apply.con commit Merge branch 'sg/t5310-empty-input-fix' (a15bfa5)
   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
  10#include "cache.h"
  11#include "config.h"
  12#include "object-store.h"
  13#include "blob.h"
  14#include "delta.h"
  15#include "diff.h"
  16#include "dir.h"
  17#include "xdiff-interface.h"
  18#include "ll-merge.h"
  19#include "lockfile.h"
  20#include "parse-options.h"
  21#include "quote.h"
  22#include "rerere.h"
  23#include "apply.h"
  24
  25static void git_apply_config(void)
  26{
  27        git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
  28        git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
  29        git_config(git_default_config, NULL);
  30}
  31
  32static int parse_whitespace_option(struct apply_state *state, const char *option)
  33{
  34        if (!option) {
  35                state->ws_error_action = warn_on_ws_error;
  36                return 0;
  37        }
  38        if (!strcmp(option, "warn")) {
  39                state->ws_error_action = warn_on_ws_error;
  40                return 0;
  41        }
  42        if (!strcmp(option, "nowarn")) {
  43                state->ws_error_action = nowarn_ws_error;
  44                return 0;
  45        }
  46        if (!strcmp(option, "error")) {
  47                state->ws_error_action = die_on_ws_error;
  48                return 0;
  49        }
  50        if (!strcmp(option, "error-all")) {
  51                state->ws_error_action = die_on_ws_error;
  52                state->squelch_whitespace_errors = 0;
  53                return 0;
  54        }
  55        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
  56                state->ws_error_action = correct_ws_error;
  57                return 0;
  58        }
  59        return error(_("unrecognized whitespace option '%s'"), option);
  60}
  61
  62static int parse_ignorewhitespace_option(struct apply_state *state,
  63                                                 const char *option)
  64{
  65        if (!option || !strcmp(option, "no") ||
  66            !strcmp(option, "false") || !strcmp(option, "never") ||
  67            !strcmp(option, "none")) {
  68                state->ws_ignore_action = ignore_ws_none;
  69                return 0;
  70        }
  71        if (!strcmp(option, "change")) {
  72                state->ws_ignore_action = ignore_ws_change;
  73                return 0;
  74        }
  75        return error(_("unrecognized whitespace ignore option '%s'"), option);
  76}
  77
  78int init_apply_state(struct apply_state *state,
  79                     const char *prefix)
  80{
  81        memset(state, 0, sizeof(*state));
  82        state->prefix = prefix;
  83        state->apply = 1;
  84        state->line_termination = '\n';
  85        state->p_value = 1;
  86        state->p_context = UINT_MAX;
  87        state->squelch_whitespace_errors = 5;
  88        state->ws_error_action = warn_on_ws_error;
  89        state->ws_ignore_action = ignore_ws_none;
  90        state->linenr = 1;
  91        string_list_init(&state->fn_table, 0);
  92        string_list_init(&state->limit_by_name, 0);
  93        string_list_init(&state->symlink_changes, 0);
  94        strbuf_init(&state->root, 0);
  95
  96        git_apply_config();
  97        if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
  98                return -1;
  99        if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
 100                return -1;
 101        return 0;
 102}
 103
 104void clear_apply_state(struct apply_state *state)
 105{
 106        string_list_clear(&state->limit_by_name, 0);
 107        string_list_clear(&state->symlink_changes, 0);
 108        strbuf_release(&state->root);
 109
 110        /* &state->fn_table is cleared at the end of apply_patch() */
 111}
 112
 113static void mute_routine(const char *msg, va_list params)
 114{
 115        /* do nothing */
 116}
 117
 118int check_apply_state(struct apply_state *state, int force_apply)
 119{
 120        int is_not_gitdir = !startup_info->have_repository;
 121
 122        if (state->apply_with_reject && state->threeway)
 123                return error(_("--reject and --3way cannot be used together."));
 124        if (state->cached && state->threeway)
 125                return error(_("--cached and --3way cannot be used together."));
 126        if (state->threeway) {
 127                if (is_not_gitdir)
 128                        return error(_("--3way outside a repository"));
 129                state->check_index = 1;
 130        }
 131        if (state->apply_with_reject) {
 132                state->apply = 1;
 133                if (state->apply_verbosity == verbosity_normal)
 134                        state->apply_verbosity = verbosity_verbose;
 135        }
 136        if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
 137                state->apply = 0;
 138        if (state->check_index && is_not_gitdir)
 139                return error(_("--index outside a repository"));
 140        if (state->cached) {
 141                if (is_not_gitdir)
 142                        return error(_("--cached outside a repository"));
 143                state->check_index = 1;
 144        }
 145        if (state->ita_only && (state->check_index || is_not_gitdir))
 146                state->ita_only = 0;
 147        if (state->check_index)
 148                state->unsafe_paths = 0;
 149
 150        if (state->apply_verbosity <= verbosity_silent) {
 151                state->saved_error_routine = get_error_routine();
 152                state->saved_warn_routine = get_warn_routine();
 153                set_error_routine(mute_routine);
 154                set_warn_routine(mute_routine);
 155        }
 156
 157        return 0;
 158}
 159
 160static void set_default_whitespace_mode(struct apply_state *state)
 161{
 162        if (!state->whitespace_option && !apply_default_whitespace)
 163                state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
 164}
 165
 166/*
 167 * This represents one "hunk" from a patch, starting with
 168 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 169 * patch text is pointed at by patch, and its byte length
 170 * is stored in size.  leading and trailing are the number
 171 * of context lines.
 172 */
 173struct fragment {
 174        unsigned long leading, trailing;
 175        unsigned long oldpos, oldlines;
 176        unsigned long newpos, newlines;
 177        /*
 178         * 'patch' is usually borrowed from buf in apply_patch(),
 179         * but some codepaths store an allocated buffer.
 180         */
 181        const char *patch;
 182        unsigned free_patch:1,
 183                rejected:1;
 184        int size;
 185        int linenr;
 186        struct fragment *next;
 187};
 188
 189/*
 190 * When dealing with a binary patch, we reuse "leading" field
 191 * to store the type of the binary hunk, either deflated "delta"
 192 * or deflated "literal".
 193 */
 194#define binary_patch_method leading
 195#define BINARY_DELTA_DEFLATED   1
 196#define BINARY_LITERAL_DEFLATED 2
 197
 198/*
 199 * This represents a "patch" to a file, both metainfo changes
 200 * such as creation/deletion, filemode and content changes represented
 201 * as a series of fragments.
 202 */
 203struct patch {
 204        char *new_name, *old_name, *def_name;
 205        unsigned int old_mode, new_mode;
 206        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 207        int rejected;
 208        unsigned ws_rule;
 209        int lines_added, lines_deleted;
 210        int score;
 211        int extension_linenr; /* first line specifying delete/new/rename/copy */
 212        unsigned int is_toplevel_relative:1;
 213        unsigned int inaccurate_eof:1;
 214        unsigned int is_binary:1;
 215        unsigned int is_copy:1;
 216        unsigned int is_rename:1;
 217        unsigned int recount:1;
 218        unsigned int conflicted_threeway:1;
 219        unsigned int direct_to_threeway:1;
 220        unsigned int crlf_in_old:1;
 221        struct fragment *fragments;
 222        char *result;
 223        size_t resultsize;
 224        char old_sha1_prefix[41];
 225        char new_sha1_prefix[41];
 226        struct patch *next;
 227
 228        /* three-way fallback result */
 229        struct object_id threeway_stage[3];
 230};
 231
 232static void free_fragment_list(struct fragment *list)
 233{
 234        while (list) {
 235                struct fragment *next = list->next;
 236                if (list->free_patch)
 237                        free((char *)list->patch);
 238                free(list);
 239                list = next;
 240        }
 241}
 242
 243static void free_patch(struct patch *patch)
 244{
 245        free_fragment_list(patch->fragments);
 246        free(patch->def_name);
 247        free(patch->old_name);
 248        free(patch->new_name);
 249        free(patch->result);
 250        free(patch);
 251}
 252
 253static void free_patch_list(struct patch *list)
 254{
 255        while (list) {
 256                struct patch *next = list->next;
 257                free_patch(list);
 258                list = next;
 259        }
 260}
 261
 262/*
 263 * A line in a file, len-bytes long (includes the terminating LF,
 264 * except for an incomplete line at the end if the file ends with
 265 * one), and its contents hashes to 'hash'.
 266 */
 267struct line {
 268        size_t len;
 269        unsigned hash : 24;
 270        unsigned flag : 8;
 271#define LINE_COMMON     1
 272#define LINE_PATCHED    2
 273};
 274
 275/*
 276 * This represents a "file", which is an array of "lines".
 277 */
 278struct image {
 279        char *buf;
 280        size_t len;
 281        size_t nr;
 282        size_t alloc;
 283        struct line *line_allocated;
 284        struct line *line;
 285};
 286
 287static uint32_t hash_line(const char *cp, size_t len)
 288{
 289        size_t i;
 290        uint32_t h;
 291        for (i = 0, h = 0; i < len; i++) {
 292                if (!isspace(cp[i])) {
 293                        h = h * 3 + (cp[i] & 0xff);
 294                }
 295        }
 296        return h;
 297}
 298
 299/*
 300 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 301 * whitespace difference. Returns 1 if they match, 0 otherwise
 302 */
 303static int fuzzy_matchlines(const char *s1, size_t n1,
 304                            const char *s2, size_t n2)
 305{
 306        const char *end1 = s1 + n1;
 307        const char *end2 = s2 + n2;
 308
 309        /* ignore line endings */
 310        while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
 311                end1--;
 312        while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
 313                end2--;
 314
 315        while (s1 < end1 && s2 < end2) {
 316                if (isspace(*s1)) {
 317                        /*
 318                         * Skip whitespace. We check on both buffers
 319                         * because we don't want "a b" to match "ab".
 320                         */
 321                        if (!isspace(*s2))
 322                                return 0;
 323                        while (s1 < end1 && isspace(*s1))
 324                                s1++;
 325                        while (s2 < end2 && isspace(*s2))
 326                                s2++;
 327                } else if (*s1++ != *s2++)
 328                        return 0;
 329        }
 330
 331        /* If we reached the end on one side only, lines don't match. */
 332        return s1 == end1 && s2 == end2;
 333}
 334
 335static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 336{
 337        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 338        img->line_allocated[img->nr].len = len;
 339        img->line_allocated[img->nr].hash = hash_line(bol, len);
 340        img->line_allocated[img->nr].flag = flag;
 341        img->nr++;
 342}
 343
 344/*
 345 * "buf" has the file contents to be patched (read from various sources).
 346 * attach it to "image" and add line-based index to it.
 347 * "image" now owns the "buf".
 348 */
 349static void prepare_image(struct image *image, char *buf, size_t len,
 350                          int prepare_linetable)
 351{
 352        const char *cp, *ep;
 353
 354        memset(image, 0, sizeof(*image));
 355        image->buf = buf;
 356        image->len = len;
 357
 358        if (!prepare_linetable)
 359                return;
 360
 361        ep = image->buf + image->len;
 362        cp = image->buf;
 363        while (cp < ep) {
 364                const char *next;
 365                for (next = cp; next < ep && *next != '\n'; next++)
 366                        ;
 367                if (next < ep)
 368                        next++;
 369                add_line_info(image, cp, next - cp, 0);
 370                cp = next;
 371        }
 372        image->line = image->line_allocated;
 373}
 374
 375static void clear_image(struct image *image)
 376{
 377        free(image->buf);
 378        free(image->line_allocated);
 379        memset(image, 0, sizeof(*image));
 380}
 381
 382/* fmt must contain _one_ %s and no other substitution */
 383static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 384{
 385        struct strbuf sb = STRBUF_INIT;
 386
 387        if (patch->old_name && patch->new_name &&
 388            strcmp(patch->old_name, patch->new_name)) {
 389                quote_c_style(patch->old_name, &sb, NULL, 0);
 390                strbuf_addstr(&sb, " => ");
 391                quote_c_style(patch->new_name, &sb, NULL, 0);
 392        } else {
 393                const char *n = patch->new_name;
 394                if (!n)
 395                        n = patch->old_name;
 396                quote_c_style(n, &sb, NULL, 0);
 397        }
 398        fprintf(output, fmt, sb.buf);
 399        fputc('\n', output);
 400        strbuf_release(&sb);
 401}
 402
 403#define SLOP (16)
 404
 405static int read_patch_file(struct strbuf *sb, int fd)
 406{
 407        if (strbuf_read(sb, fd, 0) < 0)
 408                return error_errno("git apply: failed to read");
 409
 410        /*
 411         * Make sure that we have some slop in the buffer
 412         * so that we can do speculative "memcmp" etc, and
 413         * see to it that it is NUL-filled.
 414         */
 415        strbuf_grow(sb, SLOP);
 416        memset(sb->buf + sb->len, 0, SLOP);
 417        return 0;
 418}
 419
 420static unsigned long linelen(const char *buffer, unsigned long size)
 421{
 422        unsigned long len = 0;
 423        while (size--) {
 424                len++;
 425                if (*buffer++ == '\n')
 426                        break;
 427        }
 428        return len;
 429}
 430
 431static int is_dev_null(const char *str)
 432{
 433        return skip_prefix(str, "/dev/null", &str) && isspace(*str);
 434}
 435
 436#define TERM_SPACE      1
 437#define TERM_TAB        2
 438
 439static int name_terminate(int c, int terminate)
 440{
 441        if (c == ' ' && !(terminate & TERM_SPACE))
 442                return 0;
 443        if (c == '\t' && !(terminate & TERM_TAB))
 444                return 0;
 445
 446        return 1;
 447}
 448
 449/* remove double slashes to make --index work with such filenames */
 450static char *squash_slash(char *name)
 451{
 452        int i = 0, j = 0;
 453
 454        if (!name)
 455                return NULL;
 456
 457        while (name[i]) {
 458                if ((name[j++] = name[i++]) == '/')
 459                        while (name[i] == '/')
 460                                i++;
 461        }
 462        name[j] = '\0';
 463        return name;
 464}
 465
 466static char *find_name_gnu(struct apply_state *state,
 467                           const char *line,
 468                           const char *def,
 469                           int p_value)
 470{
 471        struct strbuf name = STRBUF_INIT;
 472        char *cp;
 473
 474        /*
 475         * Proposed "new-style" GNU patch/diff format; see
 476         * http://marc.info/?l=git&m=112927316408690&w=2
 477         */
 478        if (unquote_c_style(&name, line, NULL)) {
 479                strbuf_release(&name);
 480                return NULL;
 481        }
 482
 483        for (cp = name.buf; p_value; p_value--) {
 484                cp = strchr(cp, '/');
 485                if (!cp) {
 486                        strbuf_release(&name);
 487                        return NULL;
 488                }
 489                cp++;
 490        }
 491
 492        strbuf_remove(&name, 0, cp - name.buf);
 493        if (state->root.len)
 494                strbuf_insert(&name, 0, state->root.buf, state->root.len);
 495        return squash_slash(strbuf_detach(&name, NULL));
 496}
 497
 498static size_t sane_tz_len(const char *line, size_t len)
 499{
 500        const char *tz, *p;
 501
 502        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 503                return 0;
 504        tz = line + len - strlen(" +0500");
 505
 506        if (tz[1] != '+' && tz[1] != '-')
 507                return 0;
 508
 509        for (p = tz + 2; p != line + len; p++)
 510                if (!isdigit(*p))
 511                        return 0;
 512
 513        return line + len - tz;
 514}
 515
 516static size_t tz_with_colon_len(const char *line, size_t len)
 517{
 518        const char *tz, *p;
 519
 520        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 521                return 0;
 522        tz = line + len - strlen(" +08:00");
 523
 524        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 525                return 0;
 526        p = tz + 2;
 527        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 528            !isdigit(*p++) || !isdigit(*p++))
 529                return 0;
 530
 531        return line + len - tz;
 532}
 533
 534static size_t date_len(const char *line, size_t len)
 535{
 536        const char *date, *p;
 537
 538        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 539                return 0;
 540        p = date = line + len - strlen("72-02-05");
 541
 542        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 543            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 544            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 545                return 0;
 546
 547        if (date - line >= strlen("19") &&
 548            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 549                date -= strlen("19");
 550
 551        return line + len - date;
 552}
 553
 554static size_t short_time_len(const char *line, size_t len)
 555{
 556        const char *time, *p;
 557
 558        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 559                return 0;
 560        p = time = line + len - strlen(" 07:01:32");
 561
 562        /* Permit 1-digit hours? */
 563        if (*p++ != ' ' ||
 564            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 565            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 566            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 567                return 0;
 568
 569        return line + len - time;
 570}
 571
 572static size_t fractional_time_len(const char *line, size_t len)
 573{
 574        const char *p;
 575        size_t n;
 576
 577        /* Expected format: 19:41:17.620000023 */
 578        if (!len || !isdigit(line[len - 1]))
 579                return 0;
 580        p = line + len - 1;
 581
 582        /* Fractional seconds. */
 583        while (p > line && isdigit(*p))
 584                p--;
 585        if (*p != '.')
 586                return 0;
 587
 588        /* Hours, minutes, and whole seconds. */
 589        n = short_time_len(line, p - line);
 590        if (!n)
 591                return 0;
 592
 593        return line + len - p + n;
 594}
 595
 596static size_t trailing_spaces_len(const char *line, size_t len)
 597{
 598        const char *p;
 599
 600        /* Expected format: ' ' x (1 or more)  */
 601        if (!len || line[len - 1] != ' ')
 602                return 0;
 603
 604        p = line + len;
 605        while (p != line) {
 606                p--;
 607                if (*p != ' ')
 608                        return line + len - (p + 1);
 609        }
 610
 611        /* All spaces! */
 612        return len;
 613}
 614
 615static size_t diff_timestamp_len(const char *line, size_t len)
 616{
 617        const char *end = line + len;
 618        size_t n;
 619
 620        /*
 621         * Posix: 2010-07-05 19:41:17
 622         * GNU: 2010-07-05 19:41:17.620000023 -0500
 623         */
 624
 625        if (!isdigit(end[-1]))
 626                return 0;
 627
 628        n = sane_tz_len(line, end - line);
 629        if (!n)
 630                n = tz_with_colon_len(line, end - line);
 631        end -= n;
 632
 633        n = short_time_len(line, end - line);
 634        if (!n)
 635                n = fractional_time_len(line, end - line);
 636        end -= n;
 637
 638        n = date_len(line, end - line);
 639        if (!n) /* No date.  Too bad. */
 640                return 0;
 641        end -= n;
 642
 643        if (end == line)        /* No space before date. */
 644                return 0;
 645        if (end[-1] == '\t') {  /* Success! */
 646                end--;
 647                return line + len - end;
 648        }
 649        if (end[-1] != ' ')     /* No space before date. */
 650                return 0;
 651
 652        /* Whitespace damage. */
 653        end -= trailing_spaces_len(line, end - line);
 654        return line + len - end;
 655}
 656
 657static char *find_name_common(struct apply_state *state,
 658                              const char *line,
 659                              const char *def,
 660                              int p_value,
 661                              const char *end,
 662                              int terminate)
 663{
 664        int len;
 665        const char *start = NULL;
 666
 667        if (p_value == 0)
 668                start = line;
 669        while (line != end) {
 670                char c = *line;
 671
 672                if (!end && isspace(c)) {
 673                        if (c == '\n')
 674                                break;
 675                        if (name_terminate(c, terminate))
 676                                break;
 677                }
 678                line++;
 679                if (c == '/' && !--p_value)
 680                        start = line;
 681        }
 682        if (!start)
 683                return squash_slash(xstrdup_or_null(def));
 684        len = line - start;
 685        if (!len)
 686                return squash_slash(xstrdup_or_null(def));
 687
 688        /*
 689         * Generally we prefer the shorter name, especially
 690         * if the other one is just a variation of that with
 691         * something else tacked on to the end (ie "file.orig"
 692         * or "file~").
