apply.con commit grep: fix segfault under -P + PCRE2 <=10.30 + (*NO_JIT) (a25b908)
   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 "blob.h"
  13#include "delta.h"
  14#include "diff.h"
  15#include "dir.h"
  16#include "xdiff-interface.h"
  17#include "ll-merge.h"
  18#include "lockfile.h"
  19#include "parse-options.h"
  20#include "quote.h"
  21#include "rerere.h"
  22#include "apply.h"
  23
  24static void git_apply_config(void)
  25{
  26        git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
  27        git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
  28        git_config(git_default_config, NULL);
  29}
  30
  31static int parse_whitespace_option(struct apply_state *state, const char *option)
  32{
  33        if (!option) {
  34                state->ws_error_action = warn_on_ws_error;
  35                return 0;
  36        }
  37        if (!strcmp(option, "warn")) {
  38                state->ws_error_action = warn_on_ws_error;
  39                return 0;
  40        }
  41        if (!strcmp(option, "nowarn")) {
  42                state->ws_error_action = nowarn_ws_error;
  43                return 0;
  44        }
  45        if (!strcmp(option, "error")) {
  46                state->ws_error_action = die_on_ws_error;
  47                return 0;
  48        }
  49        if (!strcmp(option, "error-all")) {
  50                state->ws_error_action = die_on_ws_error;
  51                state->squelch_whitespace_errors = 0;
  52                return 0;
  53        }
  54        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
  55                state->ws_error_action = correct_ws_error;
  56                return 0;
  57        }
  58        return error(_("unrecognized whitespace option '%s'"), option);
  59}
  60
  61static int parse_ignorewhitespace_option(struct apply_state *state,
  62                                                 const char *option)
  63{
  64        if (!option || !strcmp(option, "no") ||
  65            !strcmp(option, "false") || !strcmp(option, "never") ||
  66            !strcmp(option, "none")) {
  67                state->ws_ignore_action = ignore_ws_none;
  68                return 0;
  69        }
  70        if (!strcmp(option, "change")) {
  71                state->ws_ignore_action = ignore_ws_change;
  72                return 0;
  73        }
  74        return error(_("unrecognized whitespace ignore option '%s'"), option);
  75}
  76
  77int init_apply_state(struct apply_state *state,
  78                     const char *prefix)
  79{
  80        memset(state, 0, sizeof(*state));
  81        state->prefix = prefix;
  82        state->apply = 1;
  83        state->line_termination = '\n';
  84        state->p_value = 1;
  85        state->p_context = UINT_MAX;
  86        state->squelch_whitespace_errors = 5;
  87        state->ws_error_action = warn_on_ws_error;
  88        state->ws_ignore_action = ignore_ws_none;
  89        state->linenr = 1;
  90        string_list_init(&state->fn_table, 0);
  91        string_list_init(&state->limit_by_name, 0);
  92        string_list_init(&state->symlink_changes, 0);
  93        strbuf_init(&state->root, 0);
  94
  95        git_apply_config();
  96        if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
  97                return -1;
  98        if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
  99                return -1;
 100        return 0;
 101}
 102
 103void clear_apply_state(struct apply_state *state)
 104{
 105        string_list_clear(&state->limit_by_name, 0);
 106        string_list_clear(&state->symlink_changes, 0);
 107        strbuf_release(&state->root);
 108
 109        /* &state->fn_table is cleared at the end of apply_patch() */
 110}
 111
 112static void mute_routine(const char *msg, va_list params)
 113{
 114        /* do nothing */
 115}
 116
 117int check_apply_state(struct apply_state *state, int force_apply)
 118{
 119        int is_not_gitdir = !startup_info->have_repository;
 120
 121        if (state->apply_with_reject && state->threeway)
 122                return error(_("--reject and --3way cannot be used together."));
 123        if (state->cached && state->threeway)
 124                return error(_("--cached and --3way cannot be used together."));
 125        if (state->threeway) {
 126                if (is_not_gitdir)
 127                        return error(_("--3way outside a repository"));
 128                state->check_index = 1;
 129        }
 130        if (state->apply_with_reject) {
 131                state->apply = 1;
 132                if (state->apply_verbosity == verbosity_normal)
 133                        state->apply_verbosity = verbosity_verbose;
 134        }
 135        if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
 136                state->apply = 0;
 137        if (state->check_index && is_not_gitdir)
 138                return error(_("--index outside a repository"));
 139        if (state->cached) {
 140                if (is_not_gitdir)
 141                        return error(_("--cached outside a repository"));
 142                state->check_index = 1;
 143        }
 144        if (state->check_index)
 145                state->unsafe_paths = 0;
 146
 147        if (state->apply_verbosity <= verbosity_silent) {
 148                state->saved_error_routine = get_error_routine();
 149                state->saved_warn_routine = get_warn_routine();
 150                set_error_routine(mute_routine);
 151                set_warn_routine(mute_routine);
 152        }
 153
 154        return 0;
 155}
 156
 157static void set_default_whitespace_mode(struct apply_state *state)
 158{
 159        if (!state->whitespace_option && !apply_default_whitespace)
 160                state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
 161}
 162
 163/*
 164 * This represents one "hunk" from a patch, starting with
 165 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 166 * patch text is pointed at by patch, and its byte length
 167 * is stored in size.  leading and trailing are the number
 168 * of context lines.
 169 */
 170struct fragment {
 171        unsigned long leading, trailing;
 172        unsigned long oldpos, oldlines;
 173        unsigned long newpos, newlines;
 174        /*
 175         * 'patch' is usually borrowed from buf in apply_patch(),
 176         * but some codepaths store an allocated buffer.
 177         */
 178        const char *patch;
 179        unsigned free_patch:1,
 180                rejected:1;
 181        int size;
 182        int linenr;
 183        struct fragment *next;
 184};
 185
 186/*
 187 * When dealing with a binary patch, we reuse "leading" field
 188 * to store the type of the binary hunk, either deflated "delta"
 189 * or deflated "literal".
 190 */
 191#define binary_patch_method leading
 192#define BINARY_DELTA_DEFLATED   1
 193#define BINARY_LITERAL_DEFLATED 2
 194
 195/*
 196 * This represents a "patch" to a file, both metainfo changes
 197 * such as creation/deletion, filemode and content changes represented
 198 * as a series of fragments.
 199 */
 200struct patch {
 201        char *new_name, *old_name, *def_name;
 202        unsigned int old_mode, new_mode;
 203        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 204        int rejected;
 205        unsigned ws_rule;
 206        int lines_added, lines_deleted;
 207        int score;
 208        int extension_linenr; /* first line specifying delete/new/rename/copy */
 209        unsigned int is_toplevel_relative:1;
 210        unsigned int inaccurate_eof:1;
 211        unsigned int is_binary:1;
 212        unsigned int is_copy:1;
 213        unsigned int is_rename:1;
 214        unsigned int recount:1;
 215        unsigned int conflicted_threeway:1;
 216        unsigned int direct_to_threeway:1;
 217        unsigned int crlf_in_old:1;
 218        struct fragment *fragments;
 219        char *result;
 220        size_t resultsize;
 221        char old_sha1_prefix[41];
 222        char new_sha1_prefix[41];
 223        struct patch *next;
 224
 225        /* three-way fallback result */
 226        struct object_id threeway_stage[3];
 227};
 228
 229static void free_fragment_list(struct fragment *list)
 230{
 231        while (list) {
 232                struct fragment *next = list->next;
 233                if (list->free_patch)
 234                        free((char *)list->patch);
 235                free(list);
 236                list = next;
 237        }
 238}
 239
 240static void free_patch(struct patch *patch)
 241{
 242        free_fragment_list(patch->fragments);
 243        free(patch->def_name);
 244        free(patch->old_name);
 245        free(patch->new_name);
 246        free(patch->result);
 247        free(patch);
 248}
 249
 250static void free_patch_list(struct patch *list)
 251{
 252        while (list) {
 253                struct patch *next = list->next;
 254                free_patch(list);
 255                list = next;
 256        }
 257}
 258
 259/*
 260 * A line in a file, len-bytes long (includes the terminating LF,
 261 * except for an incomplete line at the end if the file ends with
 262 * one), and its contents hashes to 'hash'.
 263 */
 264struct line {
 265        size_t len;
 266        unsigned hash : 24;
 267        unsigned flag : 8;
 268#define LINE_COMMON     1
 269#define LINE_PATCHED    2
 270};
 271
 272/*
 273 * This represents a "file", which is an array of "lines".
 274 */
 275struct image {
 276        char *buf;
 277        size_t len;
 278        size_t nr;
 279        size_t alloc;
 280        struct line *line_allocated;
 281        struct line *line;
 282};
 283
 284static uint32_t hash_line(const char *cp, size_t len)
 285{
 286        size_t i;
 287        uint32_t h;
 288        for (i = 0, h = 0; i < len; i++) {
 289                if (!isspace(cp[i])) {
 290                        h = h * 3 + (cp[i] & 0xff);
 291                }
 292        }
 293        return h;
 294}
 295
 296/*
 297 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 298 * whitespace difference. Returns 1 if they match, 0 otherwise
 299 */
 300static int fuzzy_matchlines(const char *s1, size_t n1,
 301                            const char *s2, size_t n2)
 302{
 303        const char *end1 = s1 + n1;
 304        const char *end2 = s2 + n2;
 305
 306        /* ignore line endings */
 307        while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
 308                end1--;
 309        while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
 310                end2--;
 311
 312        while (s1 < end1 && s2 < end2) {
 313                if (isspace(*s1)) {
 314                        /*
 315                         * Skip whitespace. We check on both buffers
 316                         * because we don't want "a b" to match "ab".
 317                         */
 318                        if (!isspace(*s2))
 319                                return 0;
 320                        while (s1 < end1 && isspace(*s1))
 321                                s1++;
 322                        while (s2 < end2 && isspace(*s2))
 323                                s2++;
 324                } else if (*s1++ != *s2++)
 325                        return 0;
 326        }
 327
 328        /* If we reached the end on one side only, lines don't match. */
 329        return s1 == end1 && s2 == end2;
 330}
 331
 332static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 333{
 334        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 335        img->line_allocated[img->nr].len = len;
 336        img->line_allocated[img->nr].hash = hash_line(bol, len);
 337        img->line_allocated[img->nr].flag = flag;
 338        img->nr++;
 339}
 340
 341/*
 342 * "buf" has the file contents to be patched (read from various sources).
 343 * attach it to "image" and add line-based index to it.
 344 * "image" now owns the "buf".
