)
1891                /* otherwise we already gave an error message */
1892                return status;
1893
1894        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1895        if (reverse)
1896                used += used_1;
1897        else if (status) {
1898                /*
1899                 * Not having reverse hunk is not an error, but having
1900                 * a corrupt reverse hunk is.
1901                 */
1902                free((void*) forward->patch);
1903                free(forward);
1904                return status;
1905        }
1906        forward->next = reverse;
1907        patch->fragments = forward;
1908        patch->is_binary = 1;
1909        return used;
1910}
1911
1912/*
1913 * Read the patch text in "buffer" taht extends for "size" bytes; stop
1914 * reading after seeing a single patch (i.e. changes to a single file).
1915 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1916 * Return the number of bytes consumed, so that the caller can call us
1917 * again for the next patch.
1918 */
1919static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1920{
1921        int hdrsize, patchsize;
1922        int offset = find_header(buffer, size, &hdrsize, patch);
1923
1924        if (offset < 0)
1925                return offset;
1926
1927        patch->ws_rule = whitespace_rule(patch->new_name
1928                                         ? patch->new_name
1929                                         : patch->old_name);
1930
1931        patchsize = parse_single_patch(buffer + offset + hdrsize,
1932                                       size - offset - hdrsize, patch);
1933
1934        if (!patchsize) {
1935                static const char *binhdr[] = {
1936                        "Binary files ",
1937                        "Files ",
1938                        NULL,
1939                };
1940                static const char git_binary[] = "GIT binary patch\n";
1941                int i;
1942                int hd = hdrsize + offset;
1943                unsigned long llen = linelen(buffer + hd, size - hd);
1944
1945                if (llen == sizeof(git_binary) - 1 &&
1946                    !memcmp(git_binary, buffer + hd, llen)) {
1947                        int used;
1948                        linenr++;
1949                        used = parse_binary(buffer + hd + llen,
1950                                            size - hd - llen, patch);
1951                        if (used)
1952                                patchsize = used + llen;
1953                        else
1954                                patchsize = 0;
1955                }
1956                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1957                        for (i = 0; binhdr[i]; i++) {
1958                                int len = strlen(binhdr[i]);
1959                                if (len < size - hd &&
1960                                    !memcmp(binhdr[i], buffer + hd, len)) {
1961                                        linenr++;
1962                                        patch->is_binary = 1;
1963                                        patchsize = llen;
1964                                        break;
1965                                }
1966                        }
1967                }
1968
1969                /* Empty patch cannot be applied if it is a text patch
1970                 * without metadata change.  A binary patch appears
1971                 * empty to us here.
1972                 */
1973                if ((apply || check) &&
1974                    (!patch->is_binary && !metadata_changes(patch)))
1975                        die(_("patch with only garbage at line %d"), linenr);
1976        }
1977
1978        return offset + hdrsize + patchsize;
1979}
1980
1981#define swap(a,b) myswap((a),(b),sizeof(a))
1982
1983#define myswap(a, b, size) do {         \
1984        unsigned char mytmp[size];      \
1985        memcpy(mytmp, &a, size);                \
1986        memcpy(&a, &b, size);           \
1987        memcpy(&b, mytmp, size);                \
1988} while (0)
1989
1990static void reverse_patches(struct patch *p)
1991{
1992        for (; p; p = p->next) {
1993                struct fragment *frag = p->fragments;
1994
1995                swap(p->new_name, p->old_name);
1996                swap(p->new_mode, p->old_mode);
1997                swap(p->is_new, p->is_delete);
1998                swap(p->lines_added, p->lines_deleted);
1999                swap(p->old_sha1_prefix, p->new_sha1_prefix);
2000
2001                for (; frag; frag = frag->next) {
2002                        swap(frag->newpos, frag->oldpos);
2003                        swap(frag->newlines, frag->oldlines);
2004                }
2005        }
2006}
2007
2008static const char pluses[] =
2009"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2010static const char minuses[]=
2011"----------------------------------------------------------------------";
2012
2013static void show_stats(struct patch *patch)
2014{
2015        struct strbuf qname = STRBUF_INIT;
2016        char *cp = patch->new_name ? patch->new_name : patch->old_name;
2017        int max, add, del;
2018
2019        quote_c_style(cp, &qname, NULL, 0);
2020
2021        /*
2022         * "scale" the filename
2023         */
2024        max = max_len;
2025        if (max > 50)
2026                max = 50;
2027
2028        if (qname.len > max) {
2029                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2030                if (!cp)
2031                        cp = qname.buf + qname.len + 3 - max;
2032                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2033        }
2034
2035        if (patch->is_binary) {
2036                printf(" %-*s |  Bin\n", max, qname.buf);
2037                strbuf_release(&qname);
2038                return;
2039        }
2040
2041        printf(" %-*s |", max, qname.buf);
2042        strbuf_release(&qname);
2043
2044        /*
2045         * scale the add/delete
2046         */
2047        max = max + max_change > 70 ? 70 - max : max_change;
2048        add = patch->lines_added;
2049        del = patch->lines_deleted;
2050
2051        if (max_change > 0) {
2052                int total = ((add + del) * max + max_change / 2) / max_change;
2053                add = (add * max + max_change / 2) / max_change;
2054                del = total - add;
2055        }
2056        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2057                add, pluses, del, minuses);
2058}
2059
2060static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2061{
2062        switch (st->st_mode & S_IFMT) {
2063        case S_IFLNK:
2064                if (strbuf_readlink(buf, path, st->st_size) < 0)
2065                        return error(_("unable to read symlink %s"), path);
2066                return 0;
2067        case S_IFREG:
2068                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2069                        return error(_("unable to open or read %s"), path);
2070                convert_to_git(path, buf->buf, buf->len, buf, 0);
2071                return 0;
2072        default:
2073                return -1;
2074        }
2075}
2076
2077/*
2078 * Update the preimage, and the common lines in postimage,
2079 * from buffer buf of length len. If postlen is 0 the postimage
2080 * is updated in place, otherwise it's updated on a new buffer
2081 * of length postlen
2082 */
2083
2084static void update_pre_post_images(struct image *preimage,
2085                                   struct image *postimage,
2086                                   char *buf,
2087                                   size_t len, size_t postlen)
2088{
2089        int i, ctx;
2090        char *new, *old, *fixed;
2091        struct image fixed_preimage;
2092
2093        /*
2094         * Update the preimage with whitespace fixes.  Note that we
2095         * are not losing preimage->buf -- apply_one_fragment() will
2096         * free "oldlines".
2097         */
2098        prepare_image(&fixed_preimage, buf, len, 1);
2099        assert(fixed_preimage.nr == preimage->nr);
2100        for (i = 0; i < preimage->nr; i++)
2101                fixed_preimage.line[i].flag = preimage->line[i].flag;
2102        free(preimage->line_allocated);
2103        *preimage = fixed_preimage;
2104
2105        /*
2106         * Adjust the common context lines in postimage. This can be
2107         * done in-place when we are just doing whitespace fixing,
2108         * which does not make the string grow, but needs a new buffer
2109         * when ignoring whitespace causes the update, since in this case
2110         * we could have e.g. tabs converted to multiple spaces.
2111         * We trust the caller to tell us if the update can be done
2112         * in place (postlen==0) or not.
2113         */
2114        old = postimage->buf;
2115        if (postlen)
2116                new = postimage->buf = xmalloc(postlen);
2117        else
2118                new = old;
2119        fixed = preimage->buf;
2120        for (i = ctx = 0; i < postimage->nr; i++) {
2121                size_t len = postimage->line[i].len;
2122                if (!(postimage->line[i].flag & LINE_COMMON)) {
2123                        /* an added line -- no counterparts in preimage */
2124                        memmove(new, old, len);
2125                        old += len;
2126                        new += len;
2127                        continue;
2128                }
2129
2130                /* a common context -- skip it in the original postimage */
2131                old += len;
2132
2133                /* and find the corresponding one in the fixed preimage */
2134                while (ctx < preimage->nr &&
2135                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2136                        fixed += preimage->line[ctx].len;
2137                        ctx++;
2138                }
2139                if (preimage->nr <= ctx)
2140                        die(_("oops"));
2141
2142                /* and copy it in, while fixing the line length */
2143                len = preimage->line[ctx].len;
2144                memcpy(new, fixed, len);
2145                new += len;
2146                fixed += len;
2147                postimage->line[i].len = len;
2148                ctx++;
2149        }
2150
2151        /* Fix the length of the whole thing */
2152        postimage->len = new - postimage->buf;
2153}
2154
2155static int match_fragment(struct image *img,
2156                          struct image *preimage,
2157                          struct image *postimage,
2158                          unsigned long try,
2159                          int try_lno,
2160                          unsigned ws_rule,
2161                          int match_beginning, int match_end)
2162{
2163        int i;
2164        char *fixed_buf, *buf, *orig, *target;
2165        struct strbuf fixed;
2166        size_t fixed_len;
2167        int preimage_limit;
2168
2169        if (preimage->nr + try_lno <= img->nr) {
2170                /*
2171                 * The hunk falls within the boundaries of img.