 693         */
 694        if (def) {
 695                int deflen = strlen(def);
 696                if (deflen < len && !strncmp(start, def, deflen))
 697                        return squash_slash(xstrdup(def));
 698        }
 699
 700        if (state->root.len) {
 701                char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
 702                return squash_slash(ret);
 703        }
 704
 705        return squash_slash(xmemdupz(start, len));
 706}
 707
 708static char *find_name(struct apply_state *state,
 709                       const char *line,
 710                       char *def,
 711                       int p_value,
 712                       int terminate)
 713{
 714        if (*line == '"') {
 715                char *name = find_name_gnu(state, line, def, p_value);
 716                if (name)
 717                        return name;
 718        }
 719
 720        return find_name_common(state, line, def, p_value, NULL, terminate);
 721}
 722
 723static char *find_name_traditional(struct apply_state *state,
 724                                   const char *line,
 725                                   char *def,
 726                                   int p_value)
 727{
 728        size_t len;
 729        size_t date_len;
 730
 731        if (*line == '"') {
 732                char *name = find_name_gnu(state, line, def, p_value);
 733                if (name)
 734                        return name;
 735        }
 736
 737        len = strchrnul(line, '\n') - line;
 738        date_len = diff_timestamp_len(line, len);
 739        if (!date_len)
 740                return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
 741        len -= date_len;
 742
 743        return find_name_common(state, line, def, p_value, line + len, 0);
 744}
 745
 746/*
 747 * Given the string after "--- " or "+++ ", guess the appropriate
 748 * p_value for the given patch.
 749 */
 750static int guess_p_value(struct apply_state *state, const char *nameline)
 751{
 752        char *name, *cp;
 753        int val = -1;
 754
 755        if (is_dev_null(nameline))
 756                return -1;
 757        name = find_name_traditional(state, nameline, NULL, 0);
 758        if (!name)
 759                return -1;
 760        cp = strchr(name, '/');
 761        if (!cp)
 762                val = 0;
 763        else if (state->prefix) {
 764                /*
 765                 * Does it begin with "a/$our-prefix" and such?  Then this is
 766                 * very likely to apply to our directory.
 767                 */
 768                if (starts_with(name, state->prefix))
 769                        val = count_slashes(state->prefix);
 770                else {
 771                        cp++;
 772                        if (starts_with(cp, state->prefix))
 773                                val = count_slashes(state->prefix) + 1;
 774                }
 775        }
 776        free(name);
 777        return val;
 778}
 779
 780/*
 781 * Does the ---/+++ line have the POSIX timestamp after the last HT?
 782 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 783 * this such a timestamp?
 784 */
 785static int has_epoch_timestamp(const char *nameline)
 786{
 787        /*
 788         * We are only interested in epoch timestamp; any non-zero
 789         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 790         * For the same reason, the date must be either 1969-12-31 or
 791         * 1970-01-01, and the seconds part must be "00".
 792         */
 793        const char stamp_regexp[] =
 794                "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
 795                " "
 796                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 797        const char *timestamp = NULL, *cp, *colon;
 798        static regex_t *stamp;
 799        regmatch_t m[10];
 800        int zoneoffset, epoch_hour, hour, minute;
 801        int status;
 802
 803        for (cp = nameline; *cp != '\n'; cp++) {
 804                if (*cp == '\t')
 805                        timestamp = cp + 1;
 806        }
 807        if (!timestamp)
 808                return 0;
 809
 810        /*
 811         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 812         * (west of GMT) or 1970-01-01 (east of GMT)
 813         */
 814        if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
 815                epoch_hour = 24;
 816        else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
 817                epoch_hour = 0;
 818        else
 819                return 0;
 820
 821        if (!stamp) {
 822                stamp = xmalloc(sizeof(*stamp));
 823                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 824                        warning(_("Cannot prepare timestamp regexp %s"),
 825                                stamp_regexp);
 826                        return 0;
 827                }
 828        }
 829
 830        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 831        if (status) {
 832                if (status != REG_NOMATCH)
 833                        warning(_("regexec returned %d for input: %s"),
 834                                status, timestamp);
 835                return 0;
 836        }
 837
 838        hour = strtol(timestamp, NULL, 10);
 839        minute = strtol(timestamp + m[1].rm_so, NULL, 10);
 840
 841        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 842        if (*colon == ':')
 843                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 844        else
 845                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 846        if (timestamp[m[3].rm_so] == '-')
 847                zoneoffset = -zoneoffset;
 848
 849        return hour * 60 + minute - zoneoffset == epoch_hour * 60;
 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 int parse_traditional_patch(struct apply_state *state,
 860                                   const char *first,
 861                                   const char *second,
 862                                   struct patch *patch)
 863{
 864        char *name;
 865
 866        first += 4;     /* skip "--- " */
 867        second += 4;    /* skip "+++ " */
 868        if (!state->p_value_known) {
 869                int p, q;
 870                p = guess_p_value(state, first);
 871                q = guess_p_value(state, second);
 872                if (p < 0) p = q;
 873                if (0 <= p && p == q) {
 874                        state->p_value = p;
 875                        state->p_value_known = 1;
 876                }
 877        }
 878        if (is_dev_null(first)) {
 879                patch->is_new = 1;
 880                patch->is_delete = 0;
 881                name = find_name_traditional(state, second, NULL, state->p_value);
 882                patch->new_name = name;
 883        } else if (is_dev_null(second)) {
 884                patch->is_new = 0;
 885                patch->is_delete = 1;
 886                name = find_name_traditional(state, first, NULL, state->p_value);
 887                patch->old_name = name;
 888        } else {
 889                char *first_name;
 890                first_name = find_name_traditional(state, first, NULL, state->p_value);
 891                name = find_name_traditional(state, second, first_name, state->p_value);
 892                free(first_name);
 893                if (has_epoch_timestamp(first)) {
 894                        patch->is_new = 1;
 895                        patch->is_delete = 0;
 896                        patch->new_name = name;
 897                } else if (has_epoch_timestamp(second)) {
 898                        patch->is_new = 0;
 899                        patch->is_delete = 1;
 900                        patch->old_name = name;
 901                } else {
 902                        patch->old_name = name;
 903                        patch->new_name = xstrdup_or_null(name);
 904                }
 905        }
 906        if (!name)
 907                return error(_("unable to find filename in patch at line %d"), state->linenr);
 908
 909        return 0;
 910}
 911
 912static int gitdiff_hdrend(struct apply_state *state,
 913                          const char *line,
 914                          struct patch *patch)
 915{
 916        return 1;
 917}
 918
 919/*
 920 * We're anal about diff header consistency, to make
 921 * sure that we don't end up having strange ambiguous
 922 * patches floating around.
 923 *
 924 * As a result, gitdiff_{old|new}name() will check
 925 * their names against any previous information, just
 926 * to make sure..
 927 */
 928#define DIFF_OLD_NAME 0
 929#define DIFF_NEW_NAME 1
 930
 931static int gitdiff_verify_name(struct apply_state *state,
 932                               const char *line,
 933                               int isnull,
 934                               char **name,
 935                               int side)
 936{
 937        if (!*name && !isnull) {
 938                *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
 939                return 0;
 940        }
 941
 942        if (*name) {
 943                char *another;
 944                if (isnull)
 945                        return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
 946                                     *name, state->linenr);
 947                another = find_name(state, line, NULL, state->p_value, TERM_TAB);
 948                if (!another || strcmp(another, *name)) {
 949                        free(another);
 950                        return error((side == DIFF_NEW_NAME) ?
 951                            _("git apply: bad git-diff - inconsistent new filename on line %d") :
 952                            _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
 953                }
 954                free(another);
 955        } else {
 956                if (!is_dev_null(line))
 957                        return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
 958        }
 959
 960        return 0;
 961}
 962
 963static int gitdiff_oldname(struct apply_state *state,
 964                           const char *line,
 965                           struct patch *patch)
 966{
 967        return gitdiff_verify_name(state, line,
 968                                   patch->is_new, &patch->old_name,
 969                                   DIFF_OLD_NAME);
 970}
 971
 972static int gitdiff_newname(struct apply_state *state,
 973                           const char *line,
 974                           struct patch *patch)
 975{
 976        return gitdiff_verify_name(state, line,
 977                                   patch->is_delete, &patch->new_name,
 978                                   DIFF_NEW_NAME);
 979}
 980
 981static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
 982{
 983        char *end;
 984        *mode = strtoul(line, &end, 8);
 985        if (end == line || !isspace(*end))
 986                return error(_("invalid mode on line %d: %s"), linenr, line);
 987        return 0;
 988}
 989
 990static int gitdiff_oldmode(struct apply_state *state,
 991                           const char *line,
 992                           struct patch *patch)
 993{
 994        return parse_mode_line(line, state->linenr, &patch->old_mode);
 995}
 996
 997static int gitdiff_newmode(struct apply_state *state,
 998                           const char *line,
 999                           struct patch *patch)
1000{
1001        return parse_mode_line(line, state->linenr, &patch->new_mode);
1002}
1003
1004static int gitdiff_delete(struct apply_state *state,
1005                          const char *line,
1006                          struct patch *patch)
1007{
1008        patch->is_delete = 1;
1009        free(patch->old_name);
1010        patch->old_name = xstrdup_or_null(patch->def_name);
1011        return gitdiff_oldmode(state, line, patch);
1012}
1013
1014static int gitdiff_newfile(struct apply_state *state,
1015                           const char *line,
1016                           struct patch *patch)
1017{
1018        patch->is_new = 1;
1019        free(patch->new_name);
1020        patch->new_name = xstrdup_or_null(patch->def_name);
1021        return gitdiff_newmode(state, line, patch);
1022}
1023
1024static int gitdiff_copysrc(struct apply_state *state,
1025                           const char *line,
1026                           struct patch *patch)
1027{
1028        patch->is_copy = 1;
1029        free(patch->old_name);
1030        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1031        return 0;
1032}
1033
1034static int gitdiff_copydst(struct apply_state *state,
1035                           const char *line,
1036                           struct patch *patch)
1037{
1038        patch->is_copy = 1;
1039        free(patch->new_name);
1040        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1041        return 0;
1042}
1043
1044static int gitdiff_renamesrc(struct apply_state *state,
1045                             const char *line,
1046                             struct patch *patch)
1047{
1048        patch->is_rename = 1;
1049        free(patch->old_name);
1050        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1051        return 0;
1052}
1053
1054static int gitdiff_renamedst(struct apply_state *state,
1055                             const char *line,
1056                             struct patch *patch)
1057{
1058        patch->is_rename = 1;
1059        free(patch->new_name);
1060        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1061        return 0;
1062}
1063
1064static int gitdiff_similarity(struct apply_state *state,
1065                              const char *line,
1066                              struct patch *patch)
1067{
1068        unsigned long val = strtoul(line, NULL, 10);
1069        if (val <= 100)
1070                patch->score = val;
1071        return 0;
1072}
1073
1074static int gitdiff_dissimilarity(struct apply_state *state,
1075                                 const char *line,
1076                                 struct patch *patch)
1077{
1078        unsigned long val = strtoul(line, NULL, 10);
1079        if (val <= 100)
1080                patch->score = val;
1081        return 0;
1082}
1083
1084static int gitdiff_index(struct apply_state *state,
1085                         const char *line,
1086                         struct patch *patch)
1087{
1088        /*
1089         * index line is N hexadecimal, "..", N hexadecimal,
1090         * and optional space with octal mode.
1091         */
1092        const char *ptr, *eol;
1093        int len;
1094
1095        ptr = strchr(line, '.');
1096        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1097                return 0;
1098        len = ptr - line;
1099        memcpy(patch->old_sha1_prefix, line, len);
1100        patch->old_sha1_prefix[len] = 0;
1101
1102        line = ptr + 2;
1103        ptr = strchr(line, ' ');
1104        eol = strchrnul(line, '\n');
1105
1106        if (!ptr || eol < ptr)
1107                ptr = eol;
1108        len = ptr - line;
1109
1110        if (40 < len)
1111                return 0;
1112        memcpy(patch->new_sha1_prefix, line, len);
1113        patch->new_sha1_prefix[len] = 0;
1114        if (*ptr == ' ')
1115                return gitdiff_oldmode(state, ptr + 1, patch);
1116        return 0;
1117}
1118
1119/*
1120 * This is normal for a diff that doesn't change anything: we'll fall through
1121 * into the next diff. Tell the parser to break out.
1122 */
1123static int gitdiff_unrecognized(struct apply_state *state,
1124                                const char *line,
1125                                struct patch *patch)
1126{
1127        return 1;
1128}
1129
1130/*
1131 * Skip p_value leading components from "line"; as we do not accept
1132 * absolute paths, return NULL in that case.
1133 */
1134static const char *skip_tree_prefix(struct apply_state *state,
1135                                    const char *line,
1136                                    int llen)
1137{
1138        int nslash;
1139        int i;
1140
1141        if (!state->p_value)
1142                return (llen && line[0] == '/') ? NULL : line;
1143
1144        nslash = state->p_value;
1145        for (i = 0; i < llen; i++) {
1146                int ch = line[i];
1147                if (ch == '/' && --nslash <= 0)
1148                        return (i == 0) ? NULL : &line[i + 1];
1149        }
1150        return NULL;
1151}
1152
1153/*
1154 * This is to extract the same name that appears on "diff --git"
1155 * line.  We do not find and return anything if it is a rename
1156 * patch, and it is OK because we will find the name elsewhere.
1157 * We need to reliably find name only when it is mode-change only,
1158 * creation or deletion of an empty file.  In any of these cases,
1159 * both sides are the same name under a/ and b/ respectively.
1160 */
1161static char *git_header_name(struct apply_state *state,
1162                             const char *line,
1163                             int llen)
1164{
1165        const char *name;
1166        const char *second = NULL;
1167        size_t len, line_len;
1168
1169        line += strlen("diff --git ");
1170        llen -= strlen("diff --git ");
1171
1172        if (*line == '"') {
1173                const char *cp;
1174                struct strbuf first = STRBUF_INIT;
1175                struct strbuf sp = STRBUF_INIT;
1176
1177                if (unquote_c_style(&first, line, &second))
1178                        goto free_and_fail1;
1179
1180                /* strip the a/b prefix including trailing slash */
1181                cp = skip_tree_prefix(state, first.buf, first.len);
1182                if (!cp)
1183                        goto free_and_fail1;
1184                strbuf_remove(&first, 0, cp - first.buf);
1185
1186                /*
1187                 * second points at one past closing dq of name.
1188                 * find the second name.
1189                 */
1190                while ((second < line + llen) && isspace(*second))
1191                        second++;
1192
1193                if (line + llen <= second)
1194                        goto free_and_fail1;
1195                if (*second == '"') {
1196                        if (unquote_c_style(&sp, second, NULL))
1197                                goto free_and_fail1;
1198                        cp = skip_tree_prefix(state, sp.buf, sp.len);
1199                        if (!cp)
1200                                goto free_and_fail1;
1201                        /* They must match, otherwise ignore */
1202                        if (strcmp(cp, first.buf))
1203                                goto free_and_fail1;
1204                        strbuf_release(&sp);
1205                        return strbuf_detach(&first, NULL);
1206                }
1207
1208                /* unquoted second */
1209                cp = skip_tree_prefix(state, second, line + llen - second);
1210                if (!cp)
1211                        goto free_and_fail1;
1212                if (line + llen - cp != first.len ||
1213                    memcmp(first.buf, cp, first.len))
1214                        goto free_and_fail1;
1215                return strbuf_detach(&first, NULL);
1216
1217        free_and_fail1:
1218                strbuf_release(&first);
1219                strbuf_release(&sp);
1220                return NULL;
1221        }
1222
1223        /* unquoted first name */
1224        name = skip_tree_prefix(state, line, llen);
1225        if (!name)
1226                return NULL;
1227
1228        /*
1229         * since the first name is unquoted, a dq if exists must be
1230         * the beginning of the second name.
1231         */
1232        for (second = name; second < line + llen; second++) {
1233                if (*second == '"') {
1234                        struct strbuf sp = STRBUF_INIT;
1235                        const char *np;
1236
1237                        if (unquote_c_style(&sp, second, NULL))
1238                                goto free_and_fail2;
1239
1240                        np = skip_tree_prefix(state, sp.buf, sp.len);
1241                        if (!np)
1242                                goto free_and_fail2;
1243
1244                        len = sp.buf + sp.len - np;
1245                        if (len < second - name &&
1246                            !strncmp(np, name, len) &&
1247                            isspace(name[len])) {
1248                                /* Good */
1249                                strbuf_remove(&sp, 0, np - sp.buf);
1250                                return strbuf_detach(&sp, NULL);
1251                        }
1252
1253                free_and_fail2:
1254                        strbuf_release(&sp);
1255                        return NULL;
1256                }
1257        }
1258
1259        /*
1260         * Accept a name only if it shows up twice, exactly the same
1261         * form.
1262         */
1263        second = strchr(name, '\n');
1264        if (!second)
1265                return NULL;
1266        line_len = second - name;
1267        for (len = 0 ; ; len++) {
1268                switch (name[len]) {
1269                default:
1270                        continue;
1271                case '\n':
1272                        return NULL;
1273                case '\t': case ' ':
1274                        /*
1275                         * Is this the separator between the preimage
1276                         * and the postimage pathname?  Again, we are
1277                         * only interested in the case where there is
1278                         * no rename, as this is only to set def_name
1279                         * and a rename patch has the names elsewhere
1280                         * in an unambiguous form.
1281                         */
1282                        if (!name[len + 1])
1283                                return NULL; /* no postimage name */
1284                        second = skip_tree_prefix(state, name + len + 1,
1285                                                  line_len - (len + 1));
1286                        if (!second)
1287                                return NULL;
1288                        /*
1289                         * Does len bytes starting at "name" and "second"
1290                         * (that are separated by one HT or SP we just
1291                         * found) exactly match?
1292                         */
1293                        if (second[len] == '\n' && !strncmp(name, second, len))
1294                                return xmemdupz(name, len);
1295                }
1296        }
1297}
1298
1299static int check_header_line(struct apply_state *state, struct patch *patch)
1300{
1301        int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1302                         (patch->is_rename == 1) + (patch->is_copy == 1);
1303        if (extensions > 1)
1304                return error(_("inconsistent header lines %d and %d"),
1305                             patch->extension_linenr, state->linenr);
1306        if (extensions && !patch->extension_linenr)
1307                patch->extension_linenr = state->linenr;
1308        return 0;
1309}
1310
1311/* Verify that we recognize the lines following a git header */
1312static int parse_git_header(struct apply_state *state,
1313                            const char *line,
1314                            int len,
1315                            unsigned int size,
1316                            struct patch *patch)
1317{
1318        unsigned long offset;
1319
1320        /* A git diff has explicit new/delete information, so we don't guess */
1321        patch->is_new = 0;
1322        patch->is_delete = 0;
1323
1324        /*
1325         * Some things may not have the old name in the
1326         * rest of the headers anywhere (pure mode changes,
1327         * or removing or adding empty files), so we get
1328         * the default name from the header.