 345 */
 346static void prepare_image(struct image *image, char *buf, size_t len,
 347                          int prepare_linetable)
 348{
 349        const char *cp, *ep;
 350
 351        memset(image, 0, sizeof(*image));
 352        image->buf = buf;
 353        image->len = len;
 354
 355        if (!prepare_linetable)
 356                return;
 357
 358        ep = image->buf + image->len;
 359        cp = image->buf;
 360        while (cp < ep) {
 361                const char *next;
 362                for (next = cp; next < ep && *next != '\n'; next++)
 363                        ;
 364                if (next < ep)
 365                        next++;
 366                add_line_info(image, cp, next - cp, 0);
 367                cp = next;
 368        }
 369        image->line = image->line_allocated;
 370}
 371
 372static void clear_image(struct image *image)
 373{
 374        free(image->buf);
 375        free(image->line_allocated);
 376        memset(image, 0, sizeof(*image));
 377}
 378
 379/* fmt must contain _one_ %s and no other substitution */
 380static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 381{
 382        struct strbuf sb = STRBUF_INIT;
 383
 384        if (patch->old_name && patch->new_name &&
 385            strcmp(patch->old_name, patch->new_name)) {
 386                quote_c_style(patch->old_name, &sb, NULL, 0);
 387                strbuf_addstr(&sb, " => ");
 388                quote_c_style(patch->new_name, &sb, NULL, 0);
 389        } else {
 390                const char *n = patch->new_name;
 391                if (!n)
 392                        n = patch->old_name;
 393                quote_c_style(n, &sb, NULL, 0);
 394        }
 395        fprintf(output, fmt, sb.buf);
 396        fputc('\n', output);
 397        strbuf_release(&sb);
 398}
 399
 400#define SLOP (16)
 401
 402static int read_patch_file(struct strbuf *sb, int fd)
 403{
 404        if (strbuf_read(sb, fd, 0) < 0)
 405                return error_errno("git apply: failed to read");
 406
 407        /*
 408         * Make sure that we have some slop in the buffer
 409         * so that we can do speculative "memcmp" etc, and
 410         * see to it that it is NUL-filled.
 411         */
 412        strbuf_grow(sb, SLOP);
 413        memset(sb->buf + sb->len, 0, SLOP);
 414        return 0;
 415}
 416
 417static unsigned long linelen(const char *buffer, unsigned long size)
 418{
 419        unsigned long len = 0;
 420        while (size--) {
 421                len++;
 422                if (*buffer++ == '\n')
 423                        break;
 424        }
 425        return len;
 426}
 427
 428static int is_dev_null(const char *str)
 429{
 430        return skip_prefix(str, "/dev/null", &str) && isspace(*str);
 431}
 432
 433#define TERM_SPACE      1
 434#define TERM_TAB        2
 435
 436static int name_terminate(int c, int terminate)
 437{
 438        if (c == ' ' && !(terminate & TERM_SPACE))
 439                return 0;
 440        if (c == '\t' && !(terminate & TERM_TAB))
 441                return 0;
 442
 443        return 1;
 444}
 445
 446/* remove double slashes to make --index work with such filenames */
 447static char *squash_slash(char *name)
 448{
 449        int i = 0, j = 0;
 450
 451        if (!name)
 452                return NULL;
 453
 454        while (name[i]) {
 455                if ((name[j++] = name[i++]) == '/')
 456                        while (name[i] == '/')
 457                                i++;
 458        }
 459        name[j] = '\0';
 460        return name;
 461}
 462
 463static char *find_name_gnu(struct apply_state *state,
 464                           const char *line,
 465                           const char *def,
 466                           int p_value)
 467{
 468        struct strbuf name = STRBUF_INIT;
 469        char *cp;
 470
 471        /*
 472         * Proposed "new-style" GNU patch/diff format; see
 473         * http://marc.info/?l=git&m=112927316408690&w=2
 474         */
 475        if (unquote_c_style(&name, line, NULL)) {
 476                strbuf_release(&name);
 477                return NULL;
 478        }
 479
 480        for (cp = name.buf; p_value; p_value--) {
 481                cp = strchr(cp, '/');
 482                if (!cp) {
 483                        strbuf_release(&name);
 484                        return NULL;
 485                }
 486                cp++;
 487        }
 488
 489        strbuf_remove(&name, 0, cp - name.buf);
 490        if (state->root.len)
 491                strbuf_insert(&name, 0, state->root.buf, state->root.len);
 492        return squash_slash(strbuf_detach(&name, NULL));
 493}
 494
 495static size_t sane_tz_len(const char *line, size_t len)
 496{
 497        const char *tz, *p;
 498
 499        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 500                return 0;
 501        tz = line + len - strlen(" +0500");
 502
 503        if (tz[1] != '+' && tz[1] != '-')
 504                return 0;
 505
 506        for (p = tz + 2; p != line + len; p++)
 507                if (!isdigit(*p))
 508                        return 0;
 509
 510        return line + len - tz;
 511}
 512
 513static size_t tz_with_colon_len(const char *line, size_t len)
 514{
 515        const char *tz, *p;
 516
 517        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 518                return 0;
 519        tz = line + len - strlen(" +08:00");
 520
 521        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 522                return 0;
 523        p = tz + 2;
 524        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 525            !isdigit(*p++) || !isdigit(*p++))
 526                return 0;
 527
 528        return line + len - tz;
 529}
 530
 531static size_t date_len(const char *line, size_t len)
 532{
 533        const char *date, *p;
 534
 535        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 536                return 0;
 537        p = date = line + len - strlen("72-02-05");
 538
 539        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 540            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 541            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 542                return 0;
 543
 544        if (date - line >= strlen("19") &&
 545            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 546                date -= strlen("19");
 547
 548        return line + len - date;
 549}
 550
 551static size_t short_time_len(const char *line, size_t len)
 552{
 553        const char *time, *p;
 554
 555        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 556                return 0;
 557        p = time = line + len - strlen(" 07:01:32");
 558
 559        /* Permit 1-digit hours? */
 560        if (*p++ != ' ' ||
 561            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 562            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 563            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 564                return 0;
 565
 566        return line + len - time;
 567}
 568
 569static size_t fractional_time_len(const char *line, size_t len)
 570{
 571        const char *p;
 572        size_t n;
 573
 574        /* Expected format: 19:41:17.620000023 */
 575        if (!len || !isdigit(line[len - 1]))
 576                return 0;
 577        p = line + len - 1;
 578
 579        /* Fractional seconds. */
 580        while (p > line && isdigit(*p))
 581                p--;
 582        if (*p != '.')
 583                return 0;
 584
 585        /* Hours, minutes, and whole seconds. */
 586        n = short_time_len(line, p - line);
 587        if (!n)
 588                return 0;
 589
 590        return line + len - p + n;
 591}
 592
 593static size_t trailing_spaces_len(const char *line, size_t len)
 594{
 595        const char *p;
 596
 597        /* Expected format: ' ' x (1 or more)  */
 598        if (!len || line[len - 1] != ' ')
 599                return 0;
 600
 601        p = line + len;
 602        while (p != line) {
 603                p--;
 604                if (*p != ' ')
 605                        return line + len - (p + 1);
 606        }
 607
 608        /* All spaces! */
 609        return len;
 610}
 611
 612static size_t diff_timestamp_len(const char *line, size_t len)
 613{
 614        const char *end = line + len;
 615        size_t n;
 616
 617        /*
 618         * Posix: 2010-07-05 19:41:17
 619         * GNU: 2010-07-05 19:41:17.620000023 -0500
 620         */
 621
 622        if (!isdigit(end[-1]))
 623                return 0;
 624
 625        n = sane_tz_len(line, end - line);
 626        if (!n)
 627                n = tz_with_colon_len(line, end - line);
 628        end -= n;
 629
 630        n = short_time_len(line, end - line);
 631        if (!n)
 632                n = fractional_time_len(line, end - line);
 633        end -= n;
 634
 635        n = date_len(line, end - line);
 636        if (!n) /* No date.  Too bad. */
 637                return 0;
 638        end -= n;
 639
 640        if (end == line)        /* No space before date. */
 641                return 0;
 642        if (end[-1] == '\t') {  /* Success! */
 643                end--;
 644                return line + len - end;
 645        }
 646        if (end[-1] != ' ')     /* No space before date. */
 647                return 0;
 648
 649        /* Whitespace damage. */
 650        end -= trailing_spaces_len(line, end - line);
 651        return line + len - end;
 652}
 653
 654static char *find_name_common(struct apply_state *state,
 655                              const char *line,
 656                              const char *def,
 657                              int p_value,
 658                              const char *end,
 659                              int terminate)
 660{
 661        int len;
 662        const char *start = NULL;
 663
 664        if (p_value == 0)
 665                start = line;
 666        while (line != end) {
 667                char c = *line;
 668
 669                if (!end && isspace(c)) {
 670                        if (c == '\n')
 671                                break;
 672                        if (name_terminate(c, terminate))
 673                                break;
 674                }
 675                line++;
 676                if (c == '/' && !--p_value)
 677                        start = line;
 678        }
 679        if (!start)
 680                return squash_slash(xstrdup_or_null(def));
 681        len = line - start;
 682        if (!len)
 683                return squash_slash(xstrdup_or_null(def));
 684
 685        /*
 686         * Generally we prefer the shorter name, especially
 687         * if the other one is just a variation of that with
 688         * something else tacked on to the end (ie "file.orig"
 689         * or "file~").
 690         */
 691        if (def) {
 692                int deflen = strlen(def);
 693                if (deflen < len && !strncmp(start, def, deflen))
 694                        return squash_slash(xstrdup(def));
 695        }
 696
 697        if (state->root.len) {
 698                char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
 699                return squash_slash(ret);
 700        }
 701
 702        return squash_slash(xmemdupz(start, len));
 703}
 704
 705static char *find_name(struct apply_state *state,
 706                       const char *line,
 707                       char *def,
 708                       int p_value,
 709                       int terminate)
 710{
 711        if (*line == '"') {
 712                char *name = find_name_gnu(state, line, def, p_value);
 713                if (name)
 714                        return name;
 715        }
 716
 717        return find_name_common(state, line, def, p_value, NULL, terminate);
 718}
 719
 720static char *find_name_traditional(struct apply_state *state,
 721                                   const char *line,
 722                                   char *def,
 723                                   int p_value)
 724{
 725        size_t len;
 726        size_t date_len;
 727
 728        if (*line == '"') {
 729                char *name = find_name_gnu(state, line, def, p_value);
 730                if (name)
 731                        return name;
 732        }
 733
 734        len = strchrnul(line, '\n') - line;
 735        date_len = diff_timestamp_len(line, len);
 736        if (!date_len)
 737                return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
 738        len -= date_len;
 739
 740        return find_name_common(state, line, def, p_value, line + len, 0);
 741}
 742
 743/*
 744 * Given the string after "--- " or "+++ ", guess the appropriate
 745 * p_value for the given patch.
 746 */
 747static int guess_p_value(struct apply_state *state, const char *nameline)
 748{
 749        char *name, *cp;
 750        int val = -1;
 751
 752        if (is_dev_null(nameline))
 753                return -1;
 754        name = find_name_traditional(state, nameline, NULL, 0);
 755        if (!name)
 756                return -1;
 757        cp = strchr(name, '/');
 758        if (!cp)
 759                val = 0;
 760        else if (state->prefix) {
 761                /*
 762                 * Does it begin with "a/$our-prefix" and such?  Then this is
 763                 * very likely to apply to our directory.
 764                 */
 765                if (starts_with(name, state->prefix))
 766                        val = count_slashes(state->prefix);
 767                else {
 768                        cp++;
 769                        if (starts_with(cp, state->prefix))
 770                                val = count_slashes(state->prefix) + 1;
 771                }
 772        }
 773        free(name);
 774        return val;
 775}
 776
 777/*
 778 * Does the ---/+++ line have the POSIX timestamp after the last HT?
 779 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 780 * this such a timestamp?
 781 */
 782static int has_epoch_timestamp(const char *nameline)
 783{
 784        /*
 785         * We are only interested in epoch timestamp; any non-zero
 786         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 787         * For the same reason, the date must be either 1969-12-31 or
 788         * 1970-01-01, and the seconds part must be "00".