2172                 */
2173                preimage_limit = preimage->nr;
2174                if (match_end && (preimage->nr + try_lno != img->nr))
2175                        return 0;
2176        } else if (ws_error_action == correct_ws_error &&
2177                   (ws_rule & WS_BLANK_AT_EOF)) {
2178                /*
2179                 * This hunk extends beyond the end of img, and we are
2180                 * removing blank lines at the end of the file.  This
2181                 * many lines from the beginning of the preimage must
2182                 * match with img, and the remainder of the preimage
2183                 * must be blank.
2184                 */
2185                preimage_limit = img->nr - try_lno;
2186        } else {
2187                /*
2188                 * The hunk extends beyond the end of the img and
2189                 * we are not removing blanks at the end, so we
2190                 * should reject the hunk at this position.
2191                 */
2192                return 0;
2193        }
2194
2195        if (match_beginning && try_lno)
2196                return 0;
2197
2198        /* Quick hash check */
2199        for (i = 0; i < preimage_limit; i++)
2200                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2201                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2202                        return 0;
2203
2204        if (preimage_limit == preimage->nr) {
2205                /*
2206                 * Do we have an exact match?  If we were told to match
2207                 * at the end, size must be exactly at try+fragsize,
2208                 * otherwise try+fragsize must be still within the preimage,
2209                 * and either case, the old piece should match the preimage
2210                 * exactly.
2211                 */
2212                if ((match_end
2213                     ? (try + preimage->len == img->len)
2214                     : (try + preimage->len <= img->len)) &&
2215                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2216                        return 1;
2217        } else {
2218                /*
2219                 * The preimage extends beyond the end of img, so
2220                 * there cannot be an exact match.
2221                 *
2222                 * There must be one non-blank context line that match
2223                 * a line before the end of img.
2224                 */
2225                char *buf_end;
2226
2227                buf = preimage->buf;
2228                buf_end = buf;
2229                for (i = 0; i < preimage_limit; i++)
2230                        buf_end += preimage->line[i].len;
2231
2232                for ( ; buf < buf_end; buf++)
2233                        if (!isspace(*buf))
2234                                break;
2235                if (buf == buf_end)
2236                        return 0;
2237        }
2238
2239        /*
2240         * No exact match. If we are ignoring whitespace, run a line-by-line
2241         * fuzzy matching. We collect all the line length information because
2242         * we need it to adjust whitespace if we match.
2243         */
2244        if (ws_ignore_action == ignore_ws_change) {
2245                size_t imgoff = 0;
2246                size_t preoff = 0;
2247                size_t postlen = postimage->len;
2248                size_t extra_chars;
2249                char *preimage_eof;
2250                char *preimage_end;
2251                for (i = 0; i < preimage_limit; i++) {
2252                        size_t prelen = preimage->line[i].len;
2253                        size_t imglen = img->line[try_lno+i].len;
2254
2255                        if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2256                                              preimage->buf + preoff, prelen))
2257                                return 0;
2258                        if (preimage->line[i].flag & LINE_COMMON)
2259                                postlen += imglen - prelen;
2260                        imgoff += imglen;
2261                        preoff += prelen;
2262                }
2263
2264                /*
2265                 * Ok, the preimage matches with whitespace fuzz.
2266                 *
2267                 * imgoff now holds the true length of the target that
2268                 * matches the preimage before the end of the file.
2269                 *
2270                 * Count the number of characters in the preimage that fall
2271                 * beyond the end of the file and make sure that all of them
2272                 * are whitespace characters. (This can only happen if
2273                 * we are removing blank lines at the end of the file.)
2274                 */
2275                buf = preimage_eof = preimage->buf + preoff;
2276                for ( ; i < preimage->nr; i++)
2277                        preoff += preimage->line[i].len;
2278                preimage_end = preimage->buf + preoff;
2279                for ( ; buf < preimage_end; buf++)
2280                        if (!isspace(*buf))
2281                                return 0;
2282
2283                /*
2284                 * Update the preimage and the common postimage context
2285                 * lines to use the same whitespace as the target.
2286                 * If whitespace is missing in the target (i.e.
2287                 * if the preimage extends beyond the end of the file),
2288                 * use the whitespace from the preimage.
2289                 */
2290                extra_chars = preimage_end - preimage_eof;
2291                strbuf_init(&fixed, imgoff + extra_chars);
2292                strbuf_add(&fixed, img->buf + try, imgoff);
2293                strbuf_add(&fixed, preimage_eof, extra_chars);
2294                fixed_buf = strbuf_detach(&fixed, &fixed_len);
2295                update_pre_post_images(preimage, postimage,
2296                                fixed_buf, fixed_len, postlen);
2297                return 1;
2298        }
2299
2300        if (ws_error_action != correct_ws_error)
2301                return 0;
2302
2303        /*
2304         * The hunk does not apply byte-by-byte, but the hash says
2305         * it might with whitespace fuzz. We haven't been asked to
2306         * ignore whitespace, we were asked to correct whitespace
2307         * errors, so let's try matching after whitespace correction.
2308         *
2309         * The preimage may extend beyond the end of the file,
2310         * but in this loop we will only handle the part of the
2311         * preimage that falls within the file.
2312         */
2313        strbuf_init(&fixed, preimage->len + 1);
2314        orig = preimage->buf;
2315        target = img->buf + try;
2316        for (i = 0; i < preimage_limit; i++) {
2317                size_t oldlen = preimage->line[i].len;
2318                size_t tgtlen = img->line[try_lno + i].len;
2319                size_t fixstart = fixed.len;
2320                struct strbuf tgtfix;
2321                int match;
2322
2323                /* Try fixing the line in the preimage */
2324                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2325
2326                /* Try fixing the line in the target */
2327                strbuf_init(&tgtfix, tgtlen);
2328                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2329
2330                /*
2331                 * If they match, either the preimage was based on
2332                 * a version before our tree fixed whitespace breakage,
2333                 * or we are lacking a whitespace-fix patch the tree
2334                 * the preimage was based on already had (i.e. target
2335                 * has whitespace breakage, the preimage doesn't).
2336                 * In either case, we are fixing the whitespace breakages
2337                 * so we might as well take the fix together with their
2338                 * real change.
2339                 */
2340                match = (tgtfix.len == fixed.len - fixstart &&
2341                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2342                                             fixed.len - fixstart));
2343
2344                strbuf_release(&tgtfix);
2345                if (!match)
2346                        goto unmatch_exit;
2347
2348                orig += oldlen;
2349                target += tgtlen;
2350        }
2351
2352
2353        /*
2354         * Now handle the lines in the preimage that falls beyond the
2355         * end of the file (if any). They will only match if they are
2356         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2357         * false).
2358         */
2359        for ( ; i < preimage->nr; i++) {
2360                size_t fixstart = fixed.len; /* start of the fixed preimage */
2361                size_t oldlen = preimage->line[i].len;
2362                int j;
2363
2364                /* Try fixing the line in the preimage */
2365                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2366
2367                for (j = fixstart; j < fixed.len; j++)
2368                        if (!isspace(fixed.buf[j]))
2369                                goto unmatch_exit;
2370
2371                orig += oldlen;
2372        }
2373
2374        /*
2375         * Yes, the preimage is based on an older version that still
2376         * has whitespace breakages unfixed, and fixing them makes the
2377         * hunk match.  Update the context lines in the postimage.
2378         */
2379        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2380        update_pre_post_images(preimage, postimage,
2381                               fixed_buf, fixed_len, 0);
2382        return 1;
2383
2384 unmatch_exit:
2385        strbuf_release(&fixed);
2386        return 0;
2387}
2388
2389static int find_pos(struct image *img,
2390                    struct image *preimage,
2391                    struct image *postimage,
2392                    int line,
2393                    unsigned ws_rule,
2394                    int match_beginning, int match_end)
2395{
2396        int i;
2397        unsigned long backwards, forwards, try;
2398        int backwards_lno, forwards_lno, try_lno;
2399
2400        /*
2401         * If match_beginning or match_end is specified, there is no
2402         * point starting from a wrong line that will never match and
2403         * wander around and wait for a match at the specified end.
2404         */
2405        if (match_beginning)
2406                line = 0;
2407        else if (match_end)
2408                line = img->nr - preimage->nr;
2409
2410        /*
2411         * Because the comparison is unsigned, the following test
2412         * will also take care of a negative line number that can
2413         * result when match_end and preimage is larger than the target.
2414         */
2415        if ((size_t) line > img->nr)
2416                line = img->nr;
2417
2418        try = 0;
2419        for (i = 0; i < line; i++)
2420                try += img->line[i].len;
2421
2422        /*
2423         * There's probably some smart way to do this, but I'll leave
2424         * that to the smart and beautiful people. I'm simple and stupid.