1329         */
1330        patch->def_name = git_header_name(state, line, len);
1331        if (patch->def_name && state->root.len) {
1332                char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1333                free(patch->def_name);
1334                patch->def_name = s;
1335        }
1336
1337        line += len;
1338        size -= len;
1339        state->linenr++;
1340        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1341                static const struct opentry {
1342                        const char *str;
1343                        int (*fn)(struct apply_state *, const char *, struct patch *);
1344                } optable[] = {
1345                        { "@@ -", gitdiff_hdrend },
1346                        { "--- ", gitdiff_oldname },
1347                        { "+++ ", gitdiff_newname },
1348                        { "old mode ", gitdiff_oldmode },
1349                        { "new mode ", gitdiff_newmode },
1350                        { "deleted file mode ", gitdiff_delete },
1351                        { "new file mode ", gitdiff_newfile },
1352                        { "copy from ", gitdiff_copysrc },
1353                        { "copy to ", gitdiff_copydst },
1354                        { "rename old ", gitdiff_renamesrc },
1355                        { "rename new ", gitdiff_renamedst },
1356                        { "rename from ", gitdiff_renamesrc },
1357                        { "rename to ", gitdiff_renamedst },
1358                        { "similarity index ", gitdiff_similarity },
1359                        { "dissimilarity index ", gitdiff_dissimilarity },
1360                        { "index ", gitdiff_index },
1361                        { "", gitdiff_unrecognized },
1362                };
1363                int i;
1364
1365                len = linelen(line, size);
1366                if (!len || line[len-1] != '\n')
1367                        break;
1368                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1369                        const struct opentry *p = optable + i;
1370                        int oplen = strlen(p->str);
1371                        int res;
1372                        if (len < oplen || memcmp(p->str, line, oplen))
1373                                continue;
1374                        res = p->fn(state, line + oplen, patch);
1375                        if (res < 0)
1376                                return -1;
1377                        if (check_header_line(state, patch))
1378                                return -1;
1379                        if (res > 0)
1380                                return offset;
1381                        break;
1382                }
1383        }
1384
1385        return offset;
1386}
1387
1388static int parse_num(const char *line, unsigned long *p)
1389{
1390        char *ptr;
1391
1392        if (!isdigit(*line))
1393                return 0;
1394        *p = strtoul(line, &ptr, 10);
1395        return ptr - line;
1396}
1397
1398static int parse_range(const char *line, int len, int offset, const char *expect,
1399                       unsigned long *p1, unsigned long *p2)
1400{
1401        int digits, ex;
1402
1403        if (offset < 0 || offset >= len)
1404                return -1;
1405        line += offset;
1406        len -= offset;
1407
1408        digits = parse_num(line, p1);
1409        if (!digits)
1410                return -1;
1411
1412        offset += digits;
1413        line += digits;
1414        len -= digits;
1415
1416        *p2 = 1;
1417        if (*line == ',') {
1418                digits = parse_num(line+1, p2);
1419                if (!digits)
1420                        return -1;
1421
1422                offset += digits+1;
1423                line += digits+1;
1424                len -= digits+1;
1425        }
1426
1427        ex = strlen(expect);
1428        if (ex > len)
1429                return -1;
1430        if (memcmp(line, expect, ex))
1431                return -1;
1432
1433        return offset + ex;
1434}
1435
1436static void recount_diff(const char *line, int size, struct fragment *fragment)
1437{
1438        int oldlines = 0, newlines = 0, ret = 0;
1439
1440        if (size < 1) {
1441                warning("recount: ignore empty hunk");
1442                return;
1443        }
1444
1445        for (;;) {
1446                int len = linelen(line, size);
1447                size -= len;
1448                line += len;
1449
1450                if (size < 1)
1451                        break;
1452
1453                switch (*line) {
1454                case ' ': case '\n':
1455                        newlines++;
1456                        /* fall through */
1457                case '-':
1458                        oldlines++;
1459                        continue;
1460                case '+':
1461                        newlines++;
1462                        continue;
1463                case '\\':
1464                        continue;
1465                case '@':
1466                        ret = size < 3 || !starts_with(line, "@@ ");
1467                        break;
1468                case 'd':
1469                        ret = size < 5 || !starts_with(line, "diff ");
1470                        break;
1471                default:
1472                        ret = -1;
1473                        break;
1474                }
1475                if (ret) {
1476                        warning(_("recount: unexpected line: %.*s"),
1477                                (int)linelen(line, size), line);
1478                        return;
1479                }
1480                break;
1481        }
1482        fragment->oldlines = oldlines;
1483        fragment->newlines = newlines;
1484}
1485
1486/*
1487 * Parse a unified diff fragment header of the
1488 * form "@@ -a,b +c,d @@"
1489 */
1490static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1491{
1492        int offset;
1493
1494        if (!len || line[len-1] != '\n')
1495                return -1;
1496
1497        /* Figure out the number of lines in a fragment */
1498        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1499        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1500
1501        return offset;
1502}
1503
1504/*
1505 * Find file diff header
1506 *
1507 * Returns:
1508 *  -1 if no header was found
1509 *  -128 in case of error
1510 *   the size of the header in bytes (called "offset") otherwise
1511 */
1512static int find_header(struct apply_state *state,
1513                       const char *line,
1514                       unsigned long size,
1515                       int *hdrsize,
1516                       struct patch *patch)
1517{
1518        unsigned long offset, len;
1519
1520        patch->is_toplevel_relative = 0;
1521        patch->is_rename = patch->is_copy = 0;
1522        patch->is_new = patch->is_delete = -1;
1523        patch->old_mode = patch->new_mode = 0;
1524        patch->old_name = patch->new_name = NULL;
1525        for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1526                unsigned long nextlen;
1527
1528                len = linelen(line, size);
1529                if (!len)
1530                        break;
1531
1532                /* Testing this early allows us to take a few shortcuts.. */
1533                if (len < 6)
1534                        continue;
1535
1536                /*
1537                 * Make sure we don't find any unconnected patch fragments.
1538                 * That's a sign that we didn't find a header, and that a
1539                 * patch has become corrupted/broken up.
1540                 */
1541                if (!memcmp("@@ -", line, 4)) {
1542                        struct fragment dummy;
1543                        if (parse_fragment_header(line, len, &dummy) < 0)
1544                                continue;
1545                        error(_("patch fragment without header at line %d: %.*s"),
1546                                     state->linenr, (int)len-1, line);
1547                        return -128;
1548                }
1549
1550                if (size < len + 6)
1551                        break;
1552
1553                /*
1554                 * Git patch? It might not have a real patch, just a rename
1555                 * or mode change, so we handle that specially
1556                 */
1557                if (!memcmp("diff --git ", line, 11)) {
1558                        int git_hdr_len = parse_git_header(state, line, len, size, patch);
1559                        if (git_hdr_len < 0)
1560                                return -128;
1561                        if (git_hdr_len <= len)
1562                                continue;
1563                        if (!patch->old_name && !patch->new_name) {
1564                                if (!patch->def_name) {
1565                                        error(Q_("git diff header lacks filename information when removing "
1566                                                        "%d leading pathname component (line %d)",
1567                                                        "git diff header lacks filename information when removing "
1568                                                        "%d leading pathname components (line %d)",
1569                                                        state->p_value),
1570                                                     state->p_value, state->linenr);
1571                                        return -128;
1572                                }
1573                                patch->old_name = xstrdup(patch->def_name);
1574                                patch->new_name = xstrdup(patch->def_name);
1575                        }
1576                        if ((!patch->new_name && !patch->is_delete) ||
1577                            (!patch->old_name && !patch->is_new)) {
1578                                error(_("git diff header lacks filename information "
1579                                             "(line %d)"), state->linenr);
1580                                return -128;
1581                        }
1582                        patch->is_toplevel_relative = 1;
1583                        *hdrsize = git_hdr_len;
1584                        return offset;
1585                }
1586
1587                /* --- followed by +++ ? */
1588                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1589                        continue;
1590
1591                /*
1592                 * We only accept unified patches, so we want it to
1593                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1594                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1595                 */
1596                nextlen = linelen(line + len, size - len);
1597                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1598                        continue;
1599
1600                /* Ok, we'll consider it a patch */
1601                if (parse_traditional_patch(state, line, line+len, patch))
1602                        return -128;
1603                *hdrsize = len + nextlen;
1604                state->linenr += 2;
1605                return offset;
1606        }
1607        return -1;
1608}
1609
1610static void record_ws_error(struct apply_state *state,
1611                            unsigned result,
1612                            const char *line,
1613                            int len,
1614                            int linenr)
1615{
1616        char *err;
1617
1618        if (!result)
1619                return;
1620
1621        state->whitespace_error++;
1622        if (state->squelch_whitespace_errors &&
1623            state->squelch_whitespace_errors < state->whitespace_error)
1624                return;
1625
1626        err = whitespace_error_string(result);
1627        if (state->apply_verbosity > verbosity_silent)
1628                fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1629                        state->patch_input_file, linenr, err, len, line);
1630        free(err);
1631}
1632
1633static void check_whitespace(struct apply_state *state,
1634                             const char *line,
1635                             int len,
1636                             unsigned ws_rule)
1637{
1638        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1639
1640        record_ws_error(state, result, line + 1, len - 2, state->linenr);
1641}
1642
1643/*
1644 * Check if the patch has context lines with CRLF or
1645 * the patch wants to remove lines with CRLF.
1646 */
1647static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1648{
1649        if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1650                patch->ws_rule |= WS_CR_AT_EOL;
1651                patch->crlf_in_old = 1;
1652        }
1653}
1654
1655
1656/*
1657 * Parse a unified diff. Note that this really needs to parse each
1658 * fragment separately, since the only way to know the difference
1659 * between a "---" that is part of a patch, and a "---" that starts
1660 * the next patch is to look at the line counts..
1661 */
1662static int parse_fragment(struct apply_state *state,
1663                          const char *line,
1664                          unsigned long size,
1665                          struct patch *patch,
1666                          struct fragment *fragment)
1667{
1668        int added, deleted;
1669        int len = linelen(line, size), offset;
1670        unsigned long oldlines, newlines;
1671        unsigned long leading, trailing;
1672
1673        offset = parse_fragment_header(line, len, fragment);
1674        if (offset < 0)
1675                return -1;
1676        if (offset > 0 && patch->recount)
1677                recount_diff(line + offset, size - offset, fragment);
1678        oldlines = fragment->oldlines;
1679        newlines = fragment->newlines;
1680        leading = 0;
1681        trailing = 0;
1682
1683        /* Parse the thing.. */
1684        line += len;
1685        size -= len;
1686        state->linenr++;
1687        added = deleted = 0;
1688        for (offset = len;
1689             0 < size;
1690             offset += len, size -= len, line += len, state->linenr++) {
1691                if (!oldlines && !newlines)
1692                        break;
1693                len = linelen(line, size);
1694                if (!len || line[len-1] != '\n')
1695                        return -1;
1696                switch (*line) {
1697                default:
1698                        return -1;
1699                case '\n': /* newer GNU diff, an empty context line */
1700                case ' ':
1701                        oldlines--;
1702                        newlines--;
1703                        if (!deleted && !added)
1704                                leading++;
1705                        trailing++;
1706                        check_old_for_crlf(patch, line, len);
1707                        if (!state->apply_in_reverse &&
1708                            state->ws_error_action == correct_ws_error)
1709                                check_whitespace(state, line, len, patch->ws_rule);
1710                        break;
1711                case '-':
1712                        if (!state->apply_in_reverse)
1713                                check_old_for_crlf(patch, line, len);
1714                        if (state->apply_in_reverse &&
1715                            state->ws_error_action != nowarn_ws_error)
1716                                check_whitespace(state, line, len, patch->ws_rule);
1717                        deleted++;
1718                        oldlines--;
1719                        trailing = 0;
1720                        break;
1721                case '+':
1722                        if (state->apply_in_reverse)
1723                                check_old_for_crlf(patch, line, len);
1724                        if (!state->apply_in_reverse &&
1725                            state->ws_error_action != nowarn_ws_error)
1726                                check_whitespace(state, line, len, patch->ws_rule);
1727                        added++;
1728                        newlines--;
1729                        trailing = 0;
1730                        break;
1731
1732                /*
1733                 * We allow "\ No newline at end of file". Depending
1734                 * on locale settings when the patch was produced we
1735                 * don't know what this line looks like. The only
1736                 * thing we do know is that it begins with "\ ".
1737                 * Checking for 12 is just for sanity check -- any
1738                 * l10n of "\ No newline..." is at least that long.
1739                 */
1740                case '\\':
1741                        if (len < 12 || memcmp(line, "\\ ", 2))
1742                                return -1;
1743                        break;
1744                }
1745        }
1746        if (oldlines || newlines)
1747                return -1;
1748        if (!deleted && !added)
1749                return -1;
1750
1751        fragment->leading = leading;
1752        fragment->trailing = trailing;
1753
1754        /*
1755         * If a fragment ends with an incomplete line, we failed to include
1756         * it in the above loop because we hit oldlines == newlines == 0
1757         * before seeing it.
1758         */
1759        if (12 < size && !memcmp(line, "\\ ", 2))
1760                offset += linelen(line, size);
1761
1762        patch->lines_added += added;
1763        patch->lines_deleted += deleted;
1764
1765        if (0 < patch->is_new && oldlines)
1766                return error(_("new file depends on old contents"));
1767        if (0 < patch->is_delete && newlines)
1768                return error(_("deleted file still has contents"));
1769        return offset;
1770}
1771
1772/*
1773 * We have seen "diff --git a/... b/..." header (or a traditional patch
1774 * header).  Read hunks that belong to this patch into fragments and hang
1775 * them to the given patch structure.
1776 *
1777 * The (fragment->patch, fragment->size) pair points into the memory given
1778 * by the caller, not a copy, when we return.
1779 *
1780 * Returns:
1781 *   -1 in case of error,
1782 *   the number of bytes in the patch otherwise.
1783 */
1784static int parse_single_patch(struct apply_state *state,
1785                              const char *line,
1786                              unsigned long size,
1787                              struct patch *patch)
1788{
1789        unsigned long offset = 0;
1790        unsigned long oldlines = 0, newlines = 0, context = 0;
1791        struct fragment **fragp = &patch->fragments;
1792
1793        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1794                struct fragment *fragment;
1795                int len;
1796
1797                fragment = xcalloc(1, sizeof(*fragment));
1798                fragment->linenr = state->linenr;
1799                len = parse_fragment(state, line, size, patch, fragment);
1800                if (len <= 0) {
1801                        free(fragment);
1802                        return error(_("corrupt patch at line %d"), state->linenr);
1803                }
1804                fragment->patch = line;
1805                fragment->size = len;
1806                oldlines += fragment->oldlines;
1807                newlines += fragment->newlines;
1808                context += fragment->leading + fragment->trailing;
1809
1810                *fragp = fragment;
1811                fragp = &fragment->next;
1812
1813                offset += len;
1814                line += len;
1815                size -= len;
1816        }
1817
1818        /*
1819         * If something was removed (i.e. we have old-lines) it cannot
1820         * be creation, and if something was added it cannot be
1821         * deletion.  However, the reverse is not true; --unified=0
1822         * patches that only add are not necessarily creation even
1823         * though they do not have any old lines, and ones that only
1824         * delete are not necessarily deletion.
1825         *
1826         * Unfortunately, a real creation/deletion patch do _not_ have
1827         * any context line by definition, so we cannot safely tell it
1828         * apart with --unified=0 insanity.  At least if the patch has
1829         * more than one hunk it is not creation or deletion.
1830         */
1831        if (patch->is_new < 0 &&
1832            (oldlines || (patch->fragments && patch->fragments->next)))
1833                patch->is_new = 0;
1834        if (patch->is_delete < 0 &&
1835            (newlines || (patch->fragments && patch->fragments->next)))
1836                patch->is_delete = 0;
1837
1838        if (0 < patch->is_new && oldlines)
1839                return error(_("new file %s depends on old contents"), patch->new_name);
1840        if (0 < patch->is_delete && newlines)
1841                return error(_("deleted file %s still has contents"), patch->old_name);
1842        if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1843                fprintf_ln(stderr,
1844                           _("** warning: "
1845                             "file %s becomes empty but is not deleted"),
1846                           patch->new_name);
1847
1848        return offset;
1849}
1850
1851static inline int metadata_changes(struct patch *patch)
1852{
1853        return  patch->is_rename > 0 ||
1854                patch->is_copy > 0 ||
1855                patch->is_new > 0 ||
1856                patch->is_delete ||
1857                (patch->old_mode && patch->new_mode &&
1858                 patch->old_mode != patch->new_mode);
1859}
1860
1861static char *inflate_it(const void *data, unsigned long size,
1862                        unsigned long inflated_size)
1863{
1864        git_zstream stream;
1865        void *out;
1866        int st;
1867
1868        memset(&stream, 0, sizeof(stream));
1869
1870        stream.next_in = (unsigned char *)data;
1871        stream.avail_in = size;
1872        stream.next_out = out = xmalloc(inflated_size);
1873        stream.avail_out = inflated_size;
1874        git_inflate_init(&stream);
1875        st = git_inflate(&stream, Z_FINISH);
1876        git_inflate_end(&stream);
1877        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1878                free(out);
1879                return NULL;
1880        }
1881        return out;
1882}
1883
1884/*
1885 * Read a binary hunk and return a new fragment; fragment->patch
1886 * points at an allocated memory that the caller must free, so
1887 * it is marked as "->free_patch = 1".
1888 */
1889static struct fragment *parse_binary_hunk(struct apply_state *state,
1890                                          char **buf_p,
1891                                          unsigned long *sz_p,
1892                                          int *status_p,
1893                                          int *used_p)
1894{
1895        /*
1896         * Expect a line that begins with binary patch method ("literal"
1897         * or "delta"), followed by the length of data before deflating.
1898         * a sequence of 'length-byte' followed by base-85 encoded data
1899         * should follow, terminated by a newline.
1900         *
1901         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1902         * and we would limit the patch line to 66 characters,
1903         * so one line can fit up to 13 groups that would decode
1904         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1905         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1906         */
1907        int llen, used;
1908        unsigned long size = *sz_p;
1909        char *buffer = *buf_p;
1910        int patch_method;
1911        unsigned long origlen;
1912        char *data = NULL;
1913        int hunk_size = 0;
1914        struct fragment *frag;
1915
1916        llen = linelen(buffer, size);
1917        used = llen;
1918
1919        *status_p = 0;
1920
1921        if (starts_with(buffer, "delta ")) {
1922                patch_method = BINARY_DELTA_DEFLATED;
1923                origlen = strtoul(buffer + 6, NULL, 10);
1924        }
1925        else if (starts_with(buffer, "literal ")) {
1926                patch_method = BINARY_LITERAL_DEFLATED;
1927                origlen = strtoul(buffer + 8, NULL, 10);
1928        }
1929        else
1930                return NULL;
1931
1932        state->linenr++;
1933        buffer += llen;
1934        while (1) {
1935                int byte_length, max_byte_length, newsize;
1936                llen = linelen(buffer, size);
1937                used += llen;
1938                state->linenr++;
1939                if (llen == 1) {
1940                        /* consume the blank line */
1941                        buffer++;
1942                        size--;
1943                        break;
1944                }
1945                /*
1946                 * Minimum line is "A00000\n" which is 7-byte long,
1947                 * and the line length must be multiple of 5 plus 2.