 789         */
 790        const char stamp_regexp[] =
 791                "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
 792                " "
 793                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 794        const char *timestamp = NULL, *cp, *colon;
 795        static regex_t *stamp;
 796        regmatch_t m[10];
 797        int zoneoffset, epoch_hour, hour, minute;
 798        int status;
 799
 800        for (cp = nameline; *cp != '\n'; cp++) {
 801                if (*cp == '\t')
 802                        timestamp = cp + 1;
 803        }
 804        if (!timestamp)
 805                return 0;
 806
 807        /*
 808         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 809         * (west of GMT) or 1970-01-01 (east of GMT)
 810         */
 811        if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
 812                epoch_hour = 24;
 813        else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
 814                epoch_hour = 0;
 815        else
 816                return 0;
 817
 818        if (!stamp) {
 819                stamp = xmalloc(sizeof(*stamp));
 820                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 821                        warning(_("Cannot prepare timestamp regexp %s"),
 822                                stamp_regexp);
 823                        return 0;
 824                }
 825        }
 826
 827        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 828        if (status) {
 829                if (status != REG_NOMATCH)
 830                        warning(_("regexec returned %d for input: %s"),
 831                                status, timestamp);
 832                return 0;
 833        }
 834
 835        hour = strtol(timestamp, NULL, 10);
 836        minute = strtol(timestamp + m[1].rm_so, NULL, 10);
 837
 838        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 839        if (*colon == ':')
 840                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 841        else
 842                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 843        if (timestamp[m[3].rm_so] == '-')
 844                zoneoffset = -zoneoffset;
 845
 846        return hour * 60 + minute - zoneoffset == epoch_hour * 60;
 847}
 848
 849/*
 850 * Get the name etc info from the ---/+++ lines of a traditional patch header
 851 *
 852 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 853 * files, we can happily check the index for a match, but for creating a
 854 * new file we should try to match whatever "patch" does. I have no idea.
 855 */
 856static int parse_traditional_patch(struct apply_state *state,
 857                                   const char *first,
 858                                   const char *second,
 859                                   struct patch *patch)
 860{
 861        char *name;
 862
 863        first += 4;     /* skip "--- " */
 864        second += 4;    /* skip "+++ " */
 865        if (!state->p_value_known) {
 866                int p, q;
 867                p = guess_p_value(state, first);
 868                q = guess_p_value(state, second);
 869                if (p < 0) p = q;
 870                if (0 <= p && p == q) {
 871                        state->p_value = p;
 872                        state->p_value_known = 1;
 873                }
 874        }
 875        if (is_dev_null(first)) {
 876                patch->is_new = 1;
 877                patch->is_delete = 0;
 878                name = find_name_traditional(state, second, NULL, state->p_value);
 879                patch->new_name = name;
 880        } else if (is_dev_null(second)) {
 881                patch->is_new = 0;
 882                patch->is_delete = 1;
 883                name = find_name_traditional(state, first, NULL, state->p_value);
 884                patch->old_name = name;
 885        } else {
 886                char *first_name;
 887                first_name = find_name_traditional(state, first, NULL, state->p_value);
 888                name = find_name_traditional(state, second, first_name, state->p_value);
 889                free(first_name);
 890                if (has_epoch_timestamp(first)) {
 891                        patch->is_new = 1;
 892                        patch->is_delete = 0;
 893                        patch->new_name = name;
 894                } else if (has_epoch_timestamp(second)) {
 895                        patch->is_new = 0;
 896                        patch->is_delete = 1;
 897                        patch->old_name = name;
 898                } else {
 899                        patch->old_name = name;
 900                        patch->new_name = xstrdup_or_null(name);
 901                }
 902        }
 903        if (!name)
 904                return error(_("unable to find filename in patch at line %d"), state->linenr);
 905
 906        return 0;
 907}
 908
 909static int gitdiff_hdrend(struct apply_state *state,
 910                          const char *line,
 911                          struct patch *patch)
 912{
 913        return 1;
 914}
 915
 916/*
 917 * We're anal about diff header consistency, to make
 918 * sure that we don't end up having strange ambiguous
 919 * patches floating around.
 920 *
 921 * As a result, gitdiff_{old|new}name() will check
 922 * their names against any previous information, just
 923 * to make sure..
 924 */
 925#define DIFF_OLD_NAME 0
 926#define DIFF_NEW_NAME 1
 927
 928static int gitdiff_verify_name(struct apply_state *state,
 929                               const char *line,
 930                               int isnull,
 931                               char **name,
 932                               int side)
 933{
 934        if (!*name && !isnull) {
 935                *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
 936                return 0;
 937        }
 938
 939        if (*name) {
 940                char *another;
 941                if (isnull)
 942                        return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
 943                                     *name, state->linenr);
 944                another = find_name(state, line, NULL, state->p_value, TERM_TAB);
 945                if (!another || strcmp(another, *name)) {
 946                        free(another);
 947                        return error((side == DIFF_NEW_NAME) ?
 948                            _("git apply: bad git-diff - inconsistent new filename on line %d") :
 949                            _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
 950                }
 951                free(another);
 952        } else {
 953                if (!starts_with(line, "/dev/null\n"))
 954                        return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
 955        }
 956
 957        return 0;
 958}
 959
 960static int gitdiff_oldname(struct apply_state *state,
 961                           const char *line,
 962                           struct patch *patch)
 963{
 964        return gitdiff_verify_name(state, line,
 965                                   patch->is_new, &patch->old_name,
 966                                   DIFF_OLD_NAME);
 967}
 968
 969static int gitdiff_newname(struct apply_state *state,
 970                           const char *line,
 971                           struct patch *patch)
 972{
 973        return gitdiff_verify_name(state, line,
 974                                   patch->is_delete, &patch->new_name,
 975                                   DIFF_NEW_NAME);
 976}
 977
 978static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
 979{
 980        char *end;
 981        *mode = strtoul(line, &end, 8);
 982        if (end == line || !isspace(*end))
 983                return error(_("invalid mode on line %d: %s"), linenr, line);
 984        return 0;
 985}
 986
 987static int gitdiff_oldmode(struct apply_state *state,
 988                           const char *line,
 989                           struct patch *patch)
 990{
 991        return parse_mode_line(line, state->linenr, &patch->old_mode);
 992}
 993
 994static int gitdiff_newmode(struct apply_state *state,
 995                           const char *line,
 996                           struct patch *patch)
 997{
 998        return parse_mode_line(line, state->linenr, &patch->new_mode);
 999}
1000
1001static int gitdiff_delete(struct apply_state *state,
1002                          const char *line,
1003                          struct patch *patch)
1004{
1005        patch->is_delete = 1;
1006        free(patch->old_name);
1007        patch->old_name = xstrdup_or_null(patch->def_name);
1008        return gitdiff_oldmode(state, line, patch);
1009}
1010
1011static int gitdiff_newfile(struct apply_state *state,
1012                           const char *line,
1013                           struct patch *patch)
1014{
1015        patch->is_new = 1;
1016        free(patch->new_name);
1017        patch->new_name = xstrdup_or_null(patch->def_name);
1018        return gitdiff_newmode(state, line, patch);
1019}
1020
1021static int gitdiff_copysrc(struct apply_state *state,
1022                           const char *line,
1023                           struct patch *patch)
1024{
1025        patch->is_copy = 1;
1026        free(patch->old_name);
1027        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1028        return 0;
1029}
1030
1031static int gitdiff_copydst(struct apply_state *state,
1032                           const char *line,
1033                           struct patch *patch)
1034{
1035        patch->is_copy = 1;
1036        free(patch->new_name);
1037        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1038        return 0;
1039}
1040
1041static int gitdiff_renamesrc(struct apply_state *state,
1042                             const char *line,
1043                             struct patch *patch)
1044{
1045        patch->is_rename = 1;
1046        free(patch->old_name);
1047        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1048        return 0;
1049}
1050
1051static int gitdiff_renamedst(struct apply_state *state,
1052                             const char *line,
1053                             struct patch *patch)
1054{
1055        patch->is_rename = 1;
1056        free(patch->new_name);
1057        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1058        return 0;
1059}
1060
1061static int gitdiff_similarity(struct apply_state *state,
1062                              const char *line,
1063                              struct patch *patch)
1064{
1065        unsigned long val = strtoul(line, NULL, 10);
1066        if (val <= 100)
1067                patch->score = val;
1068        return 0;
1069}
1070
1071static int gitdiff_dissimilarity(struct apply_state *state,
1072                                 const char *line,
1073                                 struct patch *patch)
1074{
1075        unsigned long val = strtoul(line, NULL, 10);
1076        if (val <= 100)
1077                patch->score = val;
1078        return 0;
1079}
1080
1081static int gitdiff_index(struct apply_state *state,
1082                         const char *line,
1083                         struct patch *patch)
1084{
1085        /*
1086         * index line is N hexadecimal, "..", N hexadecimal,
1087         * and optional space with octal mode.
1088         */
1089        const char *ptr, *eol;
1090        int len;
1091
1092        ptr = strchr(line, '.');
1093        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1094                return 0;
1095        len = ptr - line;
1096        memcpy(patch->old_sha1_prefix, line, len);
1097        patch->old_sha1_prefix[len] = 0;
1098
1099        line = ptr + 2;
1100        ptr = strchr(line, ' ');
1101        eol = strchrnul(line, '\n');
1102
1103        if (!ptr || eol < ptr)
1104                ptr = eol;
1105        len = ptr - line;
1106
1107        if (40 < len)
1108                return 0;
1109        memcpy(patch->new_sha1_prefix, line, len);
1110        patch->new_sha1_prefix[len] = 0;
1111        if (*ptr == ' ')
1112                return gitdiff_oldmode(state, ptr + 1, patch);
1113        return 0;
1114}
1115
1116/*
1117 * This is normal for a diff that doesn't change anything: we'll fall through
1118 * into the next diff. Tell the parser to break out.
1119 */
1120static int gitdiff_unrecognized(struct apply_state *state,
1121                                const char *line,
1122                                struct patch *patch)
1123{
1124        return 1;
1125}
1126
1127/*
1128 * Skip p_value leading components from "line"; as we do not accept
1129 * absolute paths, return NULL in that case.
1130 */
1131static const char *skip_tree_prefix(struct apply_state *state,
1132                                    const char *line,
1133                                    int llen)
1134{
1135        int nslash;
1136        int i;
1137
1138        if (!state->p_value)
1139                return (llen && line[0] == '/') ? NULL : line;
1140
1141        nslash = state->p_value;
1142        for (i = 0; i < llen; i++) {
1143                int ch = line[i];
1144                if (ch == '/' && --nslash <= 0)
1145                        return (i == 0) ? NULL : &line[i + 1];
1146        }
1147        return NULL;
1148}
1149
1150/*
1151 * This is to extract the same name that appears on "diff --git"
1152 * line.  We do not find and return anything if it is a rename
1153 * patch, and it is OK because we will find the name elsewhere.
1154 * We need to reliably find name only when it is mode-change only,
1155 * creation or deletion of an empty file.  In any of these cases,
1156 * both sides are the same name under a/ and b/ respectively.