2425         */
2426        backwards = try;
2427        backwards_lno = line;
2428        forwards = try;
2429        forwards_lno = line;
2430        try_lno = line;
2431
2432        for (i = 0; ; i++) {
2433                if (match_fragment(img, preimage, postimage,
2434                                   try, try_lno, ws_rule,
2435                                   match_beginning, match_end))
2436                        return try_lno;
2437
2438        again:
2439                if (backwards_lno == 0 && forwards_lno == img->nr)
2440                        break;
2441
2442                if (i & 1) {
2443                        if (backwards_lno == 0) {
2444                                i++;
2445                                goto again;
2446                        }
2447                        backwards_lno--;
2448                        backwards -= img->line[backwards_lno].len;
2449                        try = backwards;
2450                        try_lno = backwards_lno;
2451                } else {
2452                        if (forwards_lno == img->nr) {
2453                                i++;
2454                                goto again;
2455                        }
2456                        forwards += img->line[forwards_lno].len;
2457                        forwards_lno++;
2458                        try = forwards;
2459                        try_lno = forwards_lno;
2460                }
2461
2462        }
2463        return -1;
2464}
2465
2466static void remove_first_line(struct image *img)
2467{
2468        img->buf += img->line[0].len;
2469        img->len -= img->line[0].len;
2470        img->line++;
2471        img->nr--;
2472}
2473
2474static void remove_last_line(struct image *img)
2475{
2476        img->len -= img->line[--img->nr].len;
2477}
2478
2479/*
2480 * The change from "preimage" and "postimage" has been found to
2481 * apply at applied_pos (counts in line numbers) in "img".
2482 * Update "img" to remove "preimage" and replace it with "postimage".
2483 */
2484static void update_image(struct image *img,
2485                         int applied_pos,
2486                         struct image *preimage,
2487                         struct image *postimage)
2488{
2489        /*
2490         * remove the copy of preimage at offset in img
2491         * and replace it with postimage
2492         */
2493        int i, nr;
2494        size_t remove_count, insert_count, applied_at = 0;
2495        char *result;
2496        int preimage_limit;
2497
2498        /*
2499         * If we are removing blank lines at the end of img,
2500         * the preimage may extend beyond the end.
2501         * If that is the case, we must be careful only to
2502         * remove the part of the preimage that falls within
2503         * the boundaries of img. Initialize preimage_limit
2504         * to the number of lines in the preimage that falls
2505         * within the boundaries.
2506         */
2507        preimage_limit = preimage->nr;
2508        if (preimage_limit > img->nr - applied_pos)
2509                preimage_limit = img->nr - applied_pos;
2510
2511        for (i = 0; i < applied_pos; i++)
2512                applied_at += img->line[i].len;
2513
2514        remove_count = 0;
2515        for (i = 0; i < preimage_limit; i++)
2516                remove_count += img->line[applied_pos + i].len;
2517        insert_count = postimage->len;
2518
2519        /* Adjust the contents */
2520        result = xmalloc(img->len + insert_count - remove_count + 1);
2521        memcpy(result, img->buf, applied_at);
2522        memcpy(result + applied_at, postimage->buf, postimage->len);
2523        memcpy(result + applied_at + postimage->len,
2524               img->buf + (applied_at + remove_count),
2525               img->len - (applied_at + remove_count));
2526        free(img->buf);
2527        img->buf = result;
2528        img->len += insert_count - remove_count;
2529        result[img->len] = '\0';
2530
2531        /* Adjust the line table */
2532        nr = img->nr + postimage->nr - preimage_limit;
2533        if (preimage_limit < postimage->nr) {
2534                /*
2535                 * NOTE: this knows that we never call remove_first_line()
2536                 * on anything other than pre/post image.
2537                 */
2538                img->line = xrealloc(img->line, nr * sizeof(*img->line));
2539                img->line_allocated = img->line;
2540        }
2541        if (preimage_limit != postimage->nr)
2542                memmove(img->line + applied_pos + postimage->nr,
2543                        img->line + applied_pos + preimage_limit,
2544                        (img->nr - (applied_pos + preimage_limit)) *
2545                        sizeof(*img->line));
2546        memcpy(img->line + applied_pos,
2547               postimage->line,
2548               postimage->nr * sizeof(*img->line));
2549        if (!allow_overlap)
2550                for (i = 0; i < postimage->nr; i++)
2551                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2552        img->nr = nr;
2553}
2554
2555/*
2556 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2557 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2558 * replace the part of "img" with "postimage" text.
2559 */
2560static int apply_one_fragment(struct image *img, struct fragment *frag,
2561                              int inaccurate_eof, unsigned ws_rule,
2562                              int nth_fragment)
2563{
2564        int match_beginning, match_end;
2565        const char *patch = frag->patch;
2566        int size = frag->size;
2567        char *old, *oldlines;
2568        struct strbuf newlines;
2569        int new_blank_lines_at_end = 0;
2570        int found_new_blank_lines_at_end = 0;
2571        int hunk_linenr = frag->linenr;
2572        unsigned long leading, trailing;
2573        int pos, applied_pos;
2574        struct image preimage;
2575        struct image postimage;
2576
2577        memset(&preimage, 0, sizeof(preimage));
2578        memset(&postimage, 0, sizeof(postimage));
2579        oldlines = xmalloc(size);
2580        strbuf_init(&newlines, size);
2581
2582        old = oldlines;
2583        while (size > 0) {
2584                char first;
2585                int len = linelen(patch, size);
2586                int plen;
2587                int added_blank_line = 0;
2588                int is_blank_context = 0;
2589                size_t start;
2590
2591                if (!len)
2592                        break;
2593
2594                /*
2595                 * "plen" is how much of the line we should use for
2596                 * the actual patch data. Normally we just remove the
2597                 * first character on the line, but if the line is
2598                 * followed by "\ No newline", then we also remove the
2599                 * last one (which is the newline, of course).
2600                 */
2601                plen = len - 1;
2602                if (len < size && patch[len] == '\\')
2603                        plen--;
2604                first = *patch;
2605                if (apply_in_reverse) {
2606                        if (first == '-')
2607                                first = '+';
2608                        else if (first == '+')
2609                                first = '-';
2610                }
2611
2612                switch (first) {
2613                case '\n':
2614                        /* Newer GNU diff, empty context line */
2615                        if (plen < 0)
2616                                /* ... followed by '\No newline'; nothing */
2617                                break;
2618                        *old++ = '\n';
2619                        strbuf_addch(&newlines, '\n');
2620                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2621                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2622                        is_blank_context = 1;
2623                        break;
2624                case ' ':
2625                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2626                            ws_blank_line(patch + 1, plen, ws_rule))
2627                                is_blank_context = 1;
2628                case '-':
2629                        memcpy(old, patch + 1, plen);
2630                        add_line_info(&preimage, old, plen,
2631                                      (first == ' ' ? LINE_COMMON : 0));
2632                        old += plen;
2633                        if (first == '-')
2634                                break;
2635                /* Fall-through for ' ' */
2636                case '+':
2637                        /* --no-add does not add new lines */
2638                        if (first == '+' && no_add)
2639                                break;
2640
2641                        start = newlines.len;
2642                        if (first != '+' ||
2643                            !whitespace_error ||
2644                            ws_error_action != correct_ws_error) {
2645                                strbuf_add(&newlines, patch + 1, plen);
2646                        }
2647                        else {
2648                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2649                        }
2650                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2651                                      (first == '+' ? 0 : LINE_COMMON));
2652                        if (first == '+' &&
2653                            (ws_rule & WS_BLANK_AT_EOF) &&
2654                            ws_blank_line(patch + 1, plen, ws_rule))
2655                                added_blank_line = 1;
2656                        break;
2657                case '@': case '\\':
2658                        /* Ignore it, we already handled it */
2659                        break;
2660                default:
2661                        if (apply_verbosely)
2662                                error(_("invalid start of line: '%c'"), first);
2663                        return -1;
2664                }
2665                if (added_blank_line) {
2666                        if (!new_blank_lines_at_end)
2667                                found_new_blank_lines_at_end = hunk_linenr;
2668                        new_blank_lines_at_end++;
2669                }
2670                else if (is_blank_context)
2671                        ;
2672                else
2673                        new_blank_lines_at_end = 0;
2674                patch += len;
2675                size -= len;
2676                hunk_linenr++;
2677        }
2678        if (inaccurate_eof &&
2679            old > oldlines && old[-1] == '\n' &&
2680            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2681                old--;
2682                strbuf_setlen(&newlines, newlines.len - 1);
2683        }
2684
2685        leading = frag->leading;
2686        trailing = frag->trailing;
2687
2688        /*
2689         * A hunk to change lines at the beginning would begin with
2690         * @@ -1,L +N,M @@
2691         * but we need to be careful.  -U0 that inserts before the second
2692         * line also has this pattern.
2693         *
2694         * And a hunk to add to an empty file would begin with
2695         * @@ -0,0 +N,M @@
2696         *
2697         * In other words, a hunk that is (frag->oldpos <= 1) with or
2698         * without leading context must match at the beginning.
2699         */
2700        match_beginning = (!frag->oldpos ||
2701                           (frag->oldpos == 1 && !unidiff_zero));
2702
2703        /*
2704         * A hunk without trailing lines must match at the end.
2705         * However, we simply cannot tell if a hunk must match end
2706         * from the lack of trailing lines if the patch was generated
2707         * with unidiff without any context.