1948                 */
1949                if ((llen < 7) || (llen-2) % 5)
1950                        goto corrupt;
1951                max_byte_length = (llen - 2) / 5 * 4;
1952                byte_length = *buffer;
1953                if ('A' <= byte_length && byte_length <= 'Z')
1954                        byte_length = byte_length - 'A' + 1;
1955                else if ('a' <= byte_length && byte_length <= 'z')
1956                        byte_length = byte_length - 'a' + 27;
1957                else
1958                        goto corrupt;
1959                /* if the input length was not multiple of 4, we would
1960                 * have filler at the end but the filler should never
1961                 * exceed 3 bytes
1962                 */
1963                if (max_byte_length < byte_length ||
1964                    byte_length <= max_byte_length - 4)
1965                        goto corrupt;
1966                newsize = hunk_size + byte_length;
1967                data = xrealloc(data, newsize);
1968                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1969                        goto corrupt;
1970                hunk_size = newsize;
1971                buffer += llen;
1972                size -= llen;
1973        }
1974
1975        frag = xcalloc(1, sizeof(*frag));
1976        frag->patch = inflate_it(data, hunk_size, origlen);
1977        frag->free_patch = 1;
1978        if (!frag->patch)
1979                goto corrupt;
1980        free(data);
1981        frag->size = origlen;
1982        *buf_p = buffer;
1983        *sz_p = size;
1984        *used_p = used;
1985        frag->binary_patch_method = patch_method;
1986        return frag;
1987
1988 corrupt:
1989        free(data);
1990        *status_p = -1;
1991        error(_("corrupt binary patch at line %d: %.*s"),
1992              state->linenr-1, llen-1, buffer);
1993        return NULL;
1994}
1995
1996/*
1997 * Returns:
1998 *   -1 in case of error,
1999 *   the length of the parsed binary patch otherwise
2000 */
2001static int parse_binary(struct apply_state *state,
2002                        char *buffer,
2003                        unsigned long size,
2004                        struct patch *patch)
2005{
2006        /*
2007         * We have read "GIT binary patch\n"; what follows is a line
2008         * that says the patch method (currently, either "literal" or
2009         * "delta") and the length of data before deflating; a
2010         * sequence of 'length-byte' followed by base-85 encoded data
2011         * follows.
2012         *
2013         * When a binary patch is reversible, there is another binary
2014         * hunk in the same format, starting with patch method (either
2015         * "literal" or "delta") with the length of data, and a sequence
2016         * of length-byte + base-85 encoded data, terminated with another
2017         * empty line.  This data, when applied to the postimage, produces
2018         * the preimage.
2019         */
2020        struct fragment *forward;
2021        struct fragment *reverse;
2022        int status;
2023        int used, used_1;
2024
2025        forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2026        if (!forward && !status)
2027                /* there has to be one hunk (forward hunk) */
2028                return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2029        if (status)
2030                /* otherwise we already gave an error message */
2031                return status;
2032
2033        reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2034        if (reverse)
2035                used += used_1;
2036        else if (status) {
2037                /*
2038                 * Not having reverse hunk is not an error, but having
2039                 * a corrupt reverse hunk is.
2040                 */
2041                free((void*) forward->patch);
2042                free(forward);
2043                return status;
2044        }
2045        forward->next = reverse;
2046        patch->fragments = forward;
2047        patch->is_binary = 1;
2048        return used;
2049}
2050
2051static void prefix_one(struct apply_state *state, char **name)
2052{
2053        char *old_name = *name;
2054        if (!old_name)
2055                return;
2056        *name = prefix_filename(state->prefix, *name);
2057        free(old_name);
2058}
2059
2060static void prefix_patch(struct apply_state *state, struct patch *p)
2061{
2062        if (!state->prefix || p->is_toplevel_relative)
2063                return;
2064        prefix_one(state, &p->new_name);
2065        prefix_one(state, &p->old_name);
2066}
2067
2068/*
2069 * include/exclude
2070 */
2071
2072static void add_name_limit(struct apply_state *state,
2073                           const char *name,
2074                           int exclude)
2075{
2076        struct string_list_item *it;
2077
2078        it = string_list_append(&state->limit_by_name, name);
2079        it->util = exclude ? NULL : (void *) 1;
2080}
2081
2082static int use_patch(struct apply_state *state, struct patch *p)
2083{
2084        const char *pathname = p->new_name ? p->new_name : p->old_name;
2085        int i;
2086
2087        /* Paths outside are not touched regardless of "--include" */
2088        if (state->prefix && *state->prefix) {
2089                const char *rest;
2090                if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2091                        return 0;
2092        }
2093
2094        /* See if it matches any of exclude/include rule */
2095        for (i = 0; i < state->limit_by_name.nr; i++) {
2096                struct string_list_item *it = &state->limit_by_name.items[i];
2097                if (!wildmatch(it->string, pathname, 0))
2098                        return (it->util != NULL);
2099        }
2100
2101        /*
2102         * If we had any include, a path that does not match any rule is
2103         * not used.  Otherwise, we saw bunch of exclude rules (or none)
2104         * and such a path is used.
2105         */
2106        return !state->has_include;
2107}
2108
2109/*
2110 * Read the patch text in "buffer" that extends for "size" bytes; stop
2111 * reading after seeing a single patch (i.e. changes to a single file).
2112 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2113 *
2114 * Returns:
2115 *   -1 if no header was found or parse_binary() failed,
2116 *   -128 on another error,
2117 *   the number of bytes consumed otherwise,
2118 *     so that the caller can call us again for the next patch.
2119 */
2120static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2121{
2122        int hdrsize, patchsize;
2123        int offset = find_header(state, buffer, size, &hdrsize, patch);
2124
2125        if (offset < 0)
2126                return offset;
2127
2128        prefix_patch(state, patch);
2129
2130        if (!use_patch(state, patch))
2131                patch->ws_rule = 0;
2132        else
2133                patch->ws_rule = whitespace_rule(patch->new_name
2134                                                 ? patch->new_name
2135                                                 : patch->old_name);
2136
2137        patchsize = parse_single_patch(state,
2138                                       buffer + offset + hdrsize,
2139                                       size - offset - hdrsize,
2140                                       patch);
2141
2142        if (patchsize < 0)
2143                return -128;
2144
2145        if (!patchsize) {
2146                static const char git_binary[] = "GIT binary patch\n";
2147                int hd = hdrsize + offset;
2148                unsigned long llen = linelen(buffer + hd, size - hd);
2149
2150                if (llen == sizeof(git_binary) - 1 &&
2151                    !memcmp(git_binary, buffer + hd, llen)) {
2152                        int used;
2153                        state->linenr++;
2154                        used = parse_binary(state, buffer + hd + llen,
2155                                            size - hd - llen, patch);
2156                        if (used < 0)
2157                                return -1;
2158                        if (used)
2159                                patchsize = used + llen;
2160                        else
2161                                patchsize = 0;
2162                }
2163                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2164                        static const char *binhdr[] = {
2165                                "Binary files ",
2166                                "Files ",
2167                                NULL,
2168                        };
2169                        int i;
2170                        for (i = 0; binhdr[i]; i++) {
2171                                int len = strlen(binhdr[i]);
2172                                if (len < size - hd &&
2173                                    !memcmp(binhdr[i], buffer + hd, len)) {
2174                                        state->linenr++;
2175                                        patch->is_binary = 1;
2176                                        patchsize = llen;
2177                                        break;
2178                                }
2179                        }
2180                }
2181
2182                /* Empty patch cannot be applied if it is a text patch
2183                 * without metadata change.  A binary patch appears
2184                 * empty to us here.
2185                 */
2186                if ((state->apply || state->check) &&
2187                    (!patch->is_binary && !metadata_changes(patch))) {
2188                        error(_("patch with only garbage at line %d"), state->linenr);
2189                        return -128;
2190                }
2191        }
2192
2193        return offset + hdrsize + patchsize;
2194}
2195
2196static void reverse_patches(struct patch *p)
2197{
2198        for (; p; p = p->next) {
2199                struct fragment *frag = p->fragments;
2200
2201                SWAP(p->new_name, p->old_name);
2202                SWAP(p->new_mode, p->old_mode);
2203                SWAP(p->is_new, p->is_delete);
2204                SWAP(p->lines_added, p->lines_deleted);
2205                SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2206
2207                for (; frag; frag = frag->next) {
2208                        SWAP(frag->newpos, frag->oldpos);
2209                        SWAP(frag->newlines, frag->oldlines);
2210                }
2211        }
2212}
2213
2214static const char pluses[] =
2215"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2216static const char minuses[]=
2217"----------------------------------------------------------------------";
2218
2219static void show_stats(struct apply_state *state, struct patch *patch)
2220{
2221        struct strbuf qname = STRBUF_INIT;
2222        char *cp = patch->new_name ? patch->new_name : patch->old_name;
2223        int max, add, del;
2224
2225        quote_c_style(cp, &qname, NULL, 0);
2226
2227        /*
2228         * "scale" the filename
2229         */
2230        max = state->max_len;
2231        if (max > 50)
2232                max = 50;
2233
2234        if (qname.len > max) {
2235                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2236                if (!cp)
2237                        cp = qname.buf + qname.len + 3 - max;
2238                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2239        }
2240
2241        if (patch->is_binary) {
2242                printf(" %-*s |  Bin\n", max, qname.buf);
2243                strbuf_release(&qname);
2244                return;
2245        }
2246
2247        printf(" %-*s |", max, qname.buf);
2248        strbuf_release(&qname);
2249
2250        /*
2251         * scale the add/delete
2252         */
2253        max = max + state->max_change > 70 ? 70 - max : state->max_change;
2254        add = patch->lines_added;
2255        del = patch->lines_deleted;
2256
2257        if (state->max_change > 0) {
2258                int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2259                add = (add * max + state->max_change / 2) / state->max_change;
2260                del = total - add;
2261        }
2262        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2263                add, pluses, del, minuses);
2264}
2265
2266static int read_old_data(struct stat *st, struct patch *patch,
2267                         const char *path, struct strbuf *buf)
2268{
2269        int conv_flags = patch->crlf_in_old ?
2270                CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2271        switch (st->st_mode & S_IFMT) {
2272        case S_IFLNK:
2273                if (strbuf_readlink(buf, path, st->st_size) < 0)
2274                        return error(_("unable to read symlink %s"), path);
2275                return 0;
2276        case S_IFREG:
2277                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2278                        return error(_("unable to open or read %s"), path);
2279                /*
2280                 * "git apply" without "--index/--cached" should never look
2281                 * at the index; the target file may not have been added to
2282                 * the index yet, and we may not even be in any Git repository.
2283                 * Pass NULL to convert_to_git() to stress this; the function
2284                 * should never look at the index when explicit crlf option
2285                 * is given.
2286                 */
2287                convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2288                return 0;
2289        default:
2290                return -1;
2291        }
2292}
2293
2294/*
2295 * Update the preimage, and the common lines in postimage,
2296 * from buffer buf of length len. If postlen is 0 the postimage
2297 * is updated in place, otherwise it's updated on a new buffer
2298 * of length postlen
2299 */
2300
2301static void update_pre_post_images(struct image *preimage,
2302                                   struct image *postimage,
2303                                   char *buf,
2304                                   size_t len, size_t postlen)
2305{
2306        int i, ctx, reduced;
2307        char *new_buf, *old_buf, *fixed;
2308        struct image fixed_preimage;
2309
2310        /*
2311         * Update the preimage with whitespace fixes.  Note that we
2312         * are not losing preimage->buf -- apply_one_fragment() will
2313         * free "oldlines".
2314         */
2315        prepare_image(&fixed_preimage, buf, len, 1);
2316        assert(postlen
2317               ? fixed_preimage.nr == preimage->nr
2318               : fixed_preimage.nr <= preimage->nr);
2319        for (i = 0; i < fixed_preimage.nr; i++)
2320                fixed_preimage.line[i].flag = preimage->line[i].flag;
2321        free(preimage->line_allocated);
2322        *preimage = fixed_preimage;
2323
2324        /*
2325         * Adjust the common context lines in postimage. This can be
2326         * done in-place when we are shrinking it with whitespace
2327         * fixing, but needs a new buffer when ignoring whitespace or
2328         * expanding leading tabs to spaces.
2329         *
2330         * We trust the caller to tell us if the update can be done
2331         * in place (postlen==0) or not.
2332         */
2333        old_buf = postimage->buf;
2334        if (postlen)
2335                new_buf = postimage->buf = xmalloc(postlen);
2336        else
2337                new_buf = old_buf;
2338        fixed = preimage->buf;
2339
2340        for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2341                size_t l_len = postimage->line[i].len;
2342                if (!(postimage->line[i].flag & LINE_COMMON)) {
2343                        /* an added line -- no counterparts in preimage */
2344                        memmove(new_buf, old_buf, l_len);
2345                        old_buf += l_len;
2346                        new_buf += l_len;
2347                        continue;
2348                }
2349
2350                /* a common context -- skip it in the original postimage */
2351                old_buf += l_len;
2352
2353                /* and find the corresponding one in the fixed preimage */
2354                while (ctx < preimage->nr &&
2355                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2356                        fixed += preimage->line[ctx].len;
2357                        ctx++;
2358                }
2359
2360                /*
2361                 * preimage is expected to run out, if the caller
2362                 * fixed addition of trailing blank lines.
2363                 */
2364                if (preimage->nr <= ctx) {
2365                        reduced++;
2366                        continue;
2367                }
2368
2369                /* and copy it in, while fixing the line length */
2370                l_len = preimage->line[ctx].len;
2371                memcpy(new_buf, fixed, l_len);
2372                new_buf += l_len;
2373                fixed += l_len;
2374                postimage->line[i].len = l_len;
2375                ctx++;
2376        }
2377
2378        if (postlen
2379            ? postlen < new_buf - postimage->buf
2380            : postimage->len < new_buf - postimage->buf)
2381                BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2382                    (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2383
2384        /* Fix the length of the whole thing */
2385        postimage->len = new_buf - postimage->buf;
2386        postimage->nr -= reduced;
2387}
2388
2389static int line_by_line_fuzzy_match(struct image *img,
2390                                    struct image *preimage,
2391                                    struct image *postimage,
2392                                    unsigned long current,
2393                                    int current_lno,
2394                                    int preimage_limit)
2395{
2396        int i;
2397        size_t imgoff = 0;
2398        size_t preoff = 0;
2399        size_t postlen = postimage->len;
2400        size_t extra_chars;
2401        char *buf;
2402        char *preimage_eof;
2403        char *preimage_end;
2404        struct strbuf fixed;
2405        char *fixed_buf;
2406        size_t fixed_len;
2407
2408        for (i = 0; i < preimage_limit; i++) {
2409                size_t prelen = preimage->line[i].len;
2410                size_t imglen = img->line[current_lno+i].len;
2411
2412                if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2413                                      preimage->buf + preoff, prelen))
2414                        return 0;
2415                if (preimage->line[i].flag & LINE_COMMON)
2416                        postlen += imglen - prelen;
2417                imgoff += imglen;
2418                preoff += prelen;
2419        }
2420
2421        /*
2422         * Ok, the preimage matches with whitespace fuzz.
2423         *
2424         * imgoff now holds the true length of the target that
2425         * matches the preimage before the end of the file.
2426         *
2427         * Count the number of characters in the preimage that fall
2428         * beyond the end of the file and make sure that all of them
2429         * are whitespace characters. (This can only happen if
2430         * we are removing blank lines at the end of the file.)
2431         */
2432        buf = preimage_eof = preimage->buf + preoff;
2433        for ( ; i < preimage->nr; i++)
2434                preoff += preimage->line[i].len;
2435        preimage_end = preimage->buf + preoff;
2436        for ( ; buf < preimage_end; buf++)
2437                if (!isspace(*buf))
2438                        return 0;
2439
2440        /*
2441         * Update the preimage and the common postimage context
2442         * lines to use the same whitespace as the target.
2443         * If whitespace is missing in the target (i.e.
2444         * if the preimage extends beyond the end of the file),
2445         * use the whitespace from the preimage.
2446         */
2447        extra_chars = preimage_end - preimage_eof;
2448        strbuf_init(&fixed, imgoff + extra_chars);
2449        strbuf_add(&fixed, img->buf + current, imgoff);
2450        strbuf_add(&fixed, preimage_eof, extra_chars);
2451        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2452        update_pre_post_images(preimage, postimage,
2453                               fixed_buf, fixed_len, postlen);
2454        return 1;
2455}
2456
2457static int match_fragment(struct apply_state *state,
2458                          struct image *img,
2459                          struct image *preimage,
2460                          struct image *postimage,
2461                          unsigned long current,
2462                          int current_lno,
2463                          unsigned ws_rule,
2464                          int match_beginning, int match_end)
2465{
2466        int i;
2467        char *fixed_buf, *buf, *orig, *target;
2468        struct strbuf fixed;
2469        size_t fixed_len, postlen;
2470        int preimage_limit;
2471
2472        if (preimage->nr + current_lno <= img->nr) {
2473                /*
2474                 * The hunk falls within the boundaries of img.
2475                 */
2476                preimage_limit = preimage->nr;
2477                if (match_end && (preimage->nr + current_lno != img->nr))
2478                        return 0;
2479        } else if (state->ws_error_action == correct_ws_error &&
2480                   (ws_rule & WS_BLANK_AT_EOF)) {
2481                /*
2482                 * This hunk extends beyond the end of img, and we are
2483                 * removing blank lines at the end of the file.  This
2484                 * many lines from the beginning of the preimage must
2485                 * match with img, and the remainder of the preimage
2486                 * must be blank.
2487                 */
2488                preimage_limit = img->nr - current_lno;
2489        } else {
2490                /*
2491                 * The hunk extends beyond the end of the img and
2492                 * we are not removing blanks at the end, so we
2493                 * should reject the hunk at this position.
2494                 */
2495                return 0;
2496        }
2497
2498        if (match_beginning && current_lno)
2499                return 0;
2500
2501        /* Quick hash check */
2502        for (i = 0; i < preimage_limit; i++)
2503                if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2504                    (preimage->line[i].hash != img->line[current_lno + i].hash))
2505                        return 0;
2506
2507        if (preimage_limit == preimage->nr) {
2508                /*
2509                 * Do we have an exact match?  If we were told to match
2510                 * at the end, size must be exactly at current+fragsize,
2511                 * otherwise current+fragsize must be still within the preimage,
2512                 * and either case, the old piece should match the preimage
2513                 * exactly.
2514                 */
2515                if ((match_end
2516                     ? (current + preimage->len == img->len)
2517                     : (current + preimage->len <= img->len)) &&
2518                    !memcmp(img->buf + current, preimage->buf, preimage->len))
2519                        return 1;
2520        } else {
2521                /*
2522                 * The preimage extends beyond the end of img, so
2523                 * there cannot be an exact match.
2524                 *
2525                 * There must be one non-blank context line that match
2526                 * a line before the end of img.
2527                 */
2528                char *buf_end;
2529
2530                buf = preimage->buf;
2531                buf_end = buf;
2532                for (i = 0; i < preimage_limit; i++)
2533                        buf_end += preimage->line[i].len;
2534
2535                for ( ; buf < buf_end; buf++)
2536                        if (!isspace(*buf))
2537                                break;
2538                if (buf == buf_end)
2539                        return 0;
2540        }
2541
2542        /*
2543         * No exact match. If we are ignoring whitespace, run a line-by-line
2544         * fuzzy matching. We collect all the line length information because
2545         * we need it to adjust whitespace if we match.
2546         */
2547        if (state->ws_ignore_action == ignore_ws_change)
2548                return line_by_line_fuzzy_match(img, preimage, postimage,
2549                                                current, current_lno, preimage_limit);
2550
2551        if (state->ws_error_action != correct_ws_error)
2552                return 0;
2553
2554        /*
2555         * The hunk does not apply byte-by-byte, but the hash says
2556         * it might with whitespace fuzz. We weren't asked to
2557         * ignore whitespace, we were asked to correct whitespace
2558         * errors, so let's try matching after whitespace correction.
2559         *
2560         * While checking the preimage against the target, whitespace
2561         * errors in both fixed, we count how large the corresponding
2562         * postimage needs to be.  The postimage prepared by
2563         * apply_one_fragment() has whitespace errors fixed on added
2564         * lines already, but the common lines were propagated as-is,
2565         * which may become longer when their whitespace errors are
2566         * fixed.
2567         */
2568
2569        /* First count added lines in postimage */
2570        postlen = 0;
2571        for (i = 0; i < postimage->nr; i++) {
2572                if (!(postimage->line[i].flag & LINE_COMMON))
2573                        postlen += postimage->line[i].len;
2574        }
2575
2576        /*
2577         * The preimage may extend beyond the end of the file,
2578         * but in this loop we will only handle the part of the
2579         * preimage that falls within the file.