1157 */
1158static char *git_header_name(struct apply_state *state,
1159                             const char *line,
1160                             int llen)
1161{
1162        const char *name;
1163        const char *second = NULL;
1164        size_t len, line_len;
1165
1166        line += strlen("diff --git ");
1167        llen -= strlen("diff --git ");
1168
1169        if (*line == '"') {
1170                const char *cp;
1171                struct strbuf first = STRBUF_INIT;
1172                struct strbuf sp = STRBUF_INIT;
1173
1174                if (unquote_c_style(&first, line, &second))
1175                        goto free_and_fail1;
1176
1177                /* strip the a/b prefix including trailing slash */
1178                cp = skip_tree_prefix(state, first.buf, first.len);
1179                if (!cp)
1180                        goto free_and_fail1;
1181                strbuf_remove(&first, 0, cp - first.buf);
1182
1183                /*
1184                 * second points at one past closing dq of name.
1185                 * find the second name.
1186                 */
1187                while ((second < line + llen) && isspace(*second))
1188                        second++;
1189
1190                if (line + llen <= second)
1191                        goto free_and_fail1;
1192                if (*second == '"') {
1193                        if (unquote_c_style(&sp, second, NULL))
1194                                goto free_and_fail1;
1195                        cp = skip_tree_prefix(state, sp.buf, sp.len);
1196                        if (!cp)
1197                                goto free_and_fail1;
1198                        /* They must match, otherwise ignore */
1199                        if (strcmp(cp, first.buf))
1200                                goto free_and_fail1;
1201                        strbuf_release(&sp);
1202                        return strbuf_detach(&first, NULL);
1203                }
1204
1205                /* unquoted second */
1206                cp = skip_tree_prefix(state, second, line + llen - second);
1207                if (!cp)
1208                        goto free_and_fail1;
1209                if (line + llen - cp != first.len ||
1210                    memcmp(first.buf, cp, first.len))
1211                        goto free_and_fail1;
1212                return strbuf_detach(&first, NULL);
1213
1214        free_and_fail1:
1215                strbuf_release(&first);
1216                strbuf_release(&sp);
1217                return NULL;
1218        }
1219
1220        /* unquoted first name */
1221        name = skip_tree_prefix(state, line, llen);
1222        if (!name)
1223                return NULL;
1224
1225        /*
1226         * since the first name is unquoted, a dq if exists must be
1227         * the beginning of the second name.
1228         */
1229        for (second = name; second < line + llen; second++) {
1230                if (*second == '"') {
1231                        struct strbuf sp = STRBUF_INIT;
1232                        const char *np;
1233
1234                        if (unquote_c_style(&sp, second, NULL))
1235                                goto free_and_fail2;
1236
1237                        np = skip_tree_prefix(state, sp.buf, sp.len);
1238                        if (!np)
1239                                goto free_and_fail2;
1240
1241                        len = sp.buf + sp.len - np;
1242                        if (len < second - name &&
1243                            !strncmp(np, name, len) &&
1244                            isspace(name[len])) {
1245                                /* Good */
1246                                strbuf_remove(&sp, 0, np - sp.buf);
1247                                return strbuf_detach(&sp, NULL);
1248                        }
1249
1250                free_and_fail2:
1251                        strbuf_release(&sp);
1252                        return NULL;
1253                }
1254        }
1255
1256        /*
1257         * Accept a name only if it shows up twice, exactly the same
1258         * form.
1259         */
1260        second = strchr(name, '\n');
1261        if (!second)
1262                return NULL;
1263        line_len = second - name;
1264        for (len = 0 ; ; len++) {
1265                switch (name[len]) {
1266                default:
1267                        continue;
1268                case '\n':
1269                        return NULL;
1270                case '\t': case ' ':
1271                        /*
1272                         * Is this the separator between the preimage
1273                         * and the postimage pathname?  Again, we are
1274                         * only interested in the case where there is
1275                         * no rename, as this is only to set def_name
1276                         * and a rename patch has the names elsewhere
1277                         * in an unambiguous form.
1278                         */
1279                        if (!name[len + 1])
1280                                return NULL; /* no postimage name */
1281                        second = skip_tree_prefix(state, name + len + 1,
1282                                                  line_len - (len + 1));
1283                        if (!second)
1284                                return NULL;
1285                        /*
1286                         * Does len bytes starting at "name" and "second"
1287                         * (that are separated by one HT or SP we just
1288                         * found) exactly match?
1289                         */
1290                        if (second[len] == '\n' && !strncmp(name, second, len))
1291                                return xmemdupz(name, len);
1292                }
1293        }
1294}
1295
1296static int check_header_line(struct apply_state *state, struct patch *patch)
1297{
1298        int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1299                         (patch->is_rename == 1) + (patch->is_copy == 1);
1300        if (extensions > 1)
1301                return error(_("inconsistent header lines %d and %d"),
1302                             patch->extension_linenr, state->linenr);
1303        if (extensions && !patch->extension_linenr)
1304                patch->extension_linenr = state->linenr;
1305        return 0;
1306}
1307
1308/* Verify that we recognize the lines following a git header */
1309static int parse_git_header(struct apply_state *state,
1310                            const char *line,
1311                            int len,
1312                            unsigned int size,
1313                            struct patch *patch)
1314{
1315        unsigned long offset;
1316
1317        /* A git diff has explicit new/delete information, so we don't guess */
1318        patch->is_new = 0;
1319        patch->is_delete = 0;
1320
1321        /*
1322         * Some things may not have the old name in the
1323         * rest of the headers anywhere (pure mode changes,
1324         * or removing or adding empty files), so we get
1325         * the default name from the header.
1326         */
1327        patch->def_name = git_header_name(state, line, len);
1328        if (patch->def_name && state->root.len) {
1329                char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1330                free(patch->def_name);
1331                patch->def_name = s;
1332        }
1333
1334        line += len;
1335        size -= len;
1336        state->linenr++;
1337        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1338                static const struct opentry {
1339                        const char *str;
1340                        int (*fn)(struct apply_state *, const char *, struct patch *);
1341                } optable[] = {
1342                        { "@@ -", gitdiff_hdrend },
1343                        { "--- ", gitdiff_oldname },
1344                        { "+++ ", gitdiff_newname },
1345                        { "old mode ", gitdiff_oldmode },
1346                        { "new mode ", gitdiff_newmode },
1347                        { "deleted file mode ", gitdiff_delete },
1348                        { "new file mode ", gitdiff_newfile },
1349                        { "copy from ", gitdiff_copysrc },
1350                        { "copy to ", gitdiff_copydst },
1351                        { "rename old ", gitdiff_renamesrc },
1352                        { "rename new ", gitdiff_renamedst },
1353                        { "rename from ", gitdiff_renamesrc },
1354                        { "rename to ", gitdiff_renamedst },
1355                        { "similarity index ", gitdiff_similarity },
1356                        { "dissimilarity index ", gitdiff_dissimilarity },
1357                        { "index ", gitdiff_index },
1358                        { "", gitdiff_unrecognized },
1359                };
1360                int i;
1361
1362                len = linelen(line, size);
1363                if (!len || line[len-1] != '\n')
1364                        break;
1365                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1366                        const struct opentry *p = optable + i;
1367                        int oplen = strlen(p->str);
1368                        int res;
1369                        if (len < oplen || memcmp(p->str, line, oplen))
1370                                continue;
1371                        res = p->fn(state, line + oplen, patch);
1372                        if (res < 0)
1373                                return -1;
1374                        if (check_header_line(state, patch))
1375                                return -1;
1376                        if (res > 0)
1377                                return offset;
1378                        break;
1379                }
1380        }
1381
1382        return offset;
1383}
1384
1385static int parse_num(const char *line, unsigned long *p)
1386{
1387        char *ptr;
1388
1389        if (!isdigit(*line))
1390                return 0;
1391        *p = strtoul(line, &ptr, 10);
1392        return ptr - line;
1393}
1394
1395static int parse_range(const char *line, int len, int offset, const char *expect,
1396                       unsigned long *p1, unsigned long *p2)
1397{
1398        int digits, ex;
1399
1400        if (offset < 0 || offset >= len)
1401                return -1;
1402        line += offset;
1403        len -= offset;
1404
1405        digits = parse_num(line, p1);
1406        if (!digits)
1407                return -1;
1408
1409        offset += digits;
1410        line += digits;
1411        len -= digits;
1412
1413        *p2 = 1;
1414        if (*line == ',') {
1415                digits = parse_num(line+1, p2);
1416                if (!digits)
1417                        return -1;
1418
1419                offset += digits+1;
1420                line += digits+1;
1421                len -= digits+1;
1422        }
1423
1424        ex = strlen(expect);
1425        if (ex > len)
1426                return -1;
1427        if (memcmp(line, expect, ex))
1428                return -1;
1429
1430        return offset + ex;
1431}
1432
1433static void recount_diff(const char *line, int size, struct fragment *fragment)
1434{
1435        int oldlines = 0, newlines = 0, ret = 0;
1436
1437        if (size < 1) {
1438                warning("recount: ignore empty hunk");
1439                return;
1440        }
1441
1442        for (;;) {
1443                int len = linelen(line, size);
1444                size -= len;
1445                line += len;
1446
1447                if (size < 1)
1448                        break;
1449
1450                switch (*line) {
1451                case ' ': case '\n':
1452                        newlines++;
1453                        /* fall through */
1454                case '-':
1455                        oldlines++;
1456                        continue;
1457                case '+':
1458                        newlines++;
1459                        continue;
1460                case '\\':
1461                        continue;
1462                case '@':
1463                        ret = size < 3 || !starts_with(line, "@@ ");
1464                        break;
1465                case 'd':
1466                        ret = size < 5 || !starts_with(line, "diff ");
1467                        break;
1468                default:
1469                        ret = -1;
1470                        break;
1471                }
1472                if (ret) {
1473                        warning(_("recount: unexpected line: %.*s"),
1474                                (int)linelen(line, size), line);
1475                        return;
1476                }
1477                break;
1478        }
1479        fragment->oldlines = oldlines;
1480        fragment->newlines = newlines;
1481}
1482
1483/*
1484 * Parse a unified diff fragment header of the
1485 * form "@@ -a,b +c,d @@"
1486 */
1487static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1488{
1489        int offset;
1490
1491        if (!len || line[len-1] != '\n')
1492                return -1;
1493
1494        /* Figure out the number of lines in a fragment */
1495        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1496        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1497
1498        return offset;
1499}
1500
1501/*
1502 * Find file diff header
1503 *
1504 * Returns:
1505 *  -1 if no header was found
1506 *  -128 in case of error
1507 *   the size of the header in bytes (called "offset") otherwise
1508 */
1509static int find_header(struct apply_state *state,
1510                       const char *line,
1511                       unsigned long size,
1512                       int *hdrsize,
1513                       struct patch *patch)
1514{
1515        unsigned long offset, len;
1516
1517        patch->is_toplevel_relative = 0;
1518        patch->is_rename = patch->is_copy = 0;
1519        patch->is_new = patch->is_delete = -1;
1520        patch->old_mode = patch->new_mode = 0;
1521        patch->old_name = patch->new_name = NULL;
1522        for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1523                unsigned long nextlen;
1524
1525                len = linelen(line, size);
1526                if (!len)
1527                        break;
1528
1529                /* Testing this early allows us to take a few shortcuts.. */
1530                if (len < 6)
1531                        continue;
1532
1533                /*
1534                 * Make sure we don't find any unconnected patch fragments.