2708         */
2709        match_end = !unidiff_zero && !trailing;
2710
2711        pos = frag->newpos ? (frag->newpos - 1) : 0;
2712        preimage.buf = oldlines;
2713        preimage.len = old - oldlines;
2714        postimage.buf = newlines.buf;
2715        postimage.len = newlines.len;
2716        preimage.line = preimage.line_allocated;
2717        postimage.line = postimage.line_allocated;
2718
2719        for (;;) {
2720
2721                applied_pos = find_pos(img, &preimage, &postimage, pos,
2722                                       ws_rule, match_beginning, match_end);
2723
2724                if (applied_pos >= 0)
2725                        break;
2726
2727                /* Am I at my context limits? */
2728                if ((leading <= p_context) && (trailing <= p_context))
2729                        break;
2730                if (match_beginning || match_end) {
2731                        match_beginning = match_end = 0;
2732                        continue;
2733                }
2734
2735                /*
2736                 * Reduce the number of context lines; reduce both
2737                 * leading and trailing if they are equal otherwise
2738                 * just reduce the larger context.
2739                 */
2740                if (leading >= trailing) {
2741                        remove_first_line(&preimage);
2742                        remove_first_line(&postimage);
2743                        pos--;
2744                        leading--;
2745                }
2746                if (trailing > leading) {
2747                        remove_last_line(&preimage);
2748                        remove_last_line(&postimage);
2749                        trailing--;
2750                }
2751        }
2752
2753        if (applied_pos >= 0) {
2754                if (new_blank_lines_at_end &&
2755                    preimage.nr + applied_pos >= img->nr &&
2756                    (ws_rule & WS_BLANK_AT_EOF) &&
2757                    ws_error_action != nowarn_ws_error) {
2758                        record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2759                                        found_new_blank_lines_at_end);
2760                        if (ws_error_action == correct_ws_error) {
2761                                while (new_blank_lines_at_end--)
2762                                        remove_last_line(&postimage);
2763                        }
2764                        /*
2765                         * We would want to prevent write_out_results()
2766                         * from taking place in apply_patch() that follows
2767                         * the callchain led us here, which is:
2768                         * apply_patch->check_patch_list->check_patch->
2769                         * apply_data->apply_fragments->apply_one_fragment
2770                         */
2771                        if (ws_error_action == die_on_ws_error)
2772                                apply = 0;
2773                }
2774
2775                if (apply_verbosely && applied_pos != pos) {
2776                        int offset = applied_pos - pos;
2777                        if (apply_in_reverse)
2778                                offset = 0 - offset;
2779                        fprintf_ln(stderr,
2780                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
2781                                      "Hunk #%d succeeded at %d (offset %d lines).",
2782                                      offset),
2783                                   nth_fragment, applied_pos + 1, offset);
2784                }
2785
2786                /*
2787                 * Warn if it was necessary to reduce the number
2788                 * of context lines.
2789                 */
2790                if ((leading != frag->leading) ||
2791                    (trailing != frag->trailing))
2792                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2793                                             " to apply fragment at %d"),
2794                                   leading, trailing, applied_pos+1);
2795                update_image(img, applied_pos, &preimage, &postimage);
2796        } else {
2797                if (apply_verbosely)
2798                        error(_("while searching for:\n%.*s"),
2799                              (int)(old - oldlines), oldlines);
2800        }
2801
2802        free(oldlines);
2803        strbuf_release(&newlines);
2804        free(preimage.line_allocated);
2805        free(postimage.line_allocated);
2806
2807        return (applied_pos < 0);
2808}
2809
2810static int apply_binary_fragment(struct image *img, struct patch *patch)
2811{
2812        struct fragment *fragment = patch->fragments;
2813        unsigned long len;
2814        void *dst;
2815
2816        if (!fragment)
2817                return error(_("missing binary patch data for '%s'"),
2818                             patch->new_name ?
2819                             patch->new_name :
2820                             patch->old_name);
2821
2822        /* Binary patch is irreversible without the optional second hunk */
2823        if (apply_in_reverse) {
2824                if (!fragment->next)
2825                        return error("cannot reverse-apply a binary patch "
2826                                     "without the reverse hunk to '%s'",
2827                                     patch->new_name
2828                                     ? patch->new_name : patch->old_name);
2829                fragment = fragment->next;
2830        }
2831        switch (fragment->binary_patch_method) {
2832        case BINARY_DELTA_DEFLATED:
2833                dst = patch_delta(img->buf, img->len, fragment->patch,
2834                                  fragment->size, &len);
2835                if (!dst)
2836                        return -1;
2837                clear_image(img);
2838                img->buf = dst;
2839                img->len = len;
2840                return 0;
2841        case BINARY_LITERAL_DEFLATED:
2842                clear_image(img);
2843                img->len = fragment->size;
2844                img->buf = xmalloc(img->len+1);
2845                memcpy(img->buf, fragment->patch, img->len);
2846                img->buf[img->len] = '\0';
2847                return 0;
2848        }
2849        return -1;
2850}
2851
2852/*
2853 * Replace "img" with the result of applying the binary patch.
2854 * The binary patch data itself in patch->fragment is still kept
2855 * but the preimage prepared by the caller in "img" is freed here
2856 * or in the helper function apply_binary_fragment() this calls.
2857 */
2858static int apply_binary(struct image *img, struct patch *patch)
2859{
2860        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2861        unsigned char sha1[20];
2862
2863        /*
2864         * For safety, we require patch index line to contain
2865         * full 40-byte textual SHA1 for old and new, at least for now.
2866         */
2867        if (strlen(patch->old_sha1_prefix) != 40 ||
2868            strlen(patch->new_sha1_prefix) != 40 ||
2869            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2870            get_sha1_hex(patch->new_sha1_prefix, sha1))
2871                return error("cannot apply binary patch to '%s' "
2872                             "without full index line", name);
2873
2874        if (patch->old_name) {
2875                /*
2876                 * See if the old one matches what the patch
2877                 * applies to.
2878                 */
2879                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2880                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2881                        return error("the patch applies to '%s' (%s), "
2882                                     "which does not match the "
2883                                     "current contents.",
2884                                     name, sha1_to_hex(sha1));
2885        }
2886        else {
2887                /* Otherwise, the old one must be empty. */
2888                if (img->len)
2889                        return error("the patch applies to an empty "
2890                                     "'%s' but it is not empty", name);
2891        }
2892
2893        get_sha1_hex(patch->new_sha1_prefix, sha1);
2894        if (is_null_sha1(sha1)) {
2895                clear_image(img);
2896                return 0; /* deletion patch */
2897        }
2898
2899        if (has_sha1_file(sha1)) {
2900                /* We already have the postimage */
2901                enum object_type type;
2902                unsigned long size;
2903                char *result;
2904
2905                result = read_sha1_file(sha1, &type, &size);
2906                if (!result)
2907                        return error("the necessary postimage %s for "
2908                                     "'%s' cannot be read",
2909                                     patch->new_sha1_prefix, name);
2910                clear_image(img);
2911                img->buf = result;
2912                img->len = size;
2913        } else {
2914                /*
2915                 * We have verified buf matches the preimage;
2916                 * apply the patch data to it, which is stored
2917                 * in the patch->fragments->{patch,size}.
2918                 */
2919                if (apply_binary_fragment(img, patch))
2920                        return error(_("binary patch does not apply to '%s'"),
2921                                     name);
2922
2923                /* verify that the result matches */
2924                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2925                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2926                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2927                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2928        }
2929
2930        return 0;
2931}
2932
2933static int apply_fragments(struct image *img, struct patch *patch)
2934{
2935        struct fragment *frag = patch->fragments;
2936        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2937        unsigned ws_rule = patch->ws_rule;
2938        unsigned inaccurate_eof = patch->inaccurate_eof;
2939        int nth = 0;
2940
2941        if (patch->is_binary)
2942                return apply_binary(img, patch);
2943
2944        while (frag) {
2945                nth++;
2946                if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2947                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
2948                        if (!apply_with_reject)
2949                                return -1;
2950                        frag->rejected = 1;
2951                }
2952                frag = frag->next;
2953        }
2954        return 0;
2955}
2956
2957static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2958{
2959        if (!ce)
2960                return 0;
2961
2962        if (S_ISGITLINK(ce->ce_mode)) {
2963                strbuf_grow(buf, 100);
2964                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2965        } else {
2966                enum object_type type;
2967                unsigned long sz;
2968                char *result;
2969
2970                result = read_sha1_file(ce->sha1, &type, &sz);
2971                if (!result)
2972                        return -1;
2973                /* XXX read_sha1_file NUL-terminates */
2974                strbuf_attach(buf, result, sz, sz + 1);
2975        }
2976        return 0;
2977}
2978
2979static struct patch *in_fn_table(const char *name)
2980{
2981        struct string_list_item *item;
2982
2983        if (name == NULL)
2984                return NULL;
2985
2986        item = string_list_lookup(&fn_table, name);
2987        if (item != NULL)
2988                return (struct patch *)item->util;
2989
2990        return NULL;
2991}
2992
2993/*
2994 * item->util in the filename table records the status of the path.
2995 * Usually it points at a patch (whose result records the contents
2996 * of it after applying it), but it could be PATH_WAS_DELETED for a
2997 * path that a previously applied patch has already removed.