2580         */
2581        strbuf_init(&fixed, preimage->len + 1);
2582        orig = preimage->buf;
2583        target = img->buf + current;
2584        for (i = 0; i < preimage_limit; i++) {
2585                size_t oldlen = preimage->line[i].len;
2586                size_t tgtlen = img->line[current_lno + i].len;
2587                size_t fixstart = fixed.len;
2588                struct strbuf tgtfix;
2589                int match;
2590
2591                /* Try fixing the line in the preimage */
2592                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2593
2594                /* Try fixing the line in the target */
2595                strbuf_init(&tgtfix, tgtlen);
2596                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2597
2598                /*
2599                 * If they match, either the preimage was based on
2600                 * a version before our tree fixed whitespace breakage,
2601                 * or we are lacking a whitespace-fix patch the tree
2602                 * the preimage was based on already had (i.e. target
2603                 * has whitespace breakage, the preimage doesn't).
2604                 * In either case, we are fixing the whitespace breakages
2605                 * so we might as well take the fix together with their
2606                 * real change.
2607                 */
2608                match = (tgtfix.len == fixed.len - fixstart &&
2609                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2610                                             fixed.len - fixstart));
2611
2612                /* Add the length if this is common with the postimage */
2613                if (preimage->line[i].flag & LINE_COMMON)
2614                        postlen += tgtfix.len;
2615
2616                strbuf_release(&tgtfix);
2617                if (!match)
2618                        goto unmatch_exit;
2619
2620                orig += oldlen;
2621                target += tgtlen;
2622        }
2623
2624
2625        /*
2626         * Now handle the lines in the preimage that falls beyond the
2627         * end of the file (if any). They will only match if they are
2628         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2629         * false).
2630         */
2631        for ( ; i < preimage->nr; i++) {
2632                size_t fixstart = fixed.len; /* start of the fixed preimage */
2633                size_t oldlen = preimage->line[i].len;
2634                int j;
2635
2636                /* Try fixing the line in the preimage */
2637                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2638
2639                for (j = fixstart; j < fixed.len; j++)
2640                        if (!isspace(fixed.buf[j]))
2641                                goto unmatch_exit;
2642
2643                orig += oldlen;
2644        }
2645
2646        /*
2647         * Yes, the preimage is based on an older version that still
2648         * has whitespace breakages unfixed, and fixing them makes the
2649         * hunk match.  Update the context lines in the postimage.
2650         */
2651        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2652        if (postlen < postimage->len)
2653                postlen = 0;
2654        update_pre_post_images(preimage, postimage,
2655                               fixed_buf, fixed_len, postlen);
2656        return 1;
2657
2658 unmatch_exit:
2659        strbuf_release(&fixed);
2660        return 0;
2661}
2662
2663static int find_pos(struct apply_state *state,
2664                    struct image *img,
2665                    struct image *preimage,
2666                    struct image *postimage,
2667                    int line,
2668                    unsigned ws_rule,
2669                    int match_beginning, int match_end)
2670{
2671        int i;
2672        unsigned long backwards, forwards, current;
2673        int backwards_lno, forwards_lno, current_lno;
2674
2675        /*
2676         * If match_beginning or match_end is specified, there is no
2677         * point starting from a wrong line that will never match and
2678         * wander around and wait for a match at the specified end.
2679         */
2680        if (match_beginning)
2681                line = 0;
2682        else if (match_end)
2683                line = img->nr - preimage->nr;
2684
2685        /*
2686         * Because the comparison is unsigned, the following test
2687         * will also take care of a negative line number that can
2688         * result when match_end and preimage is larger than the target.
2689         */
2690        if ((size_t) line > img->nr)
2691                line = img->nr;
2692
2693        current = 0;
2694        for (i = 0; i < line; i++)
2695                current += img->line[i].len;
2696
2697        /*
2698         * There's probably some smart way to do this, but I'll leave
2699         * that to the smart and beautiful people. I'm simple and stupid.
2700         */
2701        backwards = current;
2702        backwards_lno = line;
2703        forwards = current;
2704        forwards_lno = line;
2705        current_lno = line;
2706
2707        for (i = 0; ; i++) {
2708                if (match_fragment(state, img, preimage, postimage,
2709                                   current, current_lno, ws_rule,
2710                                   match_beginning, match_end))
2711                        return current_lno;
2712
2713        again:
2714                if (backwards_lno == 0 && forwards_lno == img->nr)
2715                        break;
2716
2717                if (i & 1) {
2718                        if (backwards_lno == 0) {
2719                                i++;
2720                                goto again;
2721                        }
2722                        backwards_lno--;
2723                        backwards -= img->line[backwards_lno].len;
2724                        current = backwards;
2725                        current_lno = backwards_lno;
2726                } else {
2727                        if (forwards_lno == img->nr) {
2728                                i++;
2729                                goto again;
2730                        }
2731                        forwards += img->line[forwards_lno].len;
2732                        forwards_lno++;
2733                        current = forwards;
2734                        current_lno = forwards_lno;
2735                }
2736
2737        }
2738        return -1;
2739}
2740
2741static void remove_first_line(struct image *img)
2742{
2743        img->buf += img->line[0].len;
2744        img->len -= img->line[0].len;
2745        img->line++;
2746        img->nr--;
2747}
2748
2749static void remove_last_line(struct image *img)
2750{
2751        img->len -= img->line[--img->nr].len;
2752}
2753
2754/*
2755 * The change from "preimage" and "postimage" has been found to
2756 * apply at applied_pos (counts in line numbers) in "img".
2757 * Update "img" to remove "preimage" and replace it with "postimage".
2758 */
2759static void update_image(struct apply_state *state,
2760                         struct image *img,
2761                         int applied_pos,
2762                         struct image *preimage,
2763                         struct image *postimage)
2764{
2765        /*
2766         * remove the copy of preimage at offset in img
2767         * and replace it with postimage
2768         */
2769        int i, nr;
2770        size_t remove_count, insert_count, applied_at = 0;
2771        char *result;
2772        int preimage_limit;
2773
2774        /*
2775         * If we are removing blank lines at the end of img,
2776         * the preimage may extend beyond the end.
2777         * If that is the case, we must be careful only to
2778         * remove the part of the preimage that falls within
2779         * the boundaries of img. Initialize preimage_limit
2780         * to the number of lines in the preimage that falls
2781         * within the boundaries.
2782         */
2783        preimage_limit = preimage->nr;
2784        if (preimage_limit > img->nr - applied_pos)
2785                preimage_limit = img->nr - applied_pos;
2786
2787        for (i = 0; i < applied_pos; i++)
2788                applied_at += img->line[i].len;
2789
2790        remove_count = 0;
2791        for (i = 0; i < preimage_limit; i++)
2792                remove_count += img->line[applied_pos + i].len;
2793        insert_count = postimage->len;
2794
2795        /* Adjust the contents */
2796        result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2797        memcpy(result, img->buf, applied_at);
2798        memcpy(result + applied_at, postimage->buf, postimage->len);
2799        memcpy(result + applied_at + postimage->len,
2800               img->buf + (applied_at + remove_count),
2801               img->len - (applied_at + remove_count));
2802        free(img->buf);
2803        img->buf = result;
2804        img->len += insert_count - remove_count;
2805        result[img->len] = '\0';
2806
2807        /* Adjust the line table */
2808        nr = img->nr + postimage->nr - preimage_limit;
2809        if (preimage_limit < postimage->nr) {
2810                /*
2811                 * NOTE: this knows that we never call remove_first_line()
2812                 * on anything other than pre/post image.
2813                 */
2814                REALLOC_ARRAY(img->line, nr);
2815                img->line_allocated = img->line;
2816        }
2817        if (preimage_limit != postimage->nr)
2818                MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2819                           img->line + applied_pos + preimage_limit,
2820                           img->nr - (applied_pos + preimage_limit));
2821        COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2822        if (!state->allow_overlap)
2823                for (i = 0; i < postimage->nr; i++)
2824                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2825        img->nr = nr;
2826}
2827
2828/*
2829 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2830 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2831 * replace the part of "img" with "postimage" text.
2832 */
2833static int apply_one_fragment(struct apply_state *state,
2834                              struct image *img, struct fragment *frag,
2835                              int inaccurate_eof, unsigned ws_rule,
2836                              int nth_fragment)
2837{
2838        int match_beginning, match_end;
2839        const char *patch = frag->patch;
2840        int size = frag->size;
2841        char *old, *oldlines;
2842        struct strbuf newlines;
2843        int new_blank_lines_at_end = 0;
2844        int found_new_blank_lines_at_end = 0;
2845        int hunk_linenr = frag->linenr;
2846        unsigned long leading, trailing;
2847        int pos, applied_pos;
2848        struct image preimage;
2849        struct image postimage;
2850
2851        memset(&preimage, 0, sizeof(preimage));
2852        memset(&postimage, 0, sizeof(postimage));
2853        oldlines = xmalloc(size);
2854        strbuf_init(&newlines, size);
2855
2856        old = oldlines;
2857        while (size > 0) {
2858                char first;
2859                int len = linelen(patch, size);
2860                int plen;
2861                int added_blank_line = 0;
2862                int is_blank_context = 0;
2863                size_t start;
2864
2865                if (!len)
2866                        break;
2867
2868                /*
2869                 * "plen" is how much of the line we should use for
2870                 * the actual patch data. Normally we just remove the
2871                 * first character on the line, but if the line is
2872                 * followed by "\ No newline", then we also remove the
2873                 * last one (which is the newline, of course).
2874                 */
2875                plen = len - 1;
2876                if (len < size && patch[len] == '\\')
2877                        plen--;
2878                first = *patch;
2879                if (state->apply_in_reverse) {
2880                        if (first == '-')
2881                                first = '+';
2882                        else if (first == '+')
2883                                first = '-';
2884                }
2885
2886                switch (first) {
2887                case '\n':
2888                        /* Newer GNU diff, empty context line */
2889                        if (plen < 0)
2890                                /* ... followed by '\No newline'; nothing */
2891                                break;
2892                        *old++ = '\n';
2893                        strbuf_addch(&newlines, '\n');
2894                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2895                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2896                        is_blank_context = 1;
2897                        break;
2898                case ' ':
2899                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2900                            ws_blank_line(patch + 1, plen, ws_rule))
2901                                is_blank_context = 1;
2902                        /* fallthrough */
2903                case '-':
2904                        memcpy(old, patch + 1, plen);
2905                        add_line_info(&preimage, old, plen,
2906                                      (first == ' ' ? LINE_COMMON : 0));
2907                        old += plen;
2908                        if (first == '-')
2909                                break;
2910                        /* fallthrough */
2911                case '+':
2912                        /* --no-add does not add new lines */
2913                        if (first == '+' && state->no_add)
2914                                break;
2915
2916                        start = newlines.len;
2917                        if (first != '+' ||
2918                            !state->whitespace_error ||
2919                            state->ws_error_action != correct_ws_error) {
2920                                strbuf_add(&newlines, patch + 1, plen);
2921                        }
2922                        else {
2923                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2924                        }
2925                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2926                                      (first == '+' ? 0 : LINE_COMMON));
2927                        if (first == '+' &&
2928                            (ws_rule & WS_BLANK_AT_EOF) &&
2929                            ws_blank_line(patch + 1, plen, ws_rule))
2930                                added_blank_line = 1;
2931                        break;
2932                case '@': case '\\':
2933                        /* Ignore it, we already handled it */
2934                        break;
2935                default:
2936                        if (state->apply_verbosity > verbosity_normal)
2937                                error(_("invalid start of line: '%c'"), first);
2938                        applied_pos = -1;
2939                        goto out;
2940                }
2941                if (added_blank_line) {
2942                        if (!new_blank_lines_at_end)
2943                                found_new_blank_lines_at_end = hunk_linenr;
2944                        new_blank_lines_at_end++;
2945                }
2946                else if (is_blank_context)
2947                        ;
2948                else
2949                        new_blank_lines_at_end = 0;
2950                patch += len;
2951                size -= len;
2952                hunk_linenr++;
2953        }
2954        if (inaccurate_eof &&
2955            old > oldlines && old[-1] == '\n' &&
2956            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2957                old--;
2958                strbuf_setlen(&newlines, newlines.len - 1);
2959                preimage.line_allocated[preimage.nr - 1].len--;
2960                postimage.line_allocated[postimage.nr - 1].len--;
2961        }
2962
2963        leading = frag->leading;
2964        trailing = frag->trailing;
2965
2966        /*
2967         * A hunk to change lines at the beginning would begin with
2968         * @@ -1,L +N,M @@
2969         * but we need to be careful.  -U0 that inserts before the second
2970         * line also has this pattern.
2971         *
2972         * And a hunk to add to an empty file would begin with
2973         * @@ -0,0 +N,M @@
2974         *
2975         * In other words, a hunk that is (frag->oldpos <= 1) with or
2976         * without leading context must match at the beginning.
2977         */
2978        match_beginning = (!frag->oldpos ||
2979                           (frag->oldpos == 1 && !state->unidiff_zero));
2980
2981        /*
2982         * A hunk without trailing lines must match at the end.
2983         * However, we simply cannot tell if a hunk must match end
2984         * from the lack of trailing lines if the patch was generated
2985         * with unidiff without any context.
2986         */
2987        match_end = !state->unidiff_zero && !trailing;
2988
2989        pos = frag->newpos ? (frag->newpos - 1) : 0;
2990        preimage.buf = oldlines;
2991        preimage.len = old - oldlines;
2992        postimage.buf = newlines.buf;
2993        postimage.len = newlines.len;
2994        preimage.line = preimage.line_allocated;
2995        postimage.line = postimage.line_allocated;
2996
2997        for (;;) {
2998
2999                applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3000                                       ws_rule, match_beginning, match_end);
3001
3002                if (applied_pos >= 0)
3003                        break;
3004
3005                /* Am I at my context limits? */
3006                if ((leading <= state->p_context) && (trailing <= state->p_context))
3007                        break;
3008                if (match_beginning || match_end) {
3009                        match_beginning = match_end = 0;
3010                        continue;
3011                }
3012
3013                /*
3014                 * Reduce the number of context lines; reduce both
3015                 * leading and trailing if they are equal otherwise
3016                 * just reduce the larger context.
3017                 */
3018                if (leading >= trailing) {
3019                        remove_first_line(&preimage);
3020                        remove_first_line(&postimage);
3021                        pos--;
3022                        leading--;
3023                }
3024                if (trailing > leading) {
3025                        remove_last_line(&preimage);
3026                        remove_last_line(&postimage);
3027                        trailing--;
3028                }
3029        }
3030
3031        if (applied_pos >= 0) {
3032                if (new_blank_lines_at_end &&
3033                    preimage.nr + applied_pos >= img->nr &&
3034                    (ws_rule & WS_BLANK_AT_EOF) &&
3035                    state->ws_error_action != nowarn_ws_error) {
3036                        record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3037                                        found_new_blank_lines_at_end);
3038                        if (state->ws_error_action == correct_ws_error) {
3039                                while (new_blank_lines_at_end--)
3040                                        remove_last_line(&postimage);
3041                        }
3042                        /*
3043                         * We would want to prevent write_out_results()
3044                         * from taking place in apply_patch() that follows
3045                         * the callchain led us here, which is:
3046                         * apply_patch->check_patch_list->check_patch->
3047                         * apply_data->apply_fragments->apply_one_fragment
3048                         */
3049                        if (state->ws_error_action == die_on_ws_error)
3050                                state->apply = 0;
3051                }
3052
3053                if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3054                        int offset = applied_pos - pos;
3055                        if (state->apply_in_reverse)
3056                                offset = 0 - offset;
3057                        fprintf_ln(stderr,
3058                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
3059                                      "Hunk #%d succeeded at %d (offset %d lines).",
3060                                      offset),
3061                                   nth_fragment, applied_pos + 1, offset);
3062                }
3063
3064                /*
3065                 * Warn if it was necessary to reduce the number
3066                 * of context lines.
3067                 */
3068                if ((leading != frag->leading ||
3069                     trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3070                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3071                                             " to apply fragment at %d"),
3072                                   leading, trailing, applied_pos+1);
3073                update_image(state, img, applied_pos, &preimage, &postimage);
3074        } else {
3075                if (state->apply_verbosity > verbosity_normal)
3076                        error(_("while searching for:\n%.*s"),
3077                              (int)(old - oldlines), oldlines);
3078        }
3079
3080out:
3081        free(oldlines);
3082        strbuf_release(&newlines);
3083        free(preimage.line_allocated);
3084        free(postimage.line_allocated);
3085
3086        return (applied_pos < 0);
3087}
3088
3089static int apply_binary_fragment(struct apply_state *state,
3090                                 struct image *img,
3091                                 struct patch *patch)
3092{
3093        struct fragment *fragment = patch->fragments;
3094        unsigned long len;
3095        void *dst;
3096
3097        if (!fragment)
3098                return error(_("missing binary patch data for '%s'"),
3099                             patch->new_name ?
3100                             patch->new_name :
3101                             patch->old_name);
3102
3103        /* Binary patch is irreversible without the optional second hunk */
3104        if (state->apply_in_reverse) {
3105                if (!fragment->next)
3106                        return error(_("cannot reverse-apply a binary patch "
3107                                       "without the reverse hunk to '%s'"),
3108                                     patch->new_name
3109                                     ? patch->new_name : patch->old_name);
3110                fragment = fragment->next;
3111        }
3112        switch (fragment->binary_patch_method) {
3113        case BINARY_DELTA_DEFLATED:
3114                dst = patch_delta(img->buf, img->len, fragment->patch,
3115                                  fragment->size, &len);
3116                if (!dst)
3117                        return -1;
3118                clear_image(img);
3119                img->buf = dst;
3120                img->len = len;
3121                return 0;
3122        case BINARY_LITERAL_DEFLATED:
3123                clear_image(img);
3124                img->len = fragment->size;
3125                img->buf = xmemdupz(fragment->patch, img->len);
3126                return 0;
3127        }
3128        return -1;
3129}
3130
3131/*
3132 * Replace "img" with the result of applying the binary patch.
3133 * The binary patch data itself in patch->fragment is still kept
3134 * but the preimage prepared by the caller in "img" is freed here
3135 * or in the helper function apply_binary_fragment() this calls.
3136 */
3137static int apply_binary(struct apply_state *state,
3138                        struct image *img,
3139                        struct patch *patch)
3140{
3141        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3142        struct object_id oid;
3143
3144        /*
3145         * For safety, we require patch index line to contain
3146         * full 40-byte textual SHA1 for old and new, at least for now.
3147         */
3148        if (strlen(patch->old_sha1_prefix) != 40 ||
3149            strlen(patch->new_sha1_prefix) != 40 ||
3150            get_oid_hex(patch->old_sha1_prefix, &oid) ||
3151            get_oid_hex(patch->new_sha1_prefix, &oid))
3152                return error(_("cannot apply binary patch to '%s' "
3153                               "without full index line"), name);
3154
3155        if (patch->old_name) {
3156                /*
3157                 * See if the old one matches what the patch
3158                 * applies to.