1535                 * That's a sign that we didn't find a header, and that a
1536                 * patch has become corrupted/broken up.
1537                 */
1538                if (!memcmp("@@ -", line, 4)) {
1539                        struct fragment dummy;
1540                        if (parse_fragment_header(line, len, &dummy) < 0)
1541                                continue;
1542                        error(_("patch fragment without header at line %d: %.*s"),
1543                                     state->linenr, (int)len-1, line);
1544                        return -128;
1545                }
1546
1547                if (size < len + 6)
1548                        break;
1549
1550                /*
1551                 * Git patch? It might not have a real patch, just a rename
1552                 * or mode change, so we handle that specially
1553                 */
1554                if (!memcmp("diff --git ", line, 11)) {
1555                        int git_hdr_len = parse_git_header(state, line, len, size, patch);
1556                        if (git_hdr_len < 0)
1557                                return -128;
1558                        if (git_hdr_len <= len)
1559                                continue;
1560                        if (!patch->old_name && !patch->new_name) {
1561                                if (!patch->def_name) {
1562                                        error(Q_("git diff header lacks filename information when removing "
1563                                                        "%d leading pathname component (line %d)",
1564                                                        "git diff header lacks filename information when removing "
1565                                                        "%d leading pathname components (line %d)",
1566                                                        state->p_value),
1567                                                     state->p_value, state->linenr);
1568                                        return -128;
1569                                }
1570                                patch->old_name = xstrdup(patch->def_name);
1571                                patch->new_name = xstrdup(patch->def_name);
1572                        }
1573                        if ((!patch->new_name && !patch->is_delete) ||
1574                            (!patch->old_name && !patch->is_new)) {
1575                                error(_("git diff header lacks filename information "
1576                                             "(line %d)"), state->linenr);
1577                                return -128;
1578                        }
1579                        patch->is_toplevel_relative = 1;
1580                        *hdrsize = git_hdr_len;
1581                        return offset;
1582                }
1583
1584                /* --- followed by +++ ? */
1585                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1586                        continue;
1587
1588                /*
1589                 * We only accept unified patches, so we want it to
1590                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1591                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1592                 */
1593                nextlen = linelen(line + len, size - len);
1594                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1595                        continue;
1596
1597                /* Ok, we'll consider it a patch */
1598                if (parse_traditional_patch(state, line, line+len, patch))
1599                        return -128;
1600                *hdrsize = len + nextlen;
1601                state->linenr += 2;
1602                return offset;
1603        }
1604        return -1;
1605}
1606
1607static void record_ws_error(struct apply_state *state,
1608                            unsigned result,
1609                            const char *line,
1610                            int len,
1611                            int linenr)
1612{
1613        char *err;
1614
1615        if (!result)
1616                return;
1617
1618        state->whitespace_error++;
1619        if (state->squelch_whitespace_errors &&
1620            state->squelch_whitespace_errors < state->whitespace_error)
1621                return;
1622
1623        err = whitespace_error_string(result);
1624        if (state->apply_verbosity > verbosity_silent)
1625                fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1626                        state->patch_input_file, linenr, err, len, line);
1627        free(err);
1628}
1629
1630static void check_whitespace(struct apply_state *state,
1631                             const char *line,
1632                             int len,
1633                             unsigned ws_rule)
1634{
1635        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1636
1637        record_ws_error(state, result, line + 1, len - 2, state->linenr);
1638}
1639
1640/*
1641 * Check if the patch has context lines with CRLF or
1642 * the patch wants to remove lines with CRLF.
1643 */
1644static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1645{
1646        if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1647                patch->ws_rule |= WS_CR_AT_EOL;
1648                patch->crlf_in_old = 1;
1649        }
1650}
1651
1652
1653/*
1654 * Parse a unified diff. Note that this really needs to parse each
1655 * fragment separately, since the only way to know the difference
1656 * between a "---" that is part of a patch, and a "---" that starts
1657 * the next patch is to look at the line counts..
1658 */
1659static int parse_fragment(struct apply_state *state,
1660                          const char *line,
1661                          unsigned long size,
1662                          struct patch *patch,
1663                          struct fragment *fragment)
1664{
1665        int added, deleted;
1666        int len = linelen(line, size), offset;
1667        unsigned long oldlines, newlines;
1668        unsigned long leading, trailing;
1669
1670        offset = parse_fragment_header(line, len, fragment);
1671        if (offset < 0)
1672                return -1;
1673        if (offset > 0 && patch->recount)
1674                recount_diff(line + offset, size - offset, fragment);
1675        oldlines = fragment->oldlines;
1676        newlines = fragment->newlines;
1677        leading = 0;
1678        trailing = 0;
1679
1680        /* Parse the thing.. */
1681        line += len;
1682        size -= len;
1683        state->linenr++;
1684        added = deleted = 0;
1685        for (offset = len;
1686             0 < size;
1687             offset += len, size -= len, line += len, state->linenr++) {
1688                if (!oldlines && !newlines)
1689                        break;
1690                len = linelen(line, size);
1691                if (!len || line[len-1] != '\n')
1692                        return -1;
1693                switch (*line) {
1694                default:
1695                        return -1;
1696                case '\n': /* newer GNU diff, an empty context line */
1697                case ' ':
1698                        oldlines--;
1699                        newlines--;
1700                        if (!deleted && !added)
1701                                leading++;
1702                        trailing++;
1703                        check_old_for_crlf(patch, line, len);
1704                        if (!state->apply_in_reverse &&
1705                            state->ws_error_action == correct_ws_error)
1706                                check_whitespace(state, line, len, patch->ws_rule);
1707                        break;
1708                case '-':
1709                        if (!state->apply_in_reverse)
1710                                check_old_for_crlf(patch, line, len);
1711                        if (state->apply_in_reverse &&
1712                            state->ws_error_action != nowarn_ws_error)
1713                                check_whitespace(state, line, len, patch->ws_rule);
1714                        deleted++;
1715                        oldlines--;
1716                        trailing = 0;
1717                        break;
1718                case '+':
1719                        if (state->apply_in_reverse)
1720                                check_old_for_crlf(patch, line, len);
1721                        if (!state->apply_in_reverse &&
1722                            state->ws_error_action != nowarn_ws_error)
1723                                check_whitespace(state, line, len, patch->ws_rule);
1724                        added++;
1725                        newlines--;
1726                        trailing = 0;
1727                        break;
1728
1729                /*
1730                 * We allow "\ No newline at end of file". Depending
1731                 * on locale settings when the patch was produced we
1732                 * don't know what this line looks like. The only
1733                 * thing we do know is that it begins with "\ ".
1734                 * Checking for 12 is just for sanity check -- any
1735                 * l10n of "\ No newline..." is at least that long.
1736                 */
1737                case '\\':
1738                        if (len < 12 || memcmp(line, "\\ ", 2))
1739                                return -1;
1740                        break;
1741                }
1742        }
1743        if (oldlines || newlines)
1744                return -1;
1745        if (!deleted && !added)
1746                return -1;
1747
1748        fragment->leading = leading;
1749        fragment->trailing = trailing;
1750
1751        /*
1752         * If a fragment ends with an incomplete line, we failed to include
1753         * it in the above loop because we hit oldlines == newlines == 0
1754         * before seeing it.
1755         */
1756        if (12 < size && !memcmp(line, "\\ ", 2))
1757                offset += linelen(line, size);
1758
1759        patch->lines_added += added;
1760        patch->lines_deleted += deleted;
1761
1762        if (0 < patch->is_new && oldlines)
1763                return error(_("new file depends on old contents"));
1764        if (0 < patch->is_delete && newlines)
1765                return error(_("deleted file still has contents"));
1766        return offset;
1767}
1768
1769/*
1770 * We have seen "diff --git a/... b/..." header (or a traditional patch
1771 * header).  Read hunks that belong to this patch into fragments and hang
1772 * them to the given patch structure.
1773 *
1774 * The (fragment->patch, fragment->size) pair points into the memory given
1775 * by the caller, not a copy, when we return.
1776 *
1777 * Returns:
1778 *   -1 in case of error,
1779 *   the number of bytes in the patch otherwise.
1780 */
1781static int parse_single_patch(struct apply_state *state,
1782                              const char *line,
1783                              unsigned long size,
1784                              struct patch *patch)
1785{
1786        unsigned long offset = 0;
1787        unsigned long oldlines = 0, newlines = 0, context = 0;
1788        struct fragment **fragp = &patch->fragments;
1789
1790        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1791                struct fragment *fragment;
1792                int len;
1793
1794                fragment = xcalloc(1, sizeof(*fragment));
1795                fragment->linenr = state->linenr;
1796                len = parse_fragment(state, line, size, patch, fragment);
1797                if (len <= 0) {
1798                        free(fragment);
1799                        return error(_("corrupt patch at line %d"), state->linenr);
1800                }
1801                fragment->patch = line;
1802                fragment->size = len;
1803                oldlines += fragment->oldlines;
1804                newlines += fragment->newlines;
1805                context += fragment->leading + fragment->trailing;
1806
1807                *fragp = fragment;
1808                fragp = &fragment->next;
1809
1810                offset += len;
1811                line += len;
1812                size -= len;
1813        }
1814
1815        /*
1816         * If something was removed (i.e. we have old-lines) it cannot
1817         * be creation, and if something was added it cannot be
1818         * deletion.  However, the reverse is not true; --unified=0
1819         * patches that only add are not necessarily creation even
1820         * though they do not have any old lines, and ones that only
1821         * delete are not necessarily deletion.
1822         *
1823         * Unfortunately, a real creation/deletion patch do _not_ have
1824         * any context line by definition, so we cannot safely tell it
1825         * apart with --unified=0 insanity.  At least if the patch has
1826         * more than one hunk it is not creation or deletion.
1827         */
1828        if (patch->is_new < 0 &&
1829            (oldlines || (patch->fragments && patch->fragments->next)))
1830                patch->is_new = 0;
1831        if (patch->is_delete < 0 &&
1832            (newlines || (patch->fragments && patch->fragments->next)))
1833                patch->is_delete = 0;
1834
1835        if (0 < patch->is_new && oldlines)
1836                return error(_("new file %s depends on old contents"), patch->new_name);
1837        if (0 < patch->is_delete && newlines)
1838                return error(_("deleted file %s still has contents"), patch->old_name);
1839        if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1840                fprintf_ln(stderr,
1841                           _("** warning: "
1842                             "file %s becomes empty but is not deleted"),
1843                           patch->new_name);
1844
1845        return offset;
1846}
1847
1848static inline int metadata_changes(struct patch *patch)
1849{
1850        return  patch->is_rename > 0 ||
1851                patch->is_copy > 0 ||
1852                patch->is_new > 0 ||
1853                patch->is_delete ||
1854                (patch->old_mode && patch->new_mode &&
1855                 patch->old_mode != patch->new_mode);
1856}
1857
1858static char *inflate_it(const void *data, unsigned long size,
1859                        unsigned long inflated_size)
1860{
1861        git_zstream stream;
1862        void *out;
1863        int st;
1864
1865        memset(&stream, 0, sizeof(stream));
1866
1867        stream.next_in = (unsigned char *)data;
1868        stream.avail_in = size;
1869        stream.next_out = out = xmalloc(inflated_size);
1870        stream.avail_out = inflated_size;
1871        git_inflate_init(&stream);
1872        st = git_inflate(&stream, Z_FINISH);
1873        git_inflate_end(&stream);
1874        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1875                free(out);
1876                return NULL;
1877        }
1878        return out;
1879}
1880
1881/*
1882 * Read a binary hunk and return a new fragment; fragment->patch
1883 * points at an allocated memory that the caller must free, so
1884 * it is marked as "->free_patch = 1".