2998 */
2999 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3000#define PATH_WAS_DELETED ((struct patch *) -1)
3001
3002static int to_be_deleted(struct patch *patch)
3003{
3004        return patch == PATH_TO_BE_DELETED;
3005}
3006
3007static int was_deleted(struct patch *patch)
3008{
3009        return patch == PATH_WAS_DELETED;
3010}
3011
3012static void add_to_fn_table(struct patch *patch)
3013{
3014        struct string_list_item *item;
3015
3016        /*
3017         * Always add new_name unless patch is a deletion
3018         * This should cover the cases for normal diffs,
3019         * file creations and copies
3020         */
3021        if (patch->new_name != NULL) {
3022                item = string_list_insert(&fn_table, patch->new_name);
3023                item->util = patch;
3024        }
3025
3026        /*
3027         * store a failure on rename/deletion cases because
3028         * later chunks shouldn't patch old names
3029         */
3030        if ((patch->new_name == NULL) || (patch->is_rename)) {
3031                item = string_list_insert(&fn_table, patch->old_name);
3032                item->util = PATH_WAS_DELETED;
3033        }
3034}
3035
3036static void prepare_fn_table(struct patch *patch)
3037{
3038        /*
3039         * store information about incoming file deletion
3040         */
3041        while (patch) {
3042                if ((patch->new_name == NULL) || (patch->is_rename)) {
3043                        struct string_list_item *item;
3044                        item = string_list_insert(&fn_table, patch->old_name);
3045                        item->util = PATH_TO_BE_DELETED;
3046                }
3047                patch = patch->next;
3048        }
3049}
3050
3051static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
3052{
3053        struct strbuf buf = STRBUF_INIT;
3054        struct image image;
3055        size_t len;
3056        char *img;
3057        struct patch *tpatch;
3058
3059        if (!(patch->is_copy || patch->is_rename) &&
3060            (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
3061                if (was_deleted(tpatch)) {
3062                        return error(_("patch %s has been renamed/deleted"),
3063                                patch->old_name);
3064                }
3065                /* We have a patched copy in memory; use that. */
3066                strbuf_add(&buf, tpatch->result, tpatch->resultsize);
3067        } else if (cached) {
3068                if (read_file_or_gitlink(ce, &buf))
3069                        return error(_("read of %s failed"), patch->old_name);
3070        } else if (patch->old_name) {
3071                if (S_ISGITLINK(patch->old_mode)) {
3072                        if (ce) {
3073                                read_file_or_gitlink(ce, &buf);
3074                        } else {
3075                                /*
3076                                 * There is no way to apply subproject
3077                                 * patch without looking at the index.
3078                                 * NEEDSWORK: shouldn't this be flagged
3079                                 * as an error???
3080                                 */
3081                                free_fragment_list(patch->fragments);
3082                                patch->fragments = NULL;
3083                        }
3084                } else {
3085                        if (read_old_data(st, patch->old_name, &buf))
3086                                return error(_("read of %s failed"), patch->old_name);
3087                }
3088        }
3089
3090        img = strbuf_detach(&buf, &len);
3091        prepare_image(&image, img, len, !patch->is_binary);
3092
3093        if (apply_fragments(&image, patch) < 0)
3094                return -1; /* note with --reject this succeeds. */
3095        patch->result = image.buf;
3096        patch->resultsize = image.len;
3097        add_to_fn_table(patch);
3098        free(image.line_allocated);
3099
3100        if (0 < patch->is_delete && patch->resultsize)
3101                return error(_("removal patch leaves file contents"));
3102
3103        return 0;
3104}
3105
3106static int check_to_create_blob(const char *new_name, int ok_if_exists)
3107{
3108        struct stat nst;
3109        if (!lstat(new_name, &nst)) {
3110                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3111                        return 0;
3112                /*
3113                 * A leading component of new_name might be a symlink
3114                 * that is going to be removed with this patch, but
3115                 * still pointing at somewhere that has the path.
3116                 * In such a case, path "new_name" does not exist as
3117                 * far as git is concerned.
3118                 */
3119                if (has_symlink_leading_path(new_name, strlen(new_name)))
3120                        return 0;
3121
3122                return error(_("%s: already exists in working directory"), new_name);
3123        }
3124        else if ((errno != ENOENT) && (errno != ENOTDIR))
3125                return error("%s: %s", new_name, strerror(errno));
3126        return 0;
3127}
3128
3129static int verify_index_match(struct cache_entry *ce, struct stat *st)
3130{
3131        if (S_ISGITLINK(ce->ce_mode)) {
3132                if (!S_ISDIR(st->st_mode))
3133                        return -1;
3134                return 0;
3135        }
3136        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3137}
3138
3139static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3140{
3141        const char *old_name = patch->old_name;
3142        struct patch *tpatch = NULL;
3143        int stat_ret = 0;
3144        unsigned st_mode = 0;
3145
3146        /*
3147         * Make sure that we do not have local modifications from the
3148         * index when we are looking at the index.  Also make sure
3149         * we have the preimage file to be patched in the work tree,
3150         * unless --cached, which tells git to apply only in the index.
3151         */
3152        if (!old_name)
3153                return 0;
3154
3155        assert(patch->is_new <= 0);
3156
3157        if (!(patch->is_copy || patch->is_rename) &&
3158            (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
3159                if (was_deleted(tpatch))
3160                        return error(_("%s: has been deleted/renamed"), old_name);
3161                st_mode = tpatch->new_mode;
3162        } else if (!cached) {
3163                stat_ret = lstat(old_name, st);
3164                if (stat_ret && errno != ENOENT)
3165                        return error(_("%s: %s"), old_name, strerror(errno));
3166        }
3167
3168        if (to_be_deleted(tpatch))
3169                tpatch = NULL;
3170
3171        if (check_index && !tpatch) {
3172                int pos = cache_name_pos(old_name, strlen(old_name));
3173                if (pos < 0) {
3174                        if (patch->is_new < 0)
3175                                goto is_new;
3176                        return error(_("%s: does not exist in index"), old_name);
3177                }
3178                *ce = active_cache[pos];
3179                if (stat_ret < 0) {
3180                        struct checkout costate;
3181                        /* checkout */
3182                        memset(&costate, 0, sizeof(costate));
3183                        costate.base_dir = "";
3184                        costate.refresh_cache = 1;
3185                        if (checkout_entry(*ce, &costate, NULL) ||
3186                            lstat(old_name, st))
3187                                return -1;
3188                }
3189                if (!cached && verify_index_match(*ce, st))
3190                        return error(_("%s: does not match index"), old_name);
3191                if (cached)
3192                        st_mode = (*ce)->ce_mode;
3193        } else if (stat_ret < 0) {
3194                if (patch->is_new < 0)
3195                        goto is_new;
3196                return error(_("%s: %s"), old_name, strerror(errno));
3197        }
3198
3199        if (!cached && !tpatch)
3200                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3201
3202        if (patch->is_new < 0)
3203                patch->is_new = 0;
3204        if (!patch->old_mode)
3205                patch->old_mode = st_mode;
3206        if ((st_mode ^ patch->old_mode) & S_IFMT)
3207                return error(_("%s: wrong type"), old_name);
3208        if (st_mode != patch->old_mode)
3209                warning(_("%s has type %o, expected %o"),
3210                        old_name, st_mode, patch->old_mode);
3211        if (!patch->new_mode && !patch->is_delete)
3212                patch->new_mode = st_mode;
3213        return 0;
3214
3215 is_new:
3216        patch->is_new = 1;
3217        patch->is_delete = 0;
3218        free(patch->old_name);
3219        patch->old_name = NULL;
3220        return 0;
3221}
3222
3223/*
3224 * Check and apply the patch in-core; leave the result in patch->result
3225 * for the caller to write it out to the final destination.
3226 */
3227static int check_patch(struct patch *patch)
3228{
3229        struct stat st;
3230        const char *old_name = patch->old_name;
3231        const char *new_name = patch->new_name;
3232        const char *name = old_name ? old_name : new_name;
3233        struct cache_entry *ce = NULL;
3234        struct patch *tpatch;
3235        int ok_if_exists;
3236        int status;
3237
3238        patch->rejected = 1; /* we will drop this after we succeed */
3239
3240        status = check_preimage(patch, &ce, &st);
3241        if (status)
3242                return status;
3243        old_name = patch->old_name;
3244
3245        if ((tpatch = in_fn_table(new_name)) &&
3246                        (was_deleted(tpatch) || to_be_deleted(tpatch)))
3247                /*
3248                 * A type-change diff is always split into a patch to
3249                 * delete old, immediately followed by a patch to
3250                 * create new (see diff.c::run_diff()); in such a case
3251                 * it is Ok that the entry to be deleted by the
3252                 * previous patch is still in the working tree and in
3253                 * the index.