3159                 */
3160                hash_object_file(img->buf, img->len, blob_type, &oid);
3161                if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3162                        return error(_("the patch applies to '%s' (%s), "
3163                                       "which does not match the "
3164                                       "current contents."),
3165                                     name, oid_to_hex(&oid));
3166        }
3167        else {
3168                /* Otherwise, the old one must be empty. */
3169                if (img->len)
3170                        return error(_("the patch applies to an empty "
3171                                       "'%s' but it is not empty"), name);
3172        }
3173
3174        get_oid_hex(patch->new_sha1_prefix, &oid);
3175        if (is_null_oid(&oid)) {
3176                clear_image(img);
3177                return 0; /* deletion patch */
3178        }
3179
3180        if (has_sha1_file(oid.hash)) {
3181                /* We already have the postimage */
3182                enum object_type type;
3183                unsigned long size;
3184                char *result;
3185
3186                result = read_object_file(&oid, &type, &size);
3187                if (!result)
3188                        return error(_("the necessary postimage %s for "
3189                                       "'%s' cannot be read"),
3190                                     patch->new_sha1_prefix, name);
3191                clear_image(img);
3192                img->buf = result;
3193                img->len = size;
3194        } else {
3195                /*
3196                 * We have verified buf matches the preimage;
3197                 * apply the patch data to it, which is stored
3198                 * in the patch->fragments->{patch,size}.
3199                 */
3200                if (apply_binary_fragment(state, img, patch))
3201                        return error(_("binary patch does not apply to '%s'"),
3202                                     name);
3203
3204                /* verify that the result matches */
3205                hash_object_file(img->buf, img->len, blob_type, &oid);
3206                if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3207                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3208                                name, patch->new_sha1_prefix, oid_to_hex(&oid));
3209        }
3210
3211        return 0;
3212}
3213
3214static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3215{
3216        struct fragment *frag = patch->fragments;
3217        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3218        unsigned ws_rule = patch->ws_rule;
3219        unsigned inaccurate_eof = patch->inaccurate_eof;
3220        int nth = 0;
3221
3222        if (patch->is_binary)
3223                return apply_binary(state, img, patch);
3224
3225        while (frag) {
3226                nth++;
3227                if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3228                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
3229                        if (!state->apply_with_reject)
3230                                return -1;
3231                        frag->rejected = 1;
3232                }
3233                frag = frag->next;
3234        }
3235        return 0;
3236}
3237
3238static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3239{
3240        if (S_ISGITLINK(mode)) {
3241                strbuf_grow(buf, 100);
3242                strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3243        } else {
3244                enum object_type type;
3245                unsigned long sz;
3246                char *result;
3247
3248                result = read_object_file(oid, &type, &sz);
3249                if (!result)
3250                        return -1;
3251                /* XXX read_sha1_file NUL-terminates */
3252                strbuf_attach(buf, result, sz, sz + 1);
3253        }
3254        return 0;
3255}
3256
3257static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3258{
3259        if (!ce)
3260                return 0;
3261        return read_blob_object(buf, &ce->oid, ce->ce_mode);
3262}
3263
3264static struct patch *in_fn_table(struct apply_state *state, const char *name)
3265{
3266        struct string_list_item *item;
3267
3268        if (name == NULL)
3269                return NULL;
3270
3271        item = string_list_lookup(&state->fn_table, name);
3272        if (item != NULL)
3273                return (struct patch *)item->util;
3274
3275        return NULL;
3276}
3277
3278/*
3279 * item->util in the filename table records the status of the path.
3280 * Usually it points at a patch (whose result records the contents
3281 * of it after applying it), but it could be PATH_WAS_DELETED for a
3282 * path that a previously applied patch has already removed, or
3283 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3284 *
3285 * The latter is needed to deal with a case where two paths A and B
3286 * are swapped by first renaming A to B and then renaming B to A;
3287 * moving A to B should not be prevented due to presence of B as we
3288 * will remove it in a later patch.
3289 */
3290#define PATH_TO_BE_DELETED ((struct patch *) -2)
3291#define PATH_WAS_DELETED ((struct patch *) -1)
3292
3293static int to_be_deleted(struct patch *patch)
3294{
3295        return patch == PATH_TO_BE_DELETED;
3296}
3297
3298static int was_deleted(struct patch *patch)
3299{
3300        return patch == PATH_WAS_DELETED;
3301}
3302
3303static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3304{
3305        struct string_list_item *item;
3306
3307        /*
3308         * Always add new_name unless patch is a deletion
3309         * This should cover the cases for normal diffs,
3310         * file creations and copies
3311         */
3312        if (patch->new_name != NULL) {
3313                item = string_list_insert(&state->fn_table, patch->new_name);
3314                item->util = patch;
3315        }
3316
3317        /*
3318         * store a failure on rename/deletion cases because
3319         * later chunks shouldn't patch old names
3320         */
3321        if ((patch->new_name == NULL) || (patch->is_rename)) {
3322                item = string_list_insert(&state->fn_table, patch->old_name);
3323                item->util = PATH_WAS_DELETED;
3324        }
3325}
3326
3327static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3328{
3329        /*
3330         * store information about incoming file deletion
3331         */
3332        while (patch) {
3333                if ((patch->new_name == NULL) || (patch->is_rename)) {
3334                        struct string_list_item *item;
3335                        item = string_list_insert(&state->fn_table, patch->old_name);
3336                        item->util = PATH_TO_BE_DELETED;
3337                }
3338                patch = patch->next;
3339        }
3340}
3341
3342static int checkout_target(struct index_state *istate,
3343                           struct cache_entry *ce, struct stat *st)
3344{
3345        struct checkout costate = CHECKOUT_INIT;
3346
3347        costate.refresh_cache = 1;
3348        costate.istate = istate;
3349        if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3350                return error(_("cannot checkout %s"), ce->name);
3351        return 0;
3352}
3353
3354static struct patch *previous_patch(struct apply_state *state,
3355                                    struct patch *patch,
3356                                    int *gone)
3357{
3358        struct patch *previous;
3359
3360        *gone = 0;
3361        if (patch->is_copy || patch->is_rename)
3362                return NULL; /* "git" patches do not depend on the order */
3363
3364        previous = in_fn_table(state, patch->old_name);
3365        if (!previous)
3366                return NULL;
3367
3368        if (to_be_deleted(previous))
3369                return NULL; /* the deletion hasn't happened yet */
3370
3371        if (was_deleted(previous))
3372                *gone = 1;
3373
3374        return previous;
3375}
3376
3377static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3378{
3379        if (S_ISGITLINK(ce->ce_mode)) {
3380                if (!S_ISDIR(st->st_mode))
3381                        return -1;
3382                return 0;
3383        }
3384        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3385}
3386
3387#define SUBMODULE_PATCH_WITHOUT_INDEX 1
3388
3389static int load_patch_target(struct apply_state *state,
3390                             struct strbuf *buf,
3391                             const struct cache_entry *ce,
3392                             struct stat *st,
3393                             struct patch *patch,
3394                             const char *name,
3395                             unsigned expected_mode)
3396{
3397        if (state->cached || state->check_index) {
3398                if (read_file_or_gitlink(ce, buf))
3399                        return error(_("failed to read %s"), name);
3400        } else if (name) {
3401                if (S_ISGITLINK(expected_mode)) {
3402                        if (ce)
3403                                return read_file_or_gitlink(ce, buf);
3404                        else
3405                                return SUBMODULE_PATCH_WITHOUT_INDEX;
3406                } else if (has_symlink_leading_path(name, strlen(name))) {
3407                        return error(_("reading from '%s' beyond a symbolic link"), name);
3408                } else {
3409                        if (read_old_data(st, patch, name, buf))
3410                                return error(_("failed to read %s"), name);
3411                }
3412        }
3413        return 0;
3414}
3415
3416/*
3417 * We are about to apply "patch"; populate the "image" with the
3418 * current version we have, from the working tree or from the index,
3419 * depending on the situation e.g. --cached/--index.  If we are
3420 * applying a non-git patch that incrementally updates the tree,
3421 * we read from the result of a previous diff.
3422 */
3423static int load_preimage(struct apply_state *state,
3424                         struct image *image,
3425                         struct patch *patch, struct stat *st,
3426                         const struct cache_entry *ce)
3427{
3428        struct strbuf buf = STRBUF_INIT;
3429        size_t len;
3430        char *img;
3431        struct patch *previous;
3432        int status;
3433
3434        previous = previous_patch(state, patch, &status);
3435        if (status)
3436                return error(_("path %s has been renamed/deleted"),
3437                             patch->old_name);
3438        if (previous) {
3439                /* We have a patched copy in memory; use that. */
3440                strbuf_add(&buf, previous->result, previous->resultsize);
3441        } else {
3442                status = load_patch_target(state, &buf, ce, st, patch,
3443                                           patch->old_name, patch->old_mode);
3444                if (status < 0)
3445                        return status;
3446                else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3447                        /*
3448                         * There is no way to apply subproject
3449                         * patch without looking at the index.
3450                         * NEEDSWORK: shouldn't this be flagged
3451                         * as an error???
3452                         */
3453                        free_fragment_list(patch->fragments);
3454                        patch->fragments = NULL;
3455                } else if (status) {
3456                        return error(_("failed to read %s"), patch->old_name);
3457                }
3458        }
3459
3460        img = strbuf_detach(&buf, &len);
3461        prepare_image(image, img, len, !patch->is_binary);
3462        return 0;
3463}
3464
3465static int three_way_merge(struct image *image,
3466                           char *path,
3467                           const struct object_id *base,
3468                           const struct object_id *ours,
3469                           const struct object_id *theirs)
3470{
3471        mmfile_t base_file, our_file, their_file;
3472        mmbuffer_t result = { NULL };
3473        int status;
3474
3475        read_mmblob(&base_file, base);
3476        read_mmblob(&our_file, ours);
3477        read_mmblob(&their_file, theirs);
3478        status = ll_merge(&result, path,
3479                          &base_file, "base",
3480                          &our_file, "ours",
3481                          &their_file, "theirs", NULL);
3482        free(base_file.ptr);
3483        free(our_file.ptr);
3484        free(their_file.ptr);
3485        if (status < 0 || !result.ptr) {
3486                free(result.ptr);
3487                return -1;
3488        }
3489        clear_image(image);
3490        image->buf = result.ptr;
3491        image->len = result.size;
3492
3493        return status;
3494}
3495
3496/*
3497 * When directly falling back to add/add three-way merge, we read from
3498 * the current contents of the new_name.  In no cases other than that
3499 * this function will be called.
3500 */
3501static int load_current(struct apply_state *state,
3502                        struct image *image,
3503                        struct patch *patch)
3504{
3505        struct strbuf buf = STRBUF_INIT;
3506        int status, pos;
3507        size_t len;
3508        char *img;
3509        struct stat st;
3510        struct cache_entry *ce;
3511        char *name = patch->new_name;
3512        unsigned mode = patch->new_mode;
3513
3514        if (!patch->is_new)
3515                BUG("patch to %s is not a creation", patch->old_name);
3516
3517        pos = cache_name_pos(name, strlen(name));
3518        if (pos < 0)
3519                return error(_("%s: does not exist in index"), name);
3520        ce = active_cache[pos];
3521        if (lstat(name, &st)) {
3522                if (errno != ENOENT)
3523                        return error_errno("%s", name);
3524                if (checkout_target(&the_index, ce, &st))
3525                        return -1;
3526        }
3527        if (verify_index_match(ce, &st))
3528                return error(_("%s: does not match index"), name);
3529
3530        status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3531        if (status < 0)
3532                return status;
3533        else if (status)
3534                return -1;
3535        img = strbuf_detach(&buf, &len);
3536        prepare_image(image, img, len, !patch->is_binary);
3537        return 0;
3538}
3539
3540static int try_threeway(struct apply_state *state,
3541                        struct image *image,
3542                        struct patch *patch,
3543                        struct stat *st,
3544                        const struct cache_entry *ce)
3545{
3546        struct object_id pre_oid, post_oid, our_oid;
3547        struct strbuf buf = STRBUF_INIT;
3548        size_t len;
3549        int status;
3550        char *img;
3551        struct image tmp_image;
3552
3553        /* No point falling back to 3-way merge in these cases */
3554        if (patch->is_delete ||
3555            S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3556                return -1;
3557
3558        /* Preimage the patch was prepared for */
3559        if (patch->is_new)
3560                write_object_file("", 0, blob_type, &pre_oid);
3561        else if (get_oid(patch->old_sha1_prefix, &pre_oid) ||
3562                 read_blob_object(&buf, &pre_oid, patch->old_mode))
3563                return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3564
3565        if (state->apply_verbosity > verbosity_silent)
3566                fprintf(stderr, _("Falling back to three-way merge...\n"));
3567
3568        img = strbuf_detach(&buf, &len);
3569        prepare_image(&tmp_image, img, len, 1);
3570        /* Apply the patch to get the post image */
3571        if (apply_fragments(state, &tmp_image, patch) < 0) {
3572                clear_image(&tmp_image);
3573                return -1;
3574        }
3575        /* post_oid is theirs */
3576        write_object_file(tmp_image.buf, tmp_image.len, blob_type, &post_oid);
3577        clear_image(&tmp_image);
3578
3579        /* our_oid is ours */
3580        if (patch->is_new) {
3581                if (load_current(state, &tmp_image, patch))
3582                        return error(_("cannot read the current contents of '%s'"),
3583                                     patch->new_name);
3584        } else {
3585                if (load_preimage(state, &tmp_image, patch, st, ce))
3586                        return error(_("cannot read the current contents of '%s'"),
3587                                     patch->old_name);
3588        }
3589        write_object_file(tmp_image.buf, tmp_image.len, blob_type, &our_oid);
3590        clear_image(&tmp_image);
3591
3592        /* in-core three-way merge between post and our using pre as base */
3593        status = three_way_merge(image, patch->new_name,
3594                                 &pre_oid, &our_oid, &post_oid);
3595        if (status < 0) {
3596                if (state->apply_verbosity > verbosity_silent)
3597                        fprintf(stderr,
3598                                _("Failed to fall back on three-way merge...\n"));
3599                return status;
3600        }
3601
3602        if (status) {
3603                patch->conflicted_threeway = 1;
3604                if (patch->is_new)
3605                        oidclr(&patch->threeway_stage[0]);
3606                else
3607                        oidcpy(&patch->threeway_stage[0], &pre_oid);
3608                oidcpy(&patch->threeway_stage[1], &our_oid);
3609                oidcpy(&patch->threeway_stage[2], &post_oid);
3610                if (state->apply_verbosity > verbosity_silent)
3611                        fprintf(stderr,
3612                                _("Applied patch to '%s' with conflicts.\n"),
3613                                patch->new_name);
3614        } else {
3615                if (state->apply_verbosity > verbosity_silent)
3616                        fprintf(stderr,
3617                                _("Applied patch to '%s' cleanly.\n"),
3618                                patch->new_name);
3619        }
3620        return 0;
3621}
3622
3623static int apply_data(struct apply_state *state, struct patch *patch,
3624                      struct stat *st, const struct cache_entry *ce)
3625{
3626        struct image image;
3627
3628        if (load_preimage(state, &image, patch, st, ce) < 0)
3629                return -1;
3630
3631        if (patch->direct_to_threeway ||
3632            apply_fragments(state, &image, patch) < 0) {
3633                /* Note: with --reject, apply_fragments() returns 0 */
3634                if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3635                        return -1;
3636        }
3637        patch->result = image.buf;
3638        patch->resultsize = image.len;
3639        add_to_fn_table(state, patch);
3640        free(image.line_allocated);
3641
3642        if (0 < patch->is_delete && patch->resultsize)
3643                return error(_("removal patch leaves file contents"));
3644
3645        return 0;
3646}
3647
3648/*
3649 * If "patch" that we are looking at modifies or deletes what we have,
3650 * we would want it not to lose any local modification we have, either
3651 * in the working tree or in the index.
3652 *
3653 * This also decides if a non-git patch is a creation patch or a
3654 * modification to an existing empty file.  We do not check the state
3655 * of the current tree for a creation patch in this function; the caller
3656 * check_patch() separately makes sure (and errors out otherwise) that
3657 * the path the patch creates does not exist in the current tree.
3658 */
3659static int check_preimage(struct apply_state *state,
3660                          struct patch *patch,
3661                          struct cache_entry **ce,
3662                          struct stat *st)
3663{
3664        const char *old_name = patch->old_name;
3665        struct patch *previous = NULL;
3666        int stat_ret = 0, status;
3667        unsigned st_mode = 0;
3668
3669        if (!old_name)
3670                return 0;
3671
3672        assert(patch->is_new <= 0);
3673        previous = previous_patch(state, patch, &status);
3674
3675        if (status)
3676                return error(_("path %s has been renamed/deleted"), old_name);
3677        if (previous) {
3678                st_mode = previous->new_mode;
3679        } else if (!state->cached) {
3680                stat_ret = lstat(old_name, st);
3681                if (stat_ret && errno != ENOENT)
3682                        return error_errno("%s", old_name);
3683        }
3684
3685        if (state->check_index && !previous) {
3686                int pos = cache_name_pos(old_name, strlen(old_name));
3687                if (pos < 0) {
3688                        if (patch->is_new < 0)
3689                                goto is_new;
3690                        return error(_("%s: does not exist in index"), old_name);
3691                }
3692                *ce = active_cache[pos];
3693                if (stat_ret < 0) {
3694                        if (checkout_target(&the_index, *ce, st))
3695                                return -1;
3696                }
3697                if (!state->cached && verify_index_match(*ce, st))
3698                        return error(_("%s: does not match index"), old_name);
3699                if (state->cached)
3700                        st_mode = (*ce)->ce_mode;
3701        } else if (stat_ret < 0) {
3702                if (patch->is_new < 0)
3703                        goto is_new;
3704                return error_errno("%s", old_name);
3705        }
3706
3707        if (!state->cached && !previous)
3708                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3709
3710        if (patch->is_new < 0)
3711                patch->is_new = 0;
3712        if (!patch->old_mode)
3713                patch->old_mode = st_mode;
3714        if ((st_mode ^ patch->old_mode) & S_IFMT)
3715                return error(_("%s: wrong type"), old_name);
3716        if (st_mode != patch->old_mode)
3717                warning(_("%s has type %o, expected %o"),
3718                        old_name, st_mode, patch->old_mode);
3719        if (!patch->new_mode && !patch->is_delete)
3720                patch->new_mode = st_mode;
3721        return 0;
3722
3723 is_new:
3724        patch->is_new = 1;
3725        patch->is_delete = 0;
3726        FREE_AND_NULL(patch->old_name);
3727        return 0;
3728}
3729
3730
3731#define EXISTS_IN_INDEX 1
3732#define EXISTS_IN_WORKTREE 2
3733
3734static int check_to_create(struct apply_state *state,
3735                           const char *new_name,
3736                           int ok_if_exists)
3737{
3738        struct stat nst;
3739
3740        if (state->check_index &&
3741            cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3742            !ok_if_exists)
3743                return EXISTS_IN_INDEX;
3744        if (state->cached)
3745                return 0;
3746
3747        if (!lstat(new_name, &nst)) {
3748                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3749                        return 0;
3750                /*
3751                 * A leading component of new_name might be a symlink
3752                 * that is going to be removed with this patch, but
3753                 * still pointing at somewhere that has the path.
3754                 * In such a case, path "new_name" does not exist as
3755                 * far as git is concerned.