1885 */
1886static struct fragment *parse_binary_hunk(struct apply_state *state,
1887                                          char **buf_p,
1888                                          unsigned long *sz_p,
1889                                          int *status_p,
1890                                          int *used_p)
1891{
1892        /*
1893         * Expect a line that begins with binary patch method ("literal"
1894         * or "delta"), followed by the length of data before deflating.
1895         * a sequence of 'length-byte' followed by base-85 encoded data
1896         * should follow, terminated by a newline.
1897         *
1898         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1899         * and we would limit the patch line to 66 characters,
1900         * so one line can fit up to 13 groups that would decode
1901         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1902         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1903         */
1904        int llen, used;
1905        unsigned long size = *sz_p;
1906        char *buffer = *buf_p;
1907        int patch_method;
1908        unsigned long origlen;
1909        char *data = NULL;
1910        int hunk_size = 0;
1911        struct fragment *frag;
1912
1913        llen = linelen(buffer, size);
1914        used = llen;
1915
1916        *status_p = 0;
1917
1918        if (starts_with(buffer, "delta ")) {
1919                patch_method = BINARY_DELTA_DEFLATED;
1920                origlen = strtoul(buffer + 6, NULL, 10);
1921        }
1922        else if (starts_with(buffer, "literal ")) {
1923                patch_method = BINARY_LITERAL_DEFLATED;
1924                origlen = strtoul(buffer + 8, NULL, 10);
1925        }
1926        else
1927                return NULL;
1928
1929        state->linenr++;
1930        buffer += llen;
1931        while (1) {
1932                int byte_length, max_byte_length, newsize;
1933                llen = linelen(buffer, size);
1934                used += llen;
1935                state->linenr++;
1936                if (llen == 1) {
1937                        /* consume the blank line */
1938                        buffer++;
1939                        size--;
1940                        break;
1941                }
1942                /*
1943                 * Minimum line is "A00000\n" which is 7-byte long,
1944                 * and the line length must be multiple of 5 plus 2.
1945                 */
1946                if ((llen < 7) || (llen-2) % 5)
1947                        goto corrupt;
1948                max_byte_length = (llen - 2) / 5 * 4;
1949                byte_length = *buffer;
1950                if ('A' <= byte_length && byte_length <= 'Z')
1951                        byte_length = byte_length - 'A' + 1;
1952                else if ('a' <= byte_length && byte_length <= 'z')
1953                        byte_length = byte_length - 'a' + 27;
1954                else
1955                        goto corrupt;
1956                /* if the input length was not multiple of 4, we would
1957                 * have filler at the end but the filler should never
1958                 * exceed 3 bytes
1959                 */
1960                if (max_byte_length < byte_length ||
1961                    byte_length <= max_byte_length - 4)
1962                        goto corrupt;
1963                newsize = hunk_size + byte_length;
1964                data = xrealloc(data, newsize);
1965                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1966                        goto corrupt;
1967                hunk_size = newsize;
1968                buffer += llen;
1969                size -= llen;
1970        }
1971
1972        frag = xcalloc(1, sizeof(*frag));
1973        frag->patch = inflate_it(data, hunk_size, origlen);
1974        frag->free_patch = 1;
1975        if (!frag->patch)
1976                goto corrupt;
1977        free(data);
1978        frag->size = origlen;
1979        *buf_p = buffer;
1980        *sz_p = size;
1981        *used_p = used;
1982        frag->binary_patch_method = patch_method;
1983        return frag;
1984
1985 corrupt:
1986        free(data);
1987        *status_p = -1;
1988        error(_("corrupt binary patch at line %d: %.*s"),
1989              state->linenr-1, llen-1, buffer);
1990        return NULL;
1991}
1992
1993/*
1994 * Returns:
1995 *   -1 in case of error,
1996 *   the length of the parsed binary patch otherwise
1997 */
1998static int parse_binary(struct apply_state *state,
1999                        char *buffer,
2000                        unsigned long size,
2001                        struct patch *patch)
2002{
2003        /*
2004         * We have read "GIT binary patch\n"; what follows is a line
2005         * that says the patch method (currently, either "literal" or
2006         * "delta") and the length of data before deflating; a
2007         * sequence of 'length-byte' followed by base-85 encoded data
2008         * follows.
2009         *
2010         * When a binary patch is reversible, there is another binary
2011         * hunk in the same format, starting with patch method (either
2012         * "literal" or "delta") with the length of data, and a sequence
2013         * of length-byte + base-85 encoded data, terminated with another
2014         * empty line.  This data, when applied to the postimage, produces
2015         * the preimage.
2016         */
2017        struct fragment *forward;
2018        struct fragment *reverse;
2019        int status;
2020        int used, used_1;
2021
2022        forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2023        if (!forward && !status)
2024                /* there has to be one hunk (forward hunk) */
2025                return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2026        if (status)
2027                /* otherwise we already gave an error message */
2028                return status;
2029
2030        reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2031        if (reverse)
2032                used += used_1;
2033        else if (status) {
2034                /*
2035                 * Not having reverse hunk is not an error, but having
2036                 * a corrupt reverse hunk is.
2037                 */
2038                free((void*) forward->patch);
2039                free(forward);
2040                return status;
2041        }
2042        forward->next = reverse;
2043        patch->fragments = forward;
2044        patch->is_binary = 1;
2045        return used;
2046}
2047
2048static void prefix_one(struct apply_state *state, char **name)
2049{
2050        char *old_name = *name;
2051        if (!old_name)
2052                return;
2053        *name = prefix_filename(state->prefix, *name);
2054        free(old_name);
2055}
2056
2057static void prefix_patch(struct apply_state *state, struct patch *p)
2058{
2059        if (!state->prefix || p->is_toplevel_relative)
2060                return;
2061        prefix_one(state, &p->new_name);
2062        prefix_one(state, &p->old_name);
2063}
2064
2065/*
2066 * include/exclude
2067 */
2068
2069static void add_name_limit(struct apply_state *state,
2070                           const char *name,
2071                           int exclude)
2072{
2073        struct string_list_item *it;
2074
2075        it = string_list_append(&state->limit_by_name, name);
2076        it->util = exclude ? NULL : (void *) 1;
2077}
2078
2079static int use_patch(struct apply_state *state, struct patch *p)
2080{
2081        const char *pathname = p->new_name ? p->new_name : p->old_name;
2082        int i;
2083
2084        /* Paths outside are not touched regardless of "--include" */
2085        if (state->prefix && *state->prefix) {
2086                const char *rest;
2087                if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2088                        return 0;
2089        }
2090
2091        /* See if it matches any of exclude/include rule */
2092        for (i = 0; i < state->limit_by_name.nr; i++) {
2093                struct string_list_item *it = &state->limit_by_name.items[i];
2094                if (!wildmatch(it->string, pathname, 0))
2095                        return (it->util != NULL);
2096        }
2097
2098        /*
2099         * If we had any include, a path that does not match any rule is
2100         * not used.  Otherwise, we saw bunch of exclude rules (or none)
2101         * and such a path is used.
2102         */
2103        return !state->has_include;
2104}
2105
2106/*
2107 * Read the patch text in "buffer" that extends for "size" bytes; stop
2108 * reading after seeing a single patch (i.e. changes to a single file).
2109 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2110 *
2111 * Returns:
2112 *   -1 if no header was found or parse_binary() failed,
2113 *   -128 on another error,
2114 *   the number of bytes consumed otherwise,
2115 *     so that the caller can call us again for the next patch.
2116 */
2117static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2118{
2119        int hdrsize, patchsize;
2120        int offset = find_header(state, buffer, size, &hdrsize, patch);
2121
2122        if (offset < 0)
2123                return offset;
2124
2125        prefix_patch(state, patch);
2126
2127        if (!use_patch(state, patch))
2128                patch->ws_rule = 0;
2129        else
2130                patch->ws_rule = whitespace_rule(patch->new_name
2131                                                 ? patch->new_name
2132                                                 : patch->old_name);
2133
2134        patchsize = parse_single_patch(state,
2135                                       buffer + offset + hdrsize,
2136                                       size - offset - hdrsize,
2137                                       patch);
2138
2139        if (patchsize < 0)
2140                return -128;
2141
2142        if (!patchsize) {
2143                static const char git_binary[] = "GIT binary patch\n";
2144                int hd = hdrsize + offset;
2145                unsigned long llen = linelen(buffer + hd, size - hd);
2146
2147                if (llen == sizeof(git_binary) - 1 &&
2148                    !memcmp(git_binary, buffer + hd, llen)) {
2149                        int used;
2150                        state->linenr++;
2151                        used = parse_binary(state, buffer + hd + llen,
2152                                            size - hd - llen, patch);
2153                        if (used < 0)
2154                                return -1;
2155                        if (used)
2156                                patchsize = used + llen;
2157                        else
2158                                patchsize = 0;
2159                }
2160                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2161                        static const char *binhdr[] = {
2162                                "Binary files ",
2163                                "Files ",
2164                                NULL,
2165                        };
2166                        int i;
2167                        for (i = 0; binhdr[i]; i++) {
2168                                int len = strlen(binhdr[i]);
2169                                if (len < size - hd &&
2170                                    !memcmp(binhdr[i], buffer + hd, len)) {
2171                                        state->linenr++;
2172                                        patch->is_binary = 1;
2173                                        patchsize = llen;
2174                                        break;
2175                                }
2176                        }
2177                }
2178
2179                /* Empty patch cannot be applied if it is a text patch
2180                 * without metadata change.  A binary patch appears
2181                 * empty to us here.
2182                 */
2183                if ((state->apply || state->check) &&
2184                    (!patch->is_binary && !metadata_changes(patch))) {
2185                        error(_("patch with only garbage at line %d"), state->linenr);
2186                        return -128;
2187                }
2188        }
2189
2190        return offset + hdrsize + patchsize;
2191}
2192
2193static void reverse_patches(struct patch *p)
2194{
2195        for (; p; p = p->next) {
2196                struct fragment *frag = p->fragments;
2197
2198                SWAP(p->new_name, p->old_name);
2199                SWAP(p->new_mode, p->old_mode);
2200                SWAP(p->is_new, p->is_delete);
2201                SWAP(p->lines_added, p->lines_deleted);
2202                SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2203
2204                for (; frag; frag = frag->next) {
2205                        SWAP(frag->newpos, frag->oldpos);
2206                        SWAP(frag->newlines, frag->oldlines);
2207                }
2208        }
2209}
2210
2211static const char pluses[] =
2212"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2213static const char minuses[]=
2214"----------------------------------------------------------------------";
2215
2216static void show_stats(struct apply_state *state, struct patch *patch)
2217{
2218        struct strbuf qname = STRBUF_INIT;
2219        char *cp = patch->new_name ? patch->new_name : patch->old_name;
2220        int max, add, del;
2221
2222        quote_c_style(cp, &qname, NULL, 0);
2223
2224        /*
2225         * "scale" the filename
2226         */
2227        max = state->max_len;
2228        if (max > 50)
2229                max = 50;
2230
2231        if (qname.len > max) {
2232                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2233                if (!cp)
2234                        cp = qname.buf + qname.len + 3 - max;
2235                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2236        }
2237
2238        if (patch->is_binary) {
2239                printf(" %-*s |  Bin\n", max, qname.buf);
2240                strbuf_release(&qname);
2241                return;
2242        }
2243
2244        printf(" %-*s |", max, qname.buf);
2245        strbuf_release(&qname);
2246
2247        /*
2248         * scale the add/delete
2249         */
2250        max = max + state->max_change > 70 ? 70 - max : state->max_change;
2251        add = patch->lines_added;
2252        del = patch->lines_deleted;
2253
2254        if (state->max_change > 0) {
2255                int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2256                add = (add * max + state->max_change / 2) / state->max_change;
2257                del = total - add;
2258        }
2259        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2260                add, pluses, del, minuses);
2261}
2262
2263static int read_old_data(struct stat *st, struct patch *patch,
2264                         const char *path, struct strbuf *buf)
2265{
2266        enum safe_crlf safe_crlf = patch->crlf_in_old ?