3254                 */
3255                ok_if_exists = 1;
3256        else
3257                ok_if_exists = 0;
3258
3259        if (new_name &&
3260            ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3261                if (check_index &&
3262                    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3263                    !ok_if_exists)
3264                        return error(_("%s: already exists in index"), new_name);
3265                if (!cached) {
3266                        int err = check_to_create_blob(new_name, ok_if_exists);
3267                        if (err)
3268                                return err;
3269                }
3270                if (!patch->new_mode) {
3271                        if (0 < patch->is_new)
3272                                patch->new_mode = S_IFREG | 0644;
3273                        else
3274                                patch->new_mode = patch->old_mode;
3275                }
3276        }
3277
3278        if (new_name && old_name) {
3279                int same = !strcmp(old_name, new_name);
3280                if (!patch->new_mode)
3281                        patch->new_mode = patch->old_mode;
3282                if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3283                        if (same)
3284                                return error(_("new mode (%o) of %s does not "
3285                                               "match old mode (%o)"),
3286                                        patch->new_mode, new_name,
3287                                        patch->old_mode);
3288                        else
3289                                return error(_("new mode (%o) of %s does not "
3290                                               "match old mode (%o) of %s"),
3291                                        patch->new_mode, new_name,
3292                                        patch->old_mode, old_name);
3293                }
3294        }
3295
3296        if (apply_data(patch, &st, ce) < 0)
3297                return error(_("%s: patch does not apply"), name);
3298        patch->rejected = 0;
3299        return 0;
3300}
3301
3302static int check_patch_list(struct patch *patch)
3303{
3304        int err = 0;
3305
3306        prepare_fn_table(patch);
3307        while (patch) {
3308                if (apply_verbosely)
3309                        say_patch_name(stderr,
3310                                       _("Checking patch %s..."), patch);
3311                err |= check_patch(patch);
3312                patch = patch->next;
3313        }
3314        return err;
3315}
3316
3317/* This function tries to read the sha1 from the current index */
3318static int get_current_sha1(const char *path, unsigned char *sha1)
3319{
3320        int pos;
3321
3322        if (read_cache() < 0)
3323                return -1;
3324        pos = cache_name_pos(path, strlen(path));
3325        if (pos < 0)
3326                return -1;
3327        hashcpy(sha1, active_cache[pos]->sha1);
3328        return 0;
3329}
3330
3331/* Build an index that contains the just the files needed for a 3way merge */
3332static void build_fake_ancestor(struct patch *list, const char *filename)
3333{
3334        struct patch *patch;
3335        struct index_state result = { NULL };
3336        int fd;
3337
3338        /* Once we start supporting the reverse patch, it may be
3339         * worth showing the new sha1 prefix, but until then...
3340         */
3341        for (patch = list; patch; patch = patch->next) {
3342                const unsigned char *sha1_ptr;
3343                unsigned char sha1[20];
3344                struct cache_entry *ce;
3345                const char *name;
3346
3347                name = patch->old_name ? patch->old_name : patch->new_name;
3348                if (0 < patch->is_new)
3349                        continue;
3350                else if (get_sha1(patch->old_sha1_prefix, sha1))
3351                        /* git diff has no index line for mode/type changes */
3352                        if (!patch->lines_added && !patch->lines_deleted) {
3353                                if (get_current_sha1(patch->old_name, sha1))
3354                                        die("mode change for %s, which is not "
3355                                                "in current HEAD", name);
3356                                sha1_ptr = sha1;
3357                        } else
3358                                die("sha1 information is lacking or useless "
3359                                        "(%s).", name);
3360                else
3361                        sha1_ptr = sha1;
3362
3363                ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3364                if (!ce)
3365                        die(_("make_cache_entry failed for path '%s'"), name);
3366                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3367                        die ("Could not add %s to temporary index", name);
3368        }
3369
3370        fd = open(filename, O_WRONLY | O_CREAT, 0666);
3371        if (fd < 0 || write_index(&result, fd) || close(fd))
3372                die ("Could not write temporary index to %s", filename);
3373
3374        discard_index(&result);
3375}
3376
3377static void stat_patch_list(struct patch *patch)
3378{
3379        int files, adds, dels;
3380
3381        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3382                files++;
3383                adds += patch->lines_added;
3384                dels += patch->lines_deleted;
3385                show_stats(patch);
3386        }
3387
3388        print_stat_summary(stdout, files, adds, dels);
3389}
3390
3391static void numstat_patch_list(struct patch *patch)
3392{
3393        for ( ; patch; patch = patch->next) {
3394                const char *name;
3395                name = patch->new_name ? patch->new_name : patch->old_name;
3396                if (patch->is_binary)
3397                        printf("-\t-\t");
3398                else
3399                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3400                write_name_quoted(name, stdout, line_termination);
3401        }
3402}
3403
3404static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3405{
3406        if (mode)
3407                printf(" %s mode %06o %s\n", newdelete, mode, name);
3408        else
3409                printf(" %s %s\n", newdelete, name);
3410}
3411
3412static void show_mode_change(struct patch *p, int show_name)
3413{
3414        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3415                if (show_name)
3416                        printf(" mode change %06o => %06o %s\n",
3417                               p->old_mode, p->new_mode, p->new_name);
3418                else
3419                        printf(" mode change %06o => %06o\n",
3420                               p->old_mode, p->new_mode);
3421        }
3422}
3423
3424static void show_rename_copy(struct patch *p)
3425{
3426        const char *renamecopy = p->is_rename ? "rename" : "copy";
3427        const char *old, *new;
3428
3429        /* Find common prefix */
3430        old = p->old_name;
3431        new = p->new_name;
3432        while (1) {
3433                const char *slash_old, *slash_new;
3434                slash_old = strchr(old, '/');
3435                slash_new = strchr(new, '/');
3436                if (!slash_old ||
3437                    !slash_new ||
3438                    slash_old - old != slash_new - new ||
3439                    memcmp(old, new, slash_new - new))
3440                        break;
3441                old = slash_old + 1;
3442                new = slash_new + 1;
3443        }
3444        /* p->old_name thru old is the common prefix, and old and new
3445         * through the end of names are renames
3446         */
3447        if (old != p->old_name)
3448                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3449                       (int)(old - p->old_name), p->old_name,
3450                       old, new, p->score);
3451        else
3452                printf(" %s %s => %s (%d%%)\n", renamecopy,
3453                       p->old_name, p->new_name, p->score);
3454        show_mode_change(p, 0);
3455}
3456
3457static void summary_patch_list(struct patch *patch)
3458{
3459        struct patch *p;
3460
3461        for (p = patch; p; p = p->next) {
3462                if (p->is_new)
3463                        show_file_mode_name("create", p->new_mode, p->new_name);
3464                else if (p->is_delete)
3465                        show_file_mode_name("delete", p->old_mode, p->old_name);
3466                else {
3467                        if (p->is_rename || p->is_copy)
3468                                show_rename_copy(p);
3469                        else {
3470                                if (p->score) {
3471                                        printf(" rewrite %s (%d%%)\n",
3472                                               p->new_name, p->score);
3473                                        show_mode_change(p, 0);
3474                                }
3475                                else
3476                                        show_mode_change(p, 1);
3477                        }
3478                }
3479        }
3480}
3481
3482static void patch_stats(struct patch *patch)
3483{
3484        int lines = patch->lines_added + patch->lines_deleted;
3485
3486        if (lines > max_change)
3487                max_change = lines;
3488        if (patch->old_name) {
3489                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3490                if (!len)
3491                        len = strlen(patch->old_name);
3492                if (len > max_len)
3493                        max_len = len;
3494        }
3495        if (patch->new_name) {
3496                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3497                if (!len)
3498                        len = strlen(patch->new_name);
3499                if (len > max_len)
3500                        max_len = len;
3501        }
3502}
3503
3504static void remove_file(struct patch *patch, int rmdir_empty)
3505{
3506        if (update_index) {
3507                if (remove_file_from_cache(patch->old_name) < 0)
3508                        die(_("unable to remove %s from index"), patch->old_name);
3509        }
3510        if (!cached) {
3511                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3512                        remove_path(patch->old_name);
3513                }
3514        }
3515}
3516
3517static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3518{
3519        struct stat st;
3520        struct cache_entry *ce;
3521        int namelen = strlen(path);
3522        unsigned ce_size = cache_entry_size(namelen);
3523
3524        if (!update_index)
3525                return;
3526
3527        ce = xcalloc(1, ce_size);
3528        memcpy(ce->name, path, namelen);
3529        ce->ce_mode = create_ce_mode(mode);
3530        ce->ce_flags = namelen;
3531        if (S_ISGITLINK(mode)) {
3532                const char *s = buf;
3533
3534                if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3535                        die(_("corrupt patch for subproject %s"), path);
3536        } else {
3537                if (!cached) {
3538                        if (lstat(path, &st) < 0)
3539                                die_errno(_("unable to stat newly created file '%s'"),
3540                                          path);
3541                        fill_stat_cache_info(ce, &st);
3542                }
3543                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3544                        die(_("unable to create backing store for newly created file %s"), path);
3545        }
3546        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3547                die(_("unable to add cache entry for %s"), path);
3548}
3549
3550static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3551{
3552        int fd;
3553        struct strbuf nbuf = STRBUF_INIT;
3554
3555        if (S_ISGITLINK(mode)) {
3556                struct stat st;
3557                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3558                        return 0;
3559                return mkdir(path, 0777);
3560        }
3561
3562        if (has_symlinks && S_ISLNK(mode))
3563                /* Although buf:size is counted string, it also is NUL
3564                 * terminated.