3756                 */
3757                if (has_symlink_leading_path(new_name, strlen(new_name)))
3758                        return 0;
3759
3760                return EXISTS_IN_WORKTREE;
3761        } else if (!is_missing_file_error(errno)) {
3762                return error_errno("%s", new_name);
3763        }
3764        return 0;
3765}
3766
3767static uintptr_t register_symlink_changes(struct apply_state *state,
3768                                          const char *path,
3769                                          uintptr_t what)
3770{
3771        struct string_list_item *ent;
3772
3773        ent = string_list_lookup(&state->symlink_changes, path);
3774        if (!ent) {
3775                ent = string_list_insert(&state->symlink_changes, path);
3776                ent->util = (void *)0;
3777        }
3778        ent->util = (void *)(what | ((uintptr_t)ent->util));
3779        return (uintptr_t)ent->util;
3780}
3781
3782static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3783{
3784        struct string_list_item *ent;
3785
3786        ent = string_list_lookup(&state->symlink_changes, path);
3787        if (!ent)
3788                return 0;
3789        return (uintptr_t)ent->util;
3790}
3791
3792static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3793{
3794        for ( ; patch; patch = patch->next) {
3795                if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3796                    (patch->is_rename || patch->is_delete))
3797                        /* the symlink at patch->old_name is removed */
3798                        register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3799
3800                if (patch->new_name && S_ISLNK(patch->new_mode))
3801                        /* the symlink at patch->new_name is created or remains */
3802                        register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3803        }
3804}
3805
3806static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3807{
3808        do {
3809                unsigned int change;
3810
3811                while (--name->len && name->buf[name->len] != '/')
3812                        ; /* scan backwards */
3813                if (!name->len)
3814                        break;
3815                name->buf[name->len] = '\0';
3816                change = check_symlink_changes(state, name->buf);
3817                if (change & APPLY_SYMLINK_IN_RESULT)
3818                        return 1;
3819                if (change & APPLY_SYMLINK_GOES_AWAY)
3820                        /*
3821                         * This cannot be "return 0", because we may
3822                         * see a new one created at a higher level.
3823                         */
3824                        continue;
3825
3826                /* otherwise, check the preimage */
3827                if (state->check_index) {
3828                        struct cache_entry *ce;
3829
3830                        ce = cache_file_exists(name->buf, name->len, ignore_case);
3831                        if (ce && S_ISLNK(ce->ce_mode))
3832                                return 1;
3833                } else {
3834                        struct stat st;
3835                        if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3836                                return 1;
3837                }
3838        } while (1);
3839        return 0;
3840}
3841
3842static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3843{
3844        int ret;
3845        struct strbuf name = STRBUF_INIT;
3846
3847        assert(*name_ != '\0');
3848        strbuf_addstr(&name, name_);
3849        ret = path_is_beyond_symlink_1(state, &name);
3850        strbuf_release(&name);
3851
3852        return ret;
3853}
3854
3855static int check_unsafe_path(struct patch *patch)
3856{
3857        const char *old_name = NULL;
3858        const char *new_name = NULL;
3859        if (patch->is_delete)
3860                old_name = patch->old_name;
3861        else if (!patch->is_new && !patch->is_copy)
3862                old_name = patch->old_name;
3863        if (!patch->is_delete)
3864                new_name = patch->new_name;
3865
3866        if (old_name && !verify_path(old_name, patch->old_mode))
3867                return error(_("invalid path '%s'"), old_name);
3868        if (new_name && !verify_path(new_name, patch->new_mode))
3869                return error(_("invalid path '%s'"), new_name);
3870        return 0;
3871}
3872
3873/*
3874 * Check and apply the patch in-core; leave the result in patch->result
3875 * for the caller to write it out to the final destination.
3876 */
3877static int check_patch(struct apply_state *state, struct patch *patch)
3878{
3879        struct stat st;
3880        const char *old_name = patch->old_name;
3881        const char *new_name = patch->new_name;
3882        const char *name = old_name ? old_name : new_name;
3883        struct cache_entry *ce = NULL;
3884        struct patch *tpatch;
3885        int ok_if_exists;
3886        int status;
3887
3888        patch->rejected = 1; /* we will drop this after we succeed */
3889
3890        status = check_preimage(state, patch, &ce, &st);
3891        if (status)
3892                return status;
3893        old_name = patch->old_name;
3894
3895        /*
3896         * A type-change diff is always split into a patch to delete
3897         * old, immediately followed by a patch to create new (see
3898         * diff.c::run_diff()); in such a case it is Ok that the entry
3899         * to be deleted by the previous patch is still in the working
3900         * tree and in the index.
3901         *
3902         * A patch to swap-rename between A and B would first rename A
3903         * to B and then rename B to A.  While applying the first one,
3904         * the presence of B should not stop A from getting renamed to
3905         * B; ask to_be_deleted() about the later rename.  Removal of
3906         * B and rename from A to B is handled the same way by asking
3907         * was_deleted().
3908         */
3909        if ((tpatch = in_fn_table(state, new_name)) &&
3910            (was_deleted(tpatch) || to_be_deleted(tpatch)))
3911                ok_if_exists = 1;
3912        else
3913                ok_if_exists = 0;
3914
3915        if (new_name &&
3916            ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3917                int err = check_to_create(state, new_name, ok_if_exists);
3918
3919                if (err && state->threeway) {
3920                        patch->direct_to_threeway = 1;
3921                } else switch (err) {
3922                case 0:
3923                        break; /* happy */
3924                case EXISTS_IN_INDEX:
3925                        return error(_("%s: already exists in index"), new_name);
3926                        break;
3927                case EXISTS_IN_WORKTREE:
3928                        return error(_("%s: already exists in working directory"),
3929                                     new_name);
3930                default:
3931                        return err;
3932                }
3933
3934                if (!patch->new_mode) {
3935                        if (0 < patch->is_new)
3936                                patch->new_mode = S_IFREG | 0644;
3937                        else
3938                                patch->new_mode = patch->old_mode;
3939                }
3940        }
3941
3942        if (new_name && old_name) {
3943                int same = !strcmp(old_name, new_name);
3944                if (!patch->new_mode)
3945                        patch->new_mode = patch->old_mode;
3946                if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3947                        if (same)
3948                                return error(_("new mode (%o) of %s does not "
3949                                               "match old mode (%o)"),
3950                                        patch->new_mode, new_name,
3951                                        patch->old_mode);
3952                        else
3953                                return error(_("new mode (%o) of %s does not "
3954                                               "match old mode (%o) of %s"),
3955                                        patch->new_mode, new_name,
3956                                        patch->old_mode, old_name);
3957                }
3958        }
3959
3960        if (!state->unsafe_paths && check_unsafe_path(patch))
3961                return -128;
3962
3963        /*
3964         * An attempt to read from or delete a path that is beyond a
3965         * symbolic link will be prevented by load_patch_target() that
3966         * is called at the beginning of apply_data() so we do not
3967         * have to worry about a patch marked with "is_delete" bit
3968         * here.  We however need to make sure that the patch result
3969         * is not deposited to a path that is beyond a symbolic link
3970         * here.
3971         */
3972        if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3973                return error(_("affected file '%s' is beyond a symbolic link"),
3974                             patch->new_name);
3975
3976        if (apply_data(state, patch, &st, ce) < 0)
3977                return error(_("%s: patch does not apply"), name);
3978        patch->rejected = 0;
3979        return 0;
3980}
3981
3982static int check_patch_list(struct apply_state *state, struct patch *patch)
3983{
3984        int err = 0;
3985
3986        prepare_symlink_changes(state, patch);
3987        prepare_fn_table(state, patch);
3988        while (patch) {
3989                int res;
3990                if (state->apply_verbosity > verbosity_normal)
3991                        say_patch_name(stderr,
3992                                       _("Checking patch %s..."), patch);
3993                res = check_patch(state, patch);
3994                if (res == -128)
3995                        return -128;
3996                err |= res;
3997                patch = patch->next;
3998        }
3999        return err;
4000}
4001
4002static int read_apply_cache(struct apply_state *state)
4003{
4004        if (state->index_file)
4005                return read_cache_from(state->index_file);
4006        else
4007                return read_cache();
4008}
4009
4010/* This function tries to read the object name from the current index */
4011static int get_current_oid(struct apply_state *state, const char *path,
4012                           struct object_id *oid)
4013{
4014        int pos;
4015
4016        if (read_apply_cache(state) < 0)
4017                return -1;
4018        pos = cache_name_pos(path, strlen(path));
4019        if (pos < 0)
4020                return -1;
4021        oidcpy(oid, &active_cache[pos]->oid);
4022        return 0;
4023}
4024
4025static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4026{
4027        /*
4028         * A usable gitlink patch has only one fragment (hunk) that looks like:
4029         * @@ -1 +1 @@
4030         * -Subproject commit <old sha1>
4031         * +Subproject commit <new sha1>
4032         * or
4033         * @@ -1 +0,0 @@
4034         * -Subproject commit <old sha1>
4035         * for a removal patch.
4036         */
4037        struct fragment *hunk = p->fragments;
4038        static const char heading[] = "-Subproject commit ";
4039        char *preimage;
4040
4041        if (/* does the patch have only one hunk? */
4042            hunk && !hunk->next &&
4043            /* is its preimage one line? */
4044            hunk->oldpos == 1 && hunk->oldlines == 1 &&
4045            /* does preimage begin with the heading? */
4046            (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4047            starts_with(++preimage, heading) &&
4048            /* does it record full SHA-1? */
4049            !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4050            preimage[sizeof(heading) + GIT_SHA1_HEXSZ - 1] == '\n' &&
4051            /* does the abbreviated name on the index line agree with it? */
4052            starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
4053                return 0; /* it all looks fine */
4054
4055        /* we may have full object name on the index line */
4056        return get_oid_hex(p->old_sha1_prefix, oid);
4057}
4058
4059/* Build an index that contains just the files needed for a 3way merge */
4060static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4061{
4062        struct patch *patch;
4063        struct index_state result = { NULL };
4064        struct lock_file lock = LOCK_INIT;
4065        int res;
4066
4067        /* Once we start supporting the reverse patch, it may be
4068         * worth showing the new sha1 prefix, but until then...
4069         */
4070        for (patch = list; patch; patch = patch->next) {
4071                struct object_id oid;
4072                struct cache_entry *ce;
4073                const char *name;
4074
4075                name = patch->old_name ? patch->old_name : patch->new_name;
4076                if (0 < patch->is_new)
4077                        continue;
4078
4079                if (S_ISGITLINK(patch->old_mode)) {
4080                        if (!preimage_oid_in_gitlink_patch(patch, &oid))
4081                                ; /* ok, the textual part looks sane */
4082                        else
4083                                return error(_("sha1 information is lacking or "
4084                                               "useless for submodule %s"), name);
4085                } else if (!get_oid_blob(patch->old_sha1_prefix, &oid)) {
4086                        ; /* ok */
4087                } else if (!patch->lines_added && !patch->lines_deleted) {
4088                        /* mode-only change: update the current */
4089                        if (get_current_oid(state, patch->old_name, &oid))
4090                                return error(_("mode change for %s, which is not "
4091                                               "in current HEAD"), name);
4092                } else
4093                        return error(_("sha1 information is lacking or useless "
4094                                       "(%s)."), name);
4095
4096                ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4097                if (!ce)
4098                        return error(_("make_cache_entry failed for path '%s'"),
4099                                     name);
4100                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4101                        discard_cache_entry(ce);
4102                        return error(_("could not add %s to temporary index"),
4103                                     name);
4104                }
4105        }
4106
4107        hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4108        res = write_locked_index(&result, &lock, COMMIT_LOCK);
4109        discard_index(&result);
4110
4111        if (res)
4112                return error(_("could not write temporary index to %s"),
4113                             state->fake_ancestor);
4114
4115        return 0;
4116}
4117
4118static void stat_patch_list(struct apply_state *state, struct patch *patch)
4119{
4120        int files, adds, dels;
4121
4122        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4123                files++;
4124                adds += patch->lines_added;
4125                dels += patch->lines_deleted;
4126                show_stats(state, patch);
4127        }
4128
4129        print_stat_summary(stdout, files, adds, dels);
4130}
4131
4132static void numstat_patch_list(struct apply_state *state,
4133                               struct patch *patch)
4134{
4135        for ( ; patch; patch = patch->next) {
4136                const char *name;
4137                name = patch->new_name ? patch->new_name : patch->old_name;
4138                if (patch->is_binary)
4139                        printf("-\t-\t");
4140                else
4141                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4142                write_name_quoted(name, stdout, state->line_termination);
4143        }
4144}
4145
4146static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4147{
4148        if (mode)
4149                printf(" %s mode %06o %s\n", newdelete, mode, name);
4150        else
4151                printf(" %s %s\n", newdelete, name);
4152}
4153
4154static void show_mode_change(struct patch *p, int show_name)
4155{
4156        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4157                if (show_name)
4158                        printf(" mode change %06o => %06o %s\n",
4159                               p->old_mode, p->new_mode, p->new_name);
4160                else
4161                        printf(" mode change %06o => %06o\n",
4162                               p->old_mode, p->new_mode);
4163        }
4164}
4165
4166static void show_rename_copy(struct patch *p)
4167{
4168        const char *renamecopy = p->is_rename ? "rename" : "copy";
4169        const char *old_name, *new_name;
4170
4171        /* Find common prefix */
4172        old_name = p->old_name;
4173        new_name = p->new_name;
4174        while (1) {
4175                const char *slash_old, *slash_new;
4176                slash_old = strchr(old_name, '/');
4177                slash_new = strchr(new_name, '/');
4178                if (!slash_old ||
4179                    !slash_new ||
4180                    slash_old - old_name != slash_new - new_name ||
4181                    memcmp(old_name, new_name, slash_new - new_name))
4182                        break;
4183                old_name = slash_old + 1;
4184                new_name = slash_new + 1;
4185        }
4186        /* p->old_name thru old_name is the common prefix, and old_name and new_name
4187         * through the end of names are renames
4188         */
4189        if (old_name != p->old_name)
4190                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4191                       (int)(old_name - p->old_name), p->old_name,
4192                       old_name, new_name, p->score);
4193        else
4194                printf(" %s %s => %s (%d%%)\n", renamecopy,
4195                       p->old_name, p->new_name, p->score);
4196        show_mode_change(p, 0);
4197}
4198
4199static void summary_patch_list(struct patch *patch)
4200{
4201        struct patch *p;
4202
4203        for (p = patch; p; p = p->next) {
4204                if (p->is_new)
4205                        show_file_mode_name("create", p->new_mode, p->new_name);
4206                else if (p->is_delete)
4207                        show_file_mode_name("delete", p->old_mode, p->old_name);
4208                else {
4209                        if (p->is_rename || p->is_copy)
4210                                show_rename_copy(p);
4211                        else {
4212                                if (p->score) {
4213                                        printf(" rewrite %s (%d%%)\n",
4214                                               p->new_name, p->score);
4215                                        show_mode_change(p, 0);
4216                                }
4217                                else
4218                                        show_mode_change(p, 1);
4219                        }
4220                }
4221        }
4222}
4223
4224static void patch_stats(struct apply_state *state, struct patch *patch)
4225{
4226        int lines = patch->lines_added + patch->lines_deleted;
4227
4228        if (lines > state->max_change)
4229                state->max_change = lines;
4230        if (patch->old_name) {
4231                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4232                if (!len)
4233                        len = strlen(patch->old_name);
4234                if (len > state->max_len)
4235                        state->max_len = len;
4236        }
4237        if (patch->new_name) {
4238                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4239                if (!len)
4240                        len = strlen(patch->new_name);
4241                if (len > state->max_len)
4242                        state->max_len = len;
4243        }
4244}
4245
4246static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4247{
4248        if (state->update_index && !state->ita_only) {
4249                if (remove_file_from_cache(patch->old_name) < 0)
4250                        return error(_("unable to remove %s from index"), patch->old_name);
4251        }
4252        if (!state->cached) {
4253                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4254                        remove_path(patch->old_name);
4255                }
4256        }
4257        return 0;
4258}
4259
4260static int add_index_file(struct apply_state *state,
4261                          const char *path,
4262                          unsigned mode,
4263                          void *buf,
4264                          unsigned long size)
4265{
4266        struct stat st;
4267        struct cache_entry *ce;
4268        int namelen = strlen(path);
4269
4270        ce = make_empty_cache_entry(&the_index, namelen);
4271        memcpy(ce->name, path, namelen);
4272        ce->ce_mode = create_ce_mode(mode);
4273        ce->ce_flags = create_ce_flags(0);
4274        ce->ce_namelen = namelen;
4275        if (state->ita_only) {
4276                ce->ce_flags |= CE_INTENT_TO_ADD;
4277                set_object_name_for_intent_to_add_entry(ce);
4278        } else if (S_ISGITLINK(mode)) {
4279                const char *s;
4280
4281                if (!skip_prefix(buf, "Subproject commit ", &s) ||
4282                    get_oid_hex(s, &ce->oid)) {
4283                        discard_cache_entry(ce);
4284                        return error(_("corrupt patch for submodule %s"), path);
4285                }
4286        } else {
4287                if (!state->cached) {
4288                        if (lstat(path, &st) < 0) {
4289                                discard_cache_entry(ce);
4290                                return error_errno(_("unable to stat newly "
4291                                                     "created file '%s'"),
4292                                                   path);
4293                        }
4294                        fill_stat_cache_info(ce, &st);
4295                }
4296                if (write_object_file(buf, size, blob_type, &ce->oid) < 0) {
4297                        discard_cache_entry(ce);
4298                        return error(_("unable to create backing store "
4299                                       "for newly created file %s"), path);
4300                }
4301        }
4302        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4303                discard_cache_entry(ce);
4304                return error(_("unable to add cache entry for %s"), path);
4305        }
4306
4307        return 0;
4308}
4309
4310/*
4311 * Returns:
4312 *  -1 if an unrecoverable error happened
4313 *   0 if everything went well
4314 *   1 if a recoverable error happened
4315 */
4316static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4317{
4318        int fd, res;
4319        struct strbuf nbuf = STRBUF_INIT;
4320
4321        if (S_ISGITLINK(mode)) {
4322                struct stat st;
4323                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4324                        return 0;
4325                return !!mkdir(path, 0777);
4326        }
4327
4328        if (has_symlinks && S_ISLNK(mode))
4329                /* Although buf:size is counted string, it also is NUL
4330                 * terminated.
4331                 */
4332                return !!symlink(buf, path);
4333
4334        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4335        if (fd < 0)
4336                return 1;
4337
4338        if (convert_to_working_tree(path, buf, size, &nbuf)) {
4339                size = nbuf.len;
4340                buf  = nbuf.buf;
4341        }
4342
4343        res = write_in_full(fd, buf, size) < 0;
4344        if (res)
4345                error_errno(_("failed to write to '%s'"), path);
4346        strbuf_release(&nbuf);
4347
4348        if (close(fd) < 0 && !res)
4349                return error_errno(_("closing file '%s'"), path);
4350
4351        return res ? -1 : 0;
4352}
4353
4354/*
4355 * We optimistically assume that the directories exist,
4356 * which is true 99% of the time anyway. If they don't,
4357 * we create them and try again.
4358 *
4359 * Returns:
4360 *   -1 on error
4361 *   0 otherwise
4362 */
4363static int create_one_file(struct apply_state *state,
4364                           char *path,
4365                           unsigned mode,
4366                           const char *buf,
4367                           unsigned long size)
4368{
4369        int res;
4370
4371        if (state->cached)
4372                return 0;
4373
4374        res = try_create_file(path, mode, buf, size);
4375        if (res < 0)
4376                return -1;
4377        if (!res)
4378                return 0;
4379
4380        if (errno == ENOENT) {
4381                if (safe_create_leading_directories(path))
4382                        return 0;
4383                res = try_create_file(path, mode, buf, size);
4384                if (res < 0)
4385                        return -1;
4386                if (!res)
4387                        return 0;
4388        }
4389
4390        if (errno == EEXIST || errno == EACCES) {
4391                /* We may be trying to create a file where a directory
4392                 * used to be.