2267                SAFE_CRLF_KEEP_CRLF : SAFE_CRLF_RENORMALIZE;
2268        switch (st->st_mode & S_IFMT) {
2269        case S_IFLNK:
2270                if (strbuf_readlink(buf, path, st->st_size) < 0)
2271                        return error(_("unable to read symlink %s"), path);
2272                return 0;
2273        case S_IFREG:
2274                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2275                        return error(_("unable to open or read %s"), path);
2276                /*
2277                 * "git apply" without "--index/--cached" should never look
2278                 * at the index; the target file may not have been added to
2279                 * the index yet, and we may not even be in any Git repository.
2280                 * Pass NULL to convert_to_git() to stress this; the function
2281                 * should never look at the index when explicit crlf option
2282                 * is given.
2283                 */
2284                convert_to_git(NULL, path, buf->buf, buf->len, buf, safe_crlf);
2285                return 0;
2286        default:
2287                return -1;
2288        }
2289}
2290
2291/*
2292 * Update the preimage, and the common lines in postimage,
2293 * from buffer buf of length len. If postlen is 0 the postimage
2294 * is updated in place, otherwise it's updated on a new buffer
2295 * of length postlen
2296 */
2297
2298static void update_pre_post_images(struct image *preimage,
2299                                   struct image *postimage,
2300                                   char *buf,
2301                                   size_t len, size_t postlen)
2302{
2303        int i, ctx, reduced;
2304        char *new, *old, *fixed;
2305        struct image fixed_preimage;
2306
2307        /*
2308         * Update the preimage with whitespace fixes.  Note that we
2309         * are not losing preimage->buf -- apply_one_fragment() will
2310         * free "oldlines".
2311         */
2312        prepare_image(&fixed_preimage, buf, len, 1);
2313        assert(postlen
2314               ? fixed_preimage.nr == preimage->nr
2315               : fixed_preimage.nr <= preimage->nr);
2316        for (i = 0; i < fixed_preimage.nr; i++)
2317                fixed_preimage.line[i].flag = preimage->line[i].flag;
2318        free(preimage->line_allocated);
2319        *preimage = fixed_preimage;
2320
2321        /*
2322         * Adjust the common context lines in postimage. This can be
2323         * done in-place when we are shrinking it with whitespace
2324         * fixing, but needs a new buffer when ignoring whitespace or
2325         * expanding leading tabs to spaces.
2326         *
2327         * We trust the caller to tell us if the update can be done
2328         * in place (postlen==0) or not.
2329         */
2330        old = postimage->buf;
2331        if (postlen)
2332                new = postimage->buf = xmalloc(postlen);
2333        else
2334                new = old;
2335        fixed = preimage->buf;
2336
2337        for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2338                size_t l_len = postimage->line[i].len;
2339                if (!(postimage->line[i].flag & LINE_COMMON)) {
2340                        /* an added line -- no counterparts in preimage */
2341                        memmove(new, old, l_len);
2342                        old += l_len;
2343                        new += l_len;
2344                        continue;
2345                }
2346
2347                /* a common context -- skip it in the original postimage */
2348                old += l_len;
2349
2350                /* and find the corresponding one in the fixed preimage */
2351                while (ctx < preimage->nr &&
2352                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2353                        fixed += preimage->line[ctx].len;
2354                        ctx++;
2355                }
2356
2357                /*
2358                 * preimage is expected to run out, if the caller
2359                 * fixed addition of trailing blank lines.
2360                 */
2361                if (preimage->nr <= ctx) {
2362                        reduced++;
2363                        continue;
2364                }
2365
2366                /* and copy it in, while fixing the line length */
2367                l_len = preimage->line[ctx].len;
2368                memcpy(new, fixed, l_len);
2369                new += l_len;
2370                fixed += l_len;
2371                postimage->line[i].len = l_len;
2372                ctx++;
2373        }
2374
2375        if (postlen
2376            ? postlen < new - postimage->buf
2377            : postimage->len < new - postimage->buf)
2378                die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2379                    (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2380
2381        /* Fix the length of the whole thing */
2382        postimage->len = new - postimage->buf;
2383        postimage->nr -= reduced;
2384}
2385
2386static int line_by_line_fuzzy_match(struct image *img,
2387                                    struct image *preimage,
2388                                    struct image *postimage,
2389                                    unsigned long try,
2390                                    int try_lno,
2391                                    int preimage_limit)
2392{
2393        int i;
2394        size_t imgoff = 0;
2395        size_t preoff = 0;
2396        size_t postlen = postimage->len;
2397        size_t extra_chars;
2398        char *buf;
2399        char *preimage_eof;
2400        char *preimage_end;
2401        struct strbuf fixed;
2402        char *fixed_buf;
2403        size_t fixed_len;
2404
2405        for (i = 0; i < preimage_limit; i++) {
2406                size_t prelen = preimage->line[i].len;
2407                size_t imglen = img->line[try_lno+i].len;
2408
2409                if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2410                                      preimage->buf + preoff, prelen))
2411                        return 0;
2412                if (preimage->line[i].flag & LINE_COMMON)
2413                        postlen += imglen - prelen;
2414                imgoff += imglen;
2415                preoff += prelen;
2416        }
2417
2418        /*
2419         * Ok, the preimage matches with whitespace fuzz.
2420         *
2421         * imgoff now holds the true length of the target that
2422         * matches the preimage before the end of the file.
2423         *
2424         * Count the number of characters in the preimage that fall
2425         * beyond the end of the file and make sure that all of them
2426         * are whitespace characters. (This can only happen if
2427         * we are removing blank lines at the end of the file.)
2428         */
2429        buf = preimage_eof = preimage->buf + preoff;
2430        for ( ; i < preimage->nr; i++)
2431                preoff += preimage->line[i].len;
2432        preimage_end = preimage->buf + preoff;
2433        for ( ; buf < preimage_end; buf++)
2434                if (!isspace(*buf))
2435                        return 0;
2436
2437        /*
2438         * Update the preimage and the common postimage context
2439         * lines to use the same whitespace as the target.
2440         * If whitespace is missing in the target (i.e.
2441         * if the preimage extends beyond the end of the file),
2442         * use the whitespace from the preimage.
2443         */
2444        extra_chars = preimage_end - preimage_eof;
2445        strbuf_init(&fixed, imgoff + extra_chars);
2446        strbuf_add(&fixed, img->buf + try, imgoff);
2447        strbuf_add(&fixed, preimage_eof, extra_chars);
2448        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2449        update_pre_post_images(preimage, postimage,
2450                               fixed_buf, fixed_len, postlen);
2451        return 1;
2452}
2453
2454static int match_fragment(struct apply_state *state,
2455                          struct image *img,
2456                          struct image *preimage,
2457                          struct image *postimage,
2458                          unsigned long try,
2459                          int try_lno,
2460                          unsigned ws_rule,
2461                          int match_beginning, int match_end)
2462{
2463        int i;
2464        char *fixed_buf, *buf, *orig, *target;
2465        struct strbuf fixed;
2466        size_t fixed_len, postlen;
2467        int preimage_limit;
2468
2469        if (preimage->nr + try_lno <= img->nr) {
2470                /*
2471                 * The hunk falls within the boundaries of img.
2472                 */
2473                preimage_limit = preimage->nr;
2474                if (match_end && (preimage->nr + try_lno != img->nr))
2475                        return 0;
2476        } else if (state->ws_error_action == correct_ws_error &&
2477                   (ws_rule & WS_BLANK_AT_EOF)) {
2478                /*
2479                 * This hunk extends beyond the end of img, and we are
2480                 * removing blank lines at the end of the file.  This
2481                 * many lines from the beginning of the preimage must
2482                 * match with img, and the remainder of the preimage
2483                 * must be blank.
2484                 */
2485                preimage_limit = img->nr - try_lno;
2486        } else {
2487                /*
2488                 * The hunk extends beyond the end of the img and
2489                 * we are not removing blanks at the end, so we
2490                 * should reject the hunk at this position.
2491                 */
2492                return 0;
2493        }
2494
2495        if (match_beginning && try_lno)
2496                return 0;
2497
2498        /* Quick hash check */
2499        for (i = 0; i < preimage_limit; i++)
2500                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2501                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2502                        return 0;
2503
2504        if (preimage_limit == preimage->nr) {
2505                /*
2506                 * Do we have an exact match?  If we were told to match
2507                 * at the end, size must be exactly at try+fragsize,
2508                 * otherwise try+fragsize must be still within the preimage,
2509                 * and either case, the old piece should match the preimage
2510                 * exactly.
2511                 */
2512                if ((match_end
2513                     ? (try + preimage->len == img->len)
2514                     : (try + preimage->len <= img->len)) &&
2515                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2516                        return 1;
2517        } else {
2518                /*
2519                 * The preimage extends beyond the end of img, so
2520                 * there cannot be an exact match.
2521                 *
2522                 * There must be one non-blank context line that match
2523                 * a line before the end of img.
2524                 */
2525                char *buf_end;
2526
2527                buf = preimage->buf;
2528                buf_end = buf;
2529                for (i = 0; i < preimage_limit; i++)
2530                        buf_end += preimage->line[i].len;
2531
2532                for ( ; buf < buf_end; buf++)
2533                        if (!isspace(*buf))
2534                                break;
2535                if (buf == buf_end)
2536                        return 0;
2537        }
2538
2539        /*
2540         * No exact match. If we are ignoring whitespace, run a line-by-line
2541         * fuzzy matching. We collect all the line length information because
2542         * we need it to adjust whitespace if we match.
2543         */
2544        if (state->ws_ignore_action == ignore_ws_change)
2545                return line_by_line_fuzzy_match(img, preimage, postimage,
2546                                                try, try_lno, preimage_limit);
2547
2548        if (state->ws_error_action != correct_ws_error)
2549                return 0;
2550
2551        /*
2552         * The hunk does not apply byte-by-byte, but the hash says
2553         * it might with whitespace fuzz. We weren't asked to
2554         * ignore whitespace, we were asked to correct whitespace
2555         * errors, so let's try matching after whitespace correction.
2556         *
2557         * While checking the preimage against the target, whitespace
2558         * errors in both fixed, we count how large the corresponding
2559         * postimage needs to be.  The postimage prepared by
2560         * apply_one_fragment() has whitespace errors fixed on added
2561         * lines already, but the common lines were propagated as-is,
2562         * which may become longer when their whitespace errors are
2563         * fixed.