3565                 */
3566                return symlink(buf, path);
3567
3568        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3569        if (fd < 0)
3570                return -1;
3571
3572        if (convert_to_working_tree(path, buf, size, &nbuf)) {
3573                size = nbuf.len;
3574                buf  = nbuf.buf;
3575        }
3576        write_or_die(fd, buf, size);
3577        strbuf_release(&nbuf);
3578
3579        if (close(fd) < 0)
3580                die_errno(_("closing file '%s'"), path);
3581        return 0;
3582}
3583
3584/*
3585 * We optimistically assume that the directories exist,
3586 * which is true 99% of the time anyway. If they don't,
3587 * we create them and try again.
3588 */
3589static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3590{
3591        if (cached)
3592                return;
3593        if (!try_create_file(path, mode, buf, size))
3594                return;
3595
3596        if (errno == ENOENT) {
3597                if (safe_create_leading_directories(path))
3598                        return;
3599                if (!try_create_file(path, mode, buf, size))
3600                        return;
3601        }
3602
3603        if (errno == EEXIST || errno == EACCES) {
3604                /* We may be trying to create a file where a directory
3605                 * used to be.
3606                 */
3607                struct stat st;
3608                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3609                        errno = EEXIST;
3610        }
3611
3612        if (errno == EEXIST) {
3613                unsigned int nr = getpid();
3614
3615                for (;;) {
3616                        char newpath[PATH_MAX];
3617                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3618                        if (!try_create_file(newpath, mode, buf, size)) {
3619                                if (!rename(newpath, path))
3620                                        return;
3621                                unlink_or_warn(newpath);
3622                                break;
3623                        }
3624                        if (errno != EEXIST)
3625                                break;
3626                        ++nr;
3627                }
3628        }
3629        die_errno(_("unable to write file '%s' mode %o"), path, mode);
3630}
3631
3632static void create_file(struct patch *patch)
3633{
3634        char *path = patch->new_name;
3635        unsigned mode = patch->new_mode;
3636        unsigned long size = patch->resultsize;
3637        char *buf = patch->result;
3638
3639        if (!mode)
3640                mode = S_IFREG | 0644;
3641        create_one_file(path, mode, buf, size);
3642        add_index_file(path, mode, buf, size);
3643}
3644
3645/* phase zero is to remove, phase one is to create */
3646static void write_out_one_result(struct patch *patch, int phase)
3647{
3648        if (patch->is_delete > 0) {
3649                if (phase == 0)
3650                        remove_file(patch, 1);
3651                return;
3652        }
3653        if (patch->is_new > 0 || patch->is_copy) {
3654                if (phase == 1)
3655                        create_file(patch);
3656                return;
3657        }
3658        /*
3659         * Rename or modification boils down to the same
3660         * thing: remove the old, write the new
3661         */
3662        if (phase == 0)
3663                remove_file(patch, patch->is_rename);
3664        if (phase == 1)
3665                create_file(patch);
3666}
3667
3668static int write_out_one_reject(struct patch *patch)
3669{
3670        FILE *rej;
3671        char namebuf[PATH_MAX];
3672        struct fragment *frag;
3673        int cnt = 0;
3674        struct strbuf sb = STRBUF_INIT;
3675
3676        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3677                if (!frag->rejected)
3678                        continue;
3679                cnt++;
3680        }
3681
3682        if (!cnt) {
3683                if (apply_verbosely)
3684                        say_patch_name(stderr,
3685                                       _("Applied patch %s cleanly."), patch);
3686                return 0;
3687        }
3688
3689        /* This should not happen, because a removal patch that leaves
3690         * contents are marked "rejected" at the patch level.
3691         */
3692        if (!patch->new_name)
3693                die(_("internal error"));
3694
3695        /* Say this even without --verbose */
3696        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
3697                            "Applying patch %%s with %d rejects...",
3698                            cnt),
3699                    cnt);
3700        say_patch_name(stderr, sb.buf, patch);
3701        strbuf_release(&sb);
3702
3703        cnt = strlen(patch->new_name);
3704        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3705                cnt = ARRAY_SIZE(namebuf) - 5;
3706                warning(_("truncating .rej filename to %.*s.rej"),
3707                        cnt - 1, patch->new_name);
3708        }
3709        memcpy(namebuf, patch->new_name, cnt);
3710        memcpy(namebuf + cnt, ".rej", 5);
3711
3712        rej = fopen(namebuf, "w");
3713        if (!rej)
3714                return error(_("cannot open %s: %s"), namebuf, strerror(errno));
3715
3716        /* Normal git tools never deal with .rej, so do not pretend
3717         * this is a git patch by saying --git nor give extended
3718         * headers.  While at it, maybe please "kompare" that wants
3719         * the trailing TAB and some garbage at the end of line ;-).
3720         */
3721        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3722                patch->new_name, patch->new_name);
3723        for (cnt = 1, frag = patch->fragments;
3724             frag;
3725             cnt++, frag = frag->next) {
3726                if (!frag->rejected) {
3727                        fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
3728                        continue;
3729                }
3730                fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
3731                fprintf(rej, "%.*s", frag->size, frag->patch);
3732                if (frag->patch[frag->size-1] != '\n')
3733                        fputc('\n', rej);
3734        }
3735        fclose(rej);
3736        return -1;
3737}
3738
3739static int write_out_results(struct patch *list)
3740{
3741        int phase;
3742        int errs = 0;
3743        struct patch *l;
3744
3745        for (phase = 0; phase < 2; phase++) {
3746                l = list;
3747                while (l) {
3748                        if (l->rejected)
3749                                errs = 1;
3750                        else {
3751                                write_out_one_result(l, phase);
3752                                if (phase == 1 && write_out_one_reject(l))
3753                                        errs = 1;
3754                        }
3755                        l = l->next;
3756                }
3757        }
3758        return errs;
3759}
3760
3761static struct lock_file lock_file;
3762
3763static struct string_list limit_by_name;
3764static int has_include;
3765static void add_name_limit(const char *name, int exclude)
3766{
3767        struct string_list_item *it;
3768
3769        it = string_list_append(&limit_by_name, name);
3770        it->util = exclude ? NULL : (void *) 1;
3771}
3772
3773static int use_patch(struct patch *p)
3774{
3775        const char *pathname = p->new_name ? p->new_name : p->old_name;
3776        int i;
3777
3778        /* Paths outside are not touched regardless of "--include" */
3779        if (0 < prefix_length) {
3780                int pathlen = strlen(pathname);
3781                if (pathlen <= prefix_length ||
3782                    memcmp(prefix, pathname, prefix_length))
3783                        return 0;
3784        }
3785
3786        /* See if it matches any of exclude/include rule */
3787        for (i = 0; i < limit_by_name.nr; i++) {
3788                struct string_list_item *it = &limit_by_name.items[i];
3789                if (!fnmatch(it->string, pathname, 0))
3790                        return (it->util != NULL);
3791        }
3792
3793        /*
3794         * If we had any include, a path that does not match any rule is
3795         * not used.  Otherwise, we saw bunch of exclude rules (or none)
3796         * and such a path is used.