4393                 */
4394                struct stat st;
4395                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4396                        errno = EEXIST;
4397        }
4398
4399        if (errno == EEXIST) {
4400                unsigned int nr = getpid();
4401
4402                for (;;) {
4403                        char newpath[PATH_MAX];
4404                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4405                        res = try_create_file(newpath, mode, buf, size);
4406                        if (res < 0)
4407                                return -1;
4408                        if (!res) {
4409                                if (!rename(newpath, path))
4410                                        return 0;
4411                                unlink_or_warn(newpath);
4412                                break;
4413                        }
4414                        if (errno != EEXIST)
4415                                break;
4416                        ++nr;
4417                }
4418        }
4419        return error_errno(_("unable to write file '%s' mode %o"),
4420                           path, mode);
4421}
4422
4423static int add_conflicted_stages_file(struct apply_state *state,
4424                                       struct patch *patch)
4425{
4426        int stage, namelen;
4427        unsigned mode;
4428        struct cache_entry *ce;
4429
4430        if (!state->update_index)
4431                return 0;
4432        namelen = strlen(patch->new_name);
4433        mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4434
4435        remove_file_from_cache(patch->new_name);
4436        for (stage = 1; stage < 4; stage++) {
4437                if (is_null_oid(&patch->threeway_stage[stage - 1]))
4438                        continue;
4439                ce = make_empty_cache_entry(&the_index, namelen);
4440                memcpy(ce->name, patch->new_name, namelen);
4441                ce->ce_mode = create_ce_mode(mode);
4442                ce->ce_flags = create_ce_flags(stage);
4443                ce->ce_namelen = namelen;
4444                oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4445                if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4446                        discard_cache_entry(ce);
4447                        return error(_("unable to add cache entry for %s"),
4448                                     patch->new_name);
4449                }
4450        }
4451
4452        return 0;
4453}
4454
4455static int create_file(struct apply_state *state, struct patch *patch)
4456{
4457        char *path = patch->new_name;
4458        unsigned mode = patch->new_mode;
4459        unsigned long size = patch->resultsize;
4460        char *buf = patch->result;
4461
4462        if (!mode)
4463                mode = S_IFREG | 0644;
4464        if (create_one_file(state, path, mode, buf, size))
4465                return -1;
4466
4467        if (patch->conflicted_threeway)
4468                return add_conflicted_stages_file(state, patch);
4469        else if (state->update_index)
4470                return add_index_file(state, path, mode, buf, size);
4471        return 0;
4472}
4473
4474/* phase zero is to remove, phase one is to create */
4475static int write_out_one_result(struct apply_state *state,
4476                                struct patch *patch,
4477                                int phase)
4478{
4479        if (patch->is_delete > 0) {
4480                if (phase == 0)
4481                        return remove_file(state, patch, 1);
4482                return 0;
4483        }
4484        if (patch->is_new > 0 || patch->is_copy) {
4485                if (phase == 1)
4486                        return create_file(state, patch);
4487                return 0;
4488        }
4489        /*
4490         * Rename or modification boils down to the same
4491         * thing: remove the old, write the new
4492         */
4493        if (phase == 0)
4494                return remove_file(state, patch, patch->is_rename);
4495        if (phase == 1)
4496                return create_file(state, patch);
4497        return 0;
4498}
4499
4500static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4501{
4502        FILE *rej;
4503        char namebuf[PATH_MAX];
4504        struct fragment *frag;
4505        int cnt = 0;
4506        struct strbuf sb = STRBUF_INIT;
4507
4508        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4509                if (!frag->rejected)
4510                        continue;
4511                cnt++;
4512        }
4513
4514        if (!cnt) {
4515                if (state->apply_verbosity > verbosity_normal)
4516                        say_patch_name(stderr,
4517                                       _("Applied patch %s cleanly."), patch);
4518                return 0;
4519        }
4520
4521        /* This should not happen, because a removal patch that leaves
4522         * contents are marked "rejected" at the patch level.
4523         */
4524        if (!patch->new_name)
4525                die(_("internal error"));
4526
4527        /* Say this even without --verbose */
4528        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4529                            "Applying patch %%s with %d rejects...",
4530                            cnt),
4531                    cnt);
4532        if (state->apply_verbosity > verbosity_silent)
4533                say_patch_name(stderr, sb.buf, patch);
4534        strbuf_release(&sb);
4535
4536        cnt = strlen(patch->new_name);
4537        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4538                cnt = ARRAY_SIZE(namebuf) - 5;
4539                warning(_("truncating .rej filename to %.*s.rej"),
4540                        cnt - 1, patch->new_name);
4541        }
4542        memcpy(namebuf, patch->new_name, cnt);
4543        memcpy(namebuf + cnt, ".rej", 5);
4544
4545        rej = fopen(namebuf, "w");
4546        if (!rej)
4547                return error_errno(_("cannot open %s"), namebuf);
4548
4549        /* Normal git tools never deal with .rej, so do not pretend
4550         * this is a git patch by saying --git or giving extended
4551         * headers.  While at it, maybe please "kompare" that wants
4552         * the trailing TAB and some garbage at the end of line ;-).
4553         */
4554        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4555                patch->new_name, patch->new_name);
4556        for (cnt = 1, frag = patch->fragments;
4557             frag;
4558             cnt++, frag = frag->next) {
4559                if (!frag->rejected) {
4560                        if (state->apply_verbosity > verbosity_silent)
4561                                fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4562                        continue;
4563                }
4564                if (state->apply_verbosity > verbosity_silent)
4565                        fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4566                fprintf(rej, "%.*s", frag->size, frag->patch);
4567                if (frag->patch[frag->size-1] != '\n')
4568                        fputc('\n', rej);
4569        }
4570        fclose(rej);
4571        return -1;
4572}
4573
4574/*
4575 * Returns:
4576 *  -1 if an error happened
4577 *   0 if the patch applied cleanly
4578 *   1 if the patch did not apply cleanly
4579 */
4580static int write_out_results(struct apply_state *state, struct patch *list)
4581{
4582        int phase;
4583        int errs = 0;
4584        struct patch *l;
4585        struct string_list cpath = STRING_LIST_INIT_DUP;
4586
4587        for (phase = 0; phase < 2; phase++) {
4588                l = list;
4589                while (l) {
4590                        if (l->rejected)
4591                                errs = 1;
4592                        else {
4593                                if (write_out_one_result(state, l, phase)) {
4594                                        string_list_clear(&cpath, 0);
4595                                        return -1;
4596                                }
4597                                if (phase == 1) {
4598                                        if (write_out_one_reject(state, l))
4599                                                errs = 1;
4600                                        if (l->conflicted_threeway) {
4601                                                string_list_append(&cpath, l->new_name);
4602                                                errs = 1;
4603                                        }
4604                                }
4605                        }
4606                        l = l->next;
4607                }
4608        }
4609
4610        if (cpath.nr) {
4611                struct string_list_item *item;
4612
4613                string_list_sort(&cpath);
4614                if (state->apply_verbosity > verbosity_silent) {
4615                        for_each_string_list_item(item, &cpath)
4616                                fprintf(stderr, "U %s\n", item->string);
4617                }
4618                string_list_clear(&cpath, 0);
4619
4620                rerere(0);
4621        }
4622
4623        return errs;
4624}
4625
4626/*
4627 * Try to apply a patch.
4628 *
4629 * Returns:
4630 *  -128 if a bad error happened (like patch unreadable)
4631 *  -1 if patch did not apply and user cannot deal with it
4632 *   0 if the patch applied
4633 *   1 if the patch did not apply but user might fix it
4634 */
4635static int apply_patch(struct apply_state *state,
4636                       int fd,
4637                       const char *filename,
4638                       int options)
4639{
4640        size_t offset;
4641        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4642        struct patch *list = NULL, **listp = &list;
4643        int skipped_patch = 0;
4644        int res = 0;
4645
4646        state->patch_input_file = filename;
4647        if (read_patch_file(&buf, fd) < 0)
4648                return -128;
4649        offset = 0;
4650        while (offset < buf.len) {
4651                struct patch *patch;
4652                int nr;
4653
4654                patch = xcalloc(1, sizeof(*patch));
4655                patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4656                patch->recount =  !!(options & APPLY_OPT_RECOUNT);
4657                nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4658                if (nr < 0) {
4659                        free_patch(patch);
4660                        if (nr == -128) {
4661                                res = -128;
4662                                goto end;
4663                        }
4664                        break;
4665                }
4666                if (state->apply_in_reverse)
4667                        reverse_patches(patch);
4668                if (use_patch(state, patch)) {
4669                        patch_stats(state, patch);
4670                        *listp = patch;
4671                        listp = &patch->next;
4672                }
4673                else {
4674                        if (state->apply_verbosity > verbosity_normal)
4675                                say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4676                        free_patch(patch);
4677                        skipped_patch++;
4678                }
4679                offset += nr;
4680        }
4681
4682        if (!list && !skipped_patch) {
4683                error(_("unrecognized input"));
4684                res = -128;
4685                goto end;
4686        }
4687
4688        if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4689                state->apply = 0;
4690
4691        state->update_index = (state->check_index || state->ita_only) && state->apply;
4692        if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4693                if (state->index_file)
4694                        hold_lock_file_for_update(&state->lock_file,
4695                                                  state->index_file,
4696                                                  LOCK_DIE_ON_ERROR);
4697                else
4698                        hold_locked_index(&state->lock_file, LOCK_DIE_ON_ERROR);
4699        }
4700
4701        if (state->check_index && read_apply_cache(state) < 0) {
4702                error(_("unable to read index file"));
4703                res = -128;
4704                goto end;
4705        }
4706
4707        if (state->check || state->apply) {
4708                int r = check_patch_list(state, list);
4709                if (r == -128) {
4710                        res = -128;
4711                        goto end;
4712                }
4713                if (r < 0 && !state->apply_with_reject) {
4714                        res = -1;
4715                        goto end;
4716                }
4717        }
4718
4719        if (state->apply) {
4720                int write_res = write_out_results(state, list);
4721                if (write_res < 0) {
4722                        res = -128;
4723                        goto end;
4724                }
4725                if (write_res > 0) {
4726                        /* with --3way, we still need to write the index out */
4727                        res = state->apply_with_reject ? -1 : 1;
4728                        goto end;
4729                }
4730        }
4731
4732        if (state->fake_ancestor &&
4733            build_fake_ancestor(state, list)) {
4734                res = -128;
4735                goto end;
4736        }
4737
4738        if (state->diffstat && state->apply_verbosity > verbosity_silent)
4739                stat_patch_list(state, list);
4740
4741        if (state->numstat && state->apply_verbosity > verbosity_silent)
4742                numstat_patch_list(state, list);
4743
4744        if (state->summary && state->apply_verbosity > verbosity_silent)
4745                summary_patch_list(list);
4746
4747end:
4748        free_patch_list(list);
4749        strbuf_release(&buf);
4750        string_list_clear(&state->fn_table, 0);
4751        return res;
4752}
4753
4754static int apply_option_parse_exclude(const struct option *opt,
4755                                      const char *arg, int unset)
4756{
4757        struct apply_state *state = opt->value;
4758        add_name_limit(state, arg, 1);
4759        return 0;
4760}
4761
4762static int apply_option_parse_include(const struct option *opt,
4763                                      const char *arg, int unset)
4764{
4765        struct apply_state *state = opt->value;
4766        add_name_limit(state, arg, 0);
4767        state->has_include = 1;
4768        return 0;
4769}
4770
4771static int apply_option_parse_p(const struct option *opt,
4772                                const char *arg,
4773                                int unset)
4774{
4775        struct apply_state *state = opt->value;
4776        state->p_value = atoi(arg);
4777        state->p_value_known = 1;
4778        return 0;
4779}
4780
4781static int apply_option_parse_space_change(const struct option *opt,
4782                                           const char *arg, int unset)
4783{
4784        struct apply_state *state = opt->value;
4785        if (unset)
4786                state->ws_ignore_action = ignore_ws_none;
4787        else
4788                state->ws_ignore_action = ignore_ws_change;
4789        return 0;
4790}
4791
4792static int apply_option_parse_whitespace(const struct option *opt,
4793                                         const char *arg, int unset)
4794{
4795        struct apply_state *state = opt->value;
4796        state->whitespace_option = arg;
4797        if (parse_whitespace_option(state, arg))
4798                exit(1);
4799        return 0;
4800}
4801
4802static int apply_option_parse_directory(const struct option *opt,
4803                                        const char *arg, int unset)
4804{
4805        struct apply_state *state = opt->value;
4806        strbuf_reset(&state->root);
4807        strbuf_addstr(&state->root, arg);
4808        strbuf_complete(&state->root, '/');
4809        return 0;
4810}
4811
4812int apply_all_patches(struct apply_state *state,
4813                      int argc,
4814                      const char **argv,
4815                      int options)
4816{
4817        int i;
4818        int res;
4819        int errs = 0;
4820        int read_stdin = 1;
4821
4822        for (i = 0; i < argc; i++) {
4823                const char *arg = argv[i];
4824                char *to_free = NULL;
4825                int fd;
4826
4827                if (!strcmp(arg, "-")) {
4828                        res = apply_patch(state, 0, "<stdin>", options);
4829                        if (res < 0)
4830                                goto end;
4831                        errs |= res;
4832                        read_stdin = 0;
4833                        continue;
4834                } else
4835                        arg = to_free = prefix_filename(state->prefix, arg);
4836
4837                fd = open(arg, O_RDONLY);
4838                if (fd < 0) {
4839                        error(_("can't open patch '%s': %s"), arg, strerror(errno));
4840                        res = -128;
4841                        free(to_free);
4842                        goto end;
4843                }
4844                read_stdin = 0;
4845                set_default_whitespace_mode(state);
4846                res = apply_patch(state, fd, arg, options);
4847                close(fd);
4848                free(to_free);
4849                if (res < 0)
4850                        goto end;
4851                errs |= res;
4852        }
4853        set_default_whitespace_mode(state);
4854        if (read_stdin) {
4855                res = apply_patch(state, 0, "<stdin>", options);
4856                if (res < 0)
4857                        goto end;
4858                errs |= res;
4859        }
4860
4861        if (state->whitespace_error) {
4862                if (state->squelch_whitespace_errors &&
4863                    state->squelch_whitespace_errors < state->whitespace_error) {
4864                        int squelched =
4865                                state->whitespace_error - state->squelch_whitespace_errors;
4866                        warning(Q_("squelched %d whitespace error",
4867                                   "squelched %d whitespace errors",
4868                                   squelched),
4869                                squelched);
4870                }
4871                if (state->ws_error_action == die_on_ws_error) {
4872                        error(Q_("%d line adds whitespace errors.",
4873                                 "%d lines add whitespace errors.",
4874                                 state->whitespace_error),
4875                              state->whitespace_error);
4876                        res = -128;
4877                        goto end;
4878                }
4879                if (state->applied_after_fixing_ws && state->apply)
4880                        warning(Q_("%d line applied after"
4881                                   " fixing whitespace errors.",
4882                                   "%d lines applied after"
4883                                   " fixing whitespace errors.",
4884                                   state->applied_after_fixing_ws),
4885                                state->applied_after_fixing_ws);
4886                else if (state->whitespace_error)
4887                        warning(Q_("%d line adds whitespace errors.",
4888                                   "%d lines add whitespace errors.",
4889                                   state->whitespace_error),
4890                                state->whitespace_error);
4891        }
4892
4893        if (state->update_index) {
4894                res = write_locked_index(&the_index, &state->lock_file, COMMIT_LOCK);
4895                if (res) {
4896                        error(_("Unable to write new index file"));
4897                        res = -128;
4898                        goto end;
4899                }
4900        }
4901
4902        res = !!errs;
4903
4904end:
4905        rollback_lock_file(&state->lock_file);
4906
4907        if (state->apply_verbosity <= verbosity_silent) {
4908                set_error_routine(state->saved_error_routine);
4909                set_warn_routine(state->saved_warn_routine);
4910        }
4911
4912        if (res > -1)
4913                return res;
4914        return (res == -1 ? 1 : 128);
4915}
4916
4917int apply_parse_options(int argc, const char **argv,
4918                        struct apply_state *state,
4919                        int *force_apply, int *options,
4920                        const char * const *apply_usage)
4921{
4922        struct option builtin_apply_options[] = {
4923                { OPTION_CALLBACK, 0, "exclude", state, N_("path"),
4924                        N_("don't apply changes matching the given path"),
4925                        0, apply_option_parse_exclude },
4926                { OPTION_CALLBACK, 0, "include", state, N_("path"),
4927                        N_("apply changes matching the given path"),
4928                        0, apply_option_parse_include },
4929                { OPTION_CALLBACK, 'p', NULL, state, N_("num"),
4930                        N_("remove <num> leading slashes from traditional diff paths"),
4931                        0, apply_option_parse_p },
4932                OPT_BOOL(0, "no-add", &state->no_add,
4933                        N_("ignore additions made by the patch")),
4934                OPT_BOOL(0, "stat", &state->diffstat,
4935                        N_("instead of applying the patch, output diffstat for the input")),
4936                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4937                OPT_NOOP_NOARG(0, "binary"),
4938                OPT_BOOL(0, "numstat", &state->numstat,
4939                        N_("show number of added and deleted lines in decimal notation")),
4940                OPT_BOOL(0, "summary", &state->summary,
4941                        N_("instead of applying the patch, output a summary for the input")),
4942                OPT_BOOL(0, "check", &state->check,
4943                        N_("instead of applying the patch, see if the patch is applicable")),
4944                OPT_BOOL(0, "index", &state->check_index,
4945                        N_("make sure the patch is applicable to the current index")),
4946                OPT_BOOL('N', "intent-to-add", &state->ita_only,
4947                        N_("mark new files with `git add --intent-to-add`")),
4948                OPT_BOOL(0, "cached", &state->cached,
4949                        N_("apply a patch without touching the working tree")),
4950                OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
4951                           N_("accept a patch that touches outside the working area"),
4952                           PARSE_OPT_NOCOMPLETE),
4953                OPT_BOOL(0, "apply", force_apply,
4954                        N_("also apply the patch (use with --stat/--summary/--check)")),
4955                OPT_BOOL('3', "3way", &state->threeway,
4956                         N_( "attempt three-way merge if a patch does not apply")),
4957                OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
4958                        N_("build a temporary index based on embedded index information")),
4959                /* Think twice before adding "--nul" synonym to this */
4960                OPT_SET_INT('z', NULL, &state->line_termination,
4961                        N_("paths are separated with NUL character"), '\0'),
4962                OPT_INTEGER('C', NULL, &state->p_context,
4963                                N_("ensure at least <n> lines of context match")),
4964                { OPTION_CALLBACK, 0, "whitespace", state, N_("action"),
4965                        N_("detect new or modified lines that have whitespace errors"),
4966                        0, apply_option_parse_whitespace },
4967                { OPTION_CALLBACK, 0, "ignore-space-change", state, NULL,
4968                        N_("ignore changes in whitespace when finding context"),
4969                        PARSE_OPT_NOARG, apply_option_parse_space_change },
4970                { OPTION_CALLBACK, 0, "ignore-whitespace", state, NULL,
4971                        N_("ignore changes in whitespace when finding context"),
4972                        PARSE_OPT_NOARG, apply_option_parse_space_change },
4973                OPT_BOOL('R', "reverse", &state->apply_in_reverse,
4974                        N_("apply the patch in reverse")),
4975                OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
4976                        N_("don't expect at least one line of context")),
4977                OPT_BOOL(0, "reject", &state->apply_with_reject,
4978                        N_("leave the rejected hunks in corresponding *.rej files")),
4979                OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
4980                        N_("allow overlapping hunks")),
4981                OPT__VERBOSE(&state->apply_verbosity, N_("be verbose")),
4982                OPT_BIT(0, "inaccurate-eof", options,
4983                        N_("tolerate incorrectly detected missing new-line at the end of file"),
4984                        APPLY_OPT_INACCURATE_EOF),
4985                OPT_BIT(0, "recount", options,
4986                        N_("do not trust the line counts in the hunk headers"),
4987                        APPLY_OPT_RECOUNT),
4988                { OPTION_CALLBACK, 0, "directory", state, N_("root"),
4989                        N_("prepend <root> to all filenames"),
4990                        0, apply_option_parse_directory },
4991                OPT_END()
4992        };
4993
4994        return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);
4995}