3797         */
3798        return !has_include;
3799}
3800
3801
3802static void prefix_one(char **name)
3803{
3804        char *old_name = *name;
3805        if (!old_name)
3806                return;
3807        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3808        free(old_name);
3809}
3810
3811static void prefix_patches(struct patch *p)
3812{
3813        if (!prefix || p->is_toplevel_relative)
3814                return;
3815        for ( ; p; p = p->next) {
3816                prefix_one(&p->new_name);
3817                prefix_one(&p->old_name);
3818        }
3819}
3820
3821#define INACCURATE_EOF  (1<<0)
3822#define RECOUNT         (1<<1)
3823
3824static int apply_patch(int fd, const char *filename, int options)
3825{
3826        size_t offset;
3827        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
3828        struct patch *list = NULL, **listp = &list;
3829        int skipped_patch = 0;
3830
3831        patch_input_file = filename;
3832        read_patch_file(&buf, fd);
3833        offset = 0;
3834        while (offset < buf.len) {
3835                struct patch *patch;
3836                int nr;
3837
3838                patch = xcalloc(1, sizeof(*patch));
3839                patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3840                patch->recount =  !!(options & RECOUNT);
3841                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3842                if (nr < 0)
3843                        break;
3844                if (apply_in_reverse)
3845                        reverse_patches(patch);
3846                if (prefix)
3847                        prefix_patches(patch);
3848                if (use_patch(patch)) {
3849                        patch_stats(patch);
3850                        *listp = patch;
3851                        listp = &patch->next;
3852                }
3853                else {
3854                        free_patch(patch);
3855                        skipped_patch++;
3856                }
3857                offset += nr;
3858        }
3859
3860        if (!list && !skipped_patch)
3861                die(_("unrecognized input"));
3862
3863        if (whitespace_error && (ws_error_action == die_on_ws_error))
3864                apply = 0;
3865
3866        update_index = check_index && apply;
3867        if (update_index && newfd < 0)
3868                newfd = hold_locked_index(&lock_file, 1);
3869
3870        if (check_index) {
3871                if (read_cache() < 0)
3872                        die(_("unable to read index file"));
3873        }
3874
3875        if ((check || apply) &&
3876            check_patch_list(list) < 0 &&
3877            !apply_with_reject)
3878                exit(1);
3879
3880        if (apply && write_out_results(list))
3881                exit(1);
3882
3883        if (fake_ancestor)
3884                build_fake_ancestor(list, fake_ancestor);
3885
3886        if (diffstat)
3887                stat_patch_list(list);
3888
3889        if (numstat)
3890                numstat_patch_list(list);
3891
3892        if (summary)
3893                summary_patch_list(list);
3894
3895        free_patch_list(list);
3896        strbuf_release(&buf);
3897        string_list_clear(&fn_table, 0);
3898        return 0;
3899}
3900
3901static int git_apply_config(const char *var, const char *value, void *cb)
3902{
3903        if (!strcmp(var, "apply.whitespace"))
3904                return git_config_string(&apply_default_whitespace, var, value);
3905        else if (!strcmp(var, "apply.ignorewhitespace"))
3906                return git_config_string(&apply_default_ignorewhitespace, var, value);
3907        return git_default_config(var, value, cb);
3908}
3909
3910static int option_parse_exclude(const struct option *opt,
3911                                const char *arg, int unset)
3912{
3913        add_name_limit(arg, 1);
3914        return 0;
3915}
3916
3917static int option_parse_include(const struct option *opt,
3918                                const char *arg, int unset)
3919{
3920        add_name_limit(arg, 0);
3921        has_include = 1;
3922        return 0;
3923}
3924
3925static int option_parse_p(const struct option *opt,
3926                          const char *arg, int unset)
3927{
3928        p_value = atoi(arg);
3929        p_value_known = 1;
3930        return 0;
3931}
3932
3933static int option_parse_z(const struct option *opt,
3934                          const char *arg, int unset)
3935{
3936        if (unset)
3937                line_termination = '\n';
3938        else
3939                line_termination = 0;
3940        return 0;
3941}
3942
3943static int option_parse_space_change(const struct option *opt,
3944                          const char *arg, int unset)
3945{
3946        if (unset)
3947                ws_ignore_action = ignore_ws_none;
3948        else
3949                ws_ignore_action = ignore_ws_change;
3950        return 0;
3951}
3952
3953static int option_parse_whitespace(const struct option *opt,
3954                                   const char *arg, int unset)
3955{
3956        const char **whitespace_option = opt->value;
3957
3958        *whitespace_option = arg;
3959        parse_whitespace_option(arg);
3960        return 0;
3961}
3962
3963static int option_parse_directory(const struct option *opt,
3964                                  const char *arg, int unset)
3965{
3966        root_len = strlen(arg);
3967        if (root_len && arg[root_len - 1] != '/') {
3968                char *new_root;
3969                root = new_root = xmalloc(root_len + 2);
3970                strcpy(new_root, arg);
3971                strcpy(new_root + root_len++, "/");
3972        } else
3973                root = arg;
3974        return 0;
3975}
3976
3977int cmd_apply(int argc, const char **argv, const char *prefix_)
3978{
3979        int i;
3980        int errs = 0;
3981        int is_not_gitdir = !startup_info->have_repository;
3982        int force_apply = 0;
3983
3984        const char *whitespace_option = NULL;
3985
3986        struct option builtin_apply_options[] = {
3987                { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
3988                        N_("don't apply changes matching the given path"),
3989                        0, option_parse_exclude },
3990                { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
3991                        N_("apply changes matching the given path"),
3992                        0, option_parse_include },
3993                { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
3994                        N_("remove <num> leading slashes from traditional diff paths"),
3995                        0, option_parse_p },
3996                OPT_BOOLEAN(0, "no-add", &no_add,
3997                        N_("ignore additions made by the patch")),
3998                OPT_BOOLEAN(0, "stat", &diffstat,
3999                        N_("instead of applying the patch, output diffstat for the input")),
4000                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4001                OPT_NOOP_NOARG(0, "binary"),
4002                OPT_BOOLEAN(0, "numstat", &numstat,
4003                        N_("shows number of added and deleted lines in decimal notation")),
4004                OPT_BOOLEAN(0, "summary", &summary,
4005                        N_("instead of applying the patch, output a summary for the input")),
4006                OPT_BOOLEAN(0, "check", &check,
4007                        N_("instead of applying the patch, see if the patch is applicable")),
4008                OPT_BOOLEAN(0, "index", &check_index,
4009                        N_("make sure the patch is applicable to the current index")),
4010                OPT_BOOLEAN(0, "cached", &cached,
4011                        N_("apply a patch without touching the working tree")),
4012                OPT_BOOLEAN(0, "apply", &force_apply,
4013                        N_("also apply the patch (use with --stat/--summary/--check)")),
4014                OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4015                        N_("build a temporary index based on embedded index information")),
4016                { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4017                        N_("paths are separated with NUL character"),
4018                        PARSE_OPT_NOARG, option_parse_z },
4019                OPT_INTEGER('C', NULL, &p_context,
4020                                N_("ensure at least <n> lines of context match")),
4021                { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4022                        N_("detect new or modified lines that have whitespace errors"),
4023                        0, option_parse_whitespace },
4024                { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4025                        N_("ignore changes in whitespace when finding context"),
4026                        PARSE_OPT_NOARG, option_parse_space_change },
4027                { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4028                        N_("ignore changes in whitespace when finding context"),
4029                        PARSE_OPT_NOARG, option_parse_space_change },
4030                OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
4031                        N_("apply the patch in reverse")),
4032                OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
4033                        N_("don't expect at least one line of context")),
4034                OPT_BOOLEAN(0, "reject", &apply_with_reject,
4035                        N_("leave the rejected hunks in corresponding *.rej files")),
4036                OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
4037                        N_("allow overlapping hunks")),
4038                OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4039                OPT_BIT(0, "inaccurate-eof", &options,
4040                        N_("tolerate incorrectly detected missing new-line at the end of file"),
4041                        INACCURATE_EOF),
4042                OPT_BIT(0, "recount", &options,
4043                        N_("do not trust the line counts in the hunk headers"),
4044                        RECOUNT),
4045                { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4046                        N_("prepend <root> to all filenames"),
4047                        0, option_parse_directory },
4048                OPT_END()
4049        };
4050
4051        prefix = prefix_;
4052        prefix_length = prefix ? strlen(prefix) : 0;
4053        git_config(git_apply_config, NULL);
4054        if (apply_default_whitespace)
4055                parse_whitespace_option(apply_default_whitespace);
4056        if (apply_default_ignorewhitespace)
4057                parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4058
4059        argc = parse_options(argc, argv, prefix, builtin_apply_options,
4060                        apply_usage, 0);
4061
4062        if (apply_with_reject)
4063                apply = apply_verbosely = 1;
4064        if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4065                apply = 0;
4066        if (check_index && is_not_gitdir)
4067                die(_("--index outside a repository"));
4068        if (cached) {
4069                if (is_not_gitdir)
4070                        die(_("--cached outside a repository"));
4071                check_index = 1;
4072        }
4073        for (i = 0; i < argc; i++) {
4074                const char *arg = argv[i];
4075                int fd;
4076
4077                if (!strcmp(arg, "-")) {
4078                        errs |= apply_patch(0, "<stdin>", options);
4079                        read_stdin = 0;
4080                        continue;
4081                } else if (0 < prefix_length)
4082                        arg = prefix_filename(prefix, prefix_length, arg);
4083
4084                fd = open(arg, O_RDONLY);
4085                if (fd < 0)
4086                        die_errno(_("can't open patch '%s'"), arg);
4087                read_stdin = 0;
4088                set_default_whitespace_mode(whitespace_option);
4089                errs |= apply_patch(fd, arg, options);
4090                close(fd);
4091        }
4092        set_default_whitespace_mode(whitespace_option);
4093        if (read_stdin)
4094                errs |= apply_patch(0, "<stdin>", options);
4095        if (whitespace_error) {
4096                if (squelch_whitespace_errors &&
4097                    squelch_whitespace_errors < whitespace_error) {
4098                        int squelched =
4099                                whitespace_error - squelch_whitespace_errors;
4100                        warning(Q_("squelched %d whitespace error",
4101                                   "squelched %d whitespace errors",
4102                                   squelched),
4103                                squelched);
4104                }
4105                if (ws_error_action == die_on_ws_error)
4106                        die(Q_("%d line adds whitespace errors.",
4107                               "%d lines add whitespace errors.",
4108                               whitespace_error),
4109                            whitespace_error);
4110                if (applied_after_fixing_ws && apply)
4111                        warning("%d line%s applied after"
4112                                " fixing whitespace errors.",
4113                                applied_after_fixing_ws,
4114                                applied_after_fixing_ws == 1 ? "" : "s");
4115                else if (whitespace_error)
4116                        warning(Q_("%d line adds whitespace errors.",
4117                                   "%d lines add whitespace errors.",
4118                                   whitespace_error),
4119                                whitespace_error);
4120        }
4121
4122        if (update_index) {
4123                if (write_cache(newfd, active_cache, active_nr) ||
4124                    commit_locked_index(&lock_file))
4125                        die(_("Unable to write new index file"));
4126        }
4127
4128        return !!errs;
4129}