2564         */
2565
2566        /* First count added lines in postimage */
2567        postlen = 0;
2568        for (i = 0; i < postimage->nr; i++) {
2569                if (!(postimage->line[i].flag & LINE_COMMON))
2570                        postlen += postimage->line[i].len;
2571        }
2572
2573        /*
2574         * The preimage may extend beyond the end of the file,
2575         * but in this loop we will only handle the part of the
2576         * preimage that falls within the file.
2577         */
2578        strbuf_init(&fixed, preimage->len + 1);
2579        orig = preimage->buf;
2580        target = img->buf + try;
2581        for (i = 0; i < preimage_limit; i++) {
2582                size_t oldlen = preimage->line[i].len;
2583                size_t tgtlen = img->line[try_lno + i].len;
2584                size_t fixstart = fixed.len;
2585                struct strbuf tgtfix;
2586                int match;
2587
2588                /* Try fixing the line in the preimage */
2589                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2590
2591                /* Try fixing the line in the target */
2592                strbuf_init(&tgtfix, tgtlen);
2593                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2594
2595                /*
2596                 * If they match, either the preimage was based on
2597                 * a version before our tree fixed whitespace breakage,
2598                 * or we are lacking a whitespace-fix patch the tree
2599                 * the preimage was based on already had (i.e. target
2600                 * has whitespace breakage, the preimage doesn't).
2601                 * In either case, we are fixing the whitespace breakages
2602                 * so we might as well take the fix together with their
2603                 * real change.
2604                 */
2605                match = (tgtfix.len == fixed.len - fixstart &&
2606                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2607                                             fixed.len - fixstart));
2608
2609                /* Add the length if this is common with the postimage */
2610                if (preimage->line[i].flag & LINE_COMMON)
2611                        postlen += tgtfix.len;
2612
2613                strbuf_release(&tgtfix);
2614                if (!match)
2615                        goto unmatch_exit;
2616
2617                orig += oldlen;
2618                target += tgtlen;
2619        }
2620
2621
2622        /*
2623         * Now handle the lines in the preimage that falls beyond the
2624         * end of the file (if any). They will only match if they are
2625         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2626         * false).
2627         */
2628        for ( ; i < preimage->nr; i++) {
2629                size_t fixstart = fixed.len; /* start of the fixed preimage */
2630                size_t oldlen = preimage->line[i].len;
2631                int j;
2632
2633                /* Try fixing the line in the preimage */
2634                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2635
2636                for (j = fixstart; j < fixed.len; j++)
2637                        if (!isspace(fixed.buf[j]))
2638                                goto unmatch_exit;
2639
2640                orig += oldlen;
2641        }
2642
2643        /*
2644         * Yes, the preimage is based on an older version that still
2645         * has whitespace breakages unfixed, and fixing them makes the
2646         * hunk match.  Update the context lines in the postimage.
2647         */
2648        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2649        if (postlen < postimage->len)
2650                postlen = 0;
2651        update_pre_post_images(preimage, postimage,
2652                               fixed_buf, fixed_len, postlen);
2653        return 1;
2654
2655 unmatch_exit:
2656        strbuf_release(&fixed);
2657        return 0;
2658}
2659
2660static int find_pos(struct apply_state *state,
2661                    struct image *img,
2662                    struct image *preimage,
2663                    struct image *postimage,
2664                    int line,
2665                    unsigned ws_rule,
2666                    int match_beginning, int match_end)
2667{
2668        int i;
2669        unsigned long backwards, forwards, try;
2670        int backwards_lno, forwards_lno, try_lno;
2671
2672        /*
2673         * If match_beginning or match_end is specified, there is no
2674         * point starting from a wrong line that will never match and
2675         * wander around and wait for a match at the specified end.
2676         */
2677        if (match_beginning)
2678                line = 0;
2679        else if (match_end)
2680                line = img->nr - preimage->nr;
2681
2682        /*
2683         * Because the comparison is unsigned, the following test
2684         * will also take care of a negative line number that can
2685         * result when match_end and preimage is larger than the target.
2686         */
2687        if ((size_t) line > img->nr)
2688                line = img->nr;
2689
2690        try = 0;
2691        for (i = 0; i < line; i++)
2692                try += img->line[i].len;
2693
2694        /*
2695         * There's probably some smart way to do this, but I'll leave
2696         * that to the smart and beautiful people. I'm simple and stupid.
2697         */
2698        backwards = try;
2699        backwards_lno = line;
2700        forwards = try;
2701        forwards_lno = line;
2702        try_lno = line;
2703
2704        for (i = 0; ; i++) {
2705                if (match_fragment(state, img, preimage, postimage,
2706                                   try, try_lno, ws_rule,
2707                                   match_beginning, match_end))
2708                        return try_lno;
2709
2710        again:
2711                if (backwards_lno == 0 && forwards_lno == img->nr)
2712                        break;
2713
2714                if (i & 1) {
2715                        if (backwards_lno == 0) {
2716                                i++;
2717                                goto again;
2718                        }
2719                        backwards_lno--;
2720                        backwards -= img->line[backwards_lno].len;
2721                        try = backwards;
2722                        try_lno = backwards_lno;
2723                } else {
2724                        if (forwards_lno == img->nr) {
2725                                i++;
2726                                goto again;
2727                        }
2728                        forwards += img->line[forwards_lno].len;
2729                        forwards_lno++;
2730                        try = forwards;
2731                        try_lno = forwards_lno;
2732                }
2733
2734        }
2735        return -1;
2736}
2737
2738static void remove_first_line(struct image *img)
2739{
2740        img->buf += img->line[0].len;
2741        img->len -= img->line[0].len;
2742        img->line++;
2743        img->nr--;
2744}
2745
2746static void remove_last_line(struct image *img)
2747{
2748        img->len -= img->line[--img->nr].len;
2749}
2750
2751/*
2752 * The change from "preimage" and "postimage" has been found to
2753 * apply at applied_pos (counts in line numbers) in "img".
2754 * Update "img" to remove "preimage" and replace it with "postimage".
2755 */
2756static void update_image(struct apply_state *state,
2757                         struct image *img,
2758                         int applied_pos,
2759                         struct image *preimage,
2760                         struct image *postimage)
2761{
2762        /*
2763         * remove the copy of preimage at offset in img
2764         * and replace it with postimage
2765         */
2766        int i, nr;
2767        size_t remove_count, insert_count, applied_at = 0;
2768        char *result;
2769        int preimage_limit;
2770
2771        /*
2772         * If we are removing blank lines at the end of img,
2773         * the preimage may extend beyond the end.
2774         * If that is the case, we must be careful only to
2775         * remove the part of the preimage that falls within
2776         * the boundaries of img. Initialize preimage_limit
2777         * to the number of lines in the preimage that falls
2778         * within the boundaries.
2779         */
2780        preimage_limit = preimage->nr;
2781        if (preimage_limit > img->nr - applied_pos)
2782                preimage_limit = img->nr - applied_pos;
2783
2784        for (i = 0; i < applied_pos; i++)
2785                applied_at += img->line[i].len;
2786
2787        remove_count = 0;
2788        for (i = 0; i < preimage_limit; i++)
2789                remove_count += img->line[applied_pos + i].len;
2790        insert_count = postimage->len;
2791
2792        /* Adjust the contents */
2793        result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2794        memcpy(result, img->buf, applied_at);
2795        memcpy(result + applied_at, postimage->buf, postimage->len);
2796        memcpy(result + applied_at + postimage->len,
2797               img->buf + (applied_at + remove_count),
2798               img->len - (applied_at + remove_count));
2799        free(img->buf);
2800        img->buf = result;
2801        img->len += insert_count - remove_count;
2802        result[img->len] = '\0';
2803
2804        /* Adjust the line table */
2805        nr = img->nr + postimage->nr - preimage_limit;
2806        if (preimage_limit < postimage->nr) {
2807                /*
2808                 * NOTE: this knows that we never call remove_first_line()
2809                 * on anything other than pre/post image.
2810                 */
2811                REALLOC_ARRAY(img->line, nr);
2812                img->line_allocated = img->line;
2813        }
2814        if (preimage_limit != postimage->nr)
2815                MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2816                           img->line + applied_pos + preimage_limit,
2817                           img->nr - (applied_pos + preimage_limit));
2818        COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2819        if (!state->allow_overlap)
2820                for (i = 0; i < postimage->nr; i++)
2821                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2822        img->nr = nr;
2823}
2824
2825/*
2826 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2827 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2828 * replace the part of "img" with "postimage" text.
2829 */
2830static int apply_one_fragment(struct apply_state *state,
2831                              struct image *img, struct fragment *frag,
2832                              int inaccurate_eof, unsigned ws_rule,
2833                              int nth_fragment)
2834{
2835        int match_beginning, match_end;
2836        const char *patch = frag->patch;
2837        int size = frag->size;
2838        char *old, *oldlines;
2839        struct strbuf newlines;
2840        int new_blank_lines_at_end = 0;
2841        int found_new_blank_lines_at_end = 0;
2842        int hunk_linenr = frag->linenr;
2843        unsigned long leading, trailing;
2844        int pos, applied_pos;
2845        struct image preimage;
2846        struct image postimage;
2847
2848        memset(&preimage, 0, sizeof(preimage));
2849        memset(&postimage, 0, sizeof(postimage));
2850        oldlines = xmalloc(size);
2851        strbuf_init(&newlines, size);
2852
2853        old = oldlines;
2854        while (size > 0) {
2855                char first;
2856                int len = linelen(patch, size);
2857                int plen;
2858                int added_blank_line = 0;
2859                int is_blank_context = 0;
2860                size_t start;
2861
2862                if (!len)
2863                        break;
2864
2865                /*
2866                 * "plen" is how much of the line we should use for
2867                 * the actual patch data. Normally we just remove the
2868                 * first character on the line, but if the line is
2869                 * followed by "\ No newline", then we also remove the
2870                 * last one (which is the newline, of course).
2871                 */
2872                plen = len - 1;
2873                if (len < size && patch[len] == '\\')
2874                        plen--;
2875                first = *patch;
2876                if (state->apply_in_reverse) {
2877                        if (first == '-')
2878                                first = '+';
2879                        else if (first == '+')
2880                                first = '-';
2881                }
2882
2883                switch (first) {
2884                case '\n':
2885                        /* Newer GNU diff, empty context line */
2886                        if (plen < 0)
2887                                /* ... followed by '\No newline'; nothing */
2888                                break;
2889                        *old++ = '\n';
2890                        strbuf_addch(&newlines, '\n');
2891                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2892                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2893                        is_blank_context = 1;
2894                        break;
2895                case ' ':
2896                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2897                            ws_blank_line(patch + 1, plen, ws_rule))
2898                                is_blank_context = 1;
2899                        /* fallthrough */
2900                case '-':
2901                        memcpy(old, patch + 1, plen);
2902                        add_line_info(&preimage, old, plen,
2903                                      (first == ' ' ? LINE_COMMON : 0));
2904                        old += plen;
2905                        if (first == '-')
2906                                break;
2907                        /* fallthrough */
2908                case '+':
2909                        /* --no-add does not add new lines */
2910                        if (first == '+' && state->no_add)
2911                                break;
2912
2913                        start = newlines.len;
2914                        if (first != '+' ||
2915                            !state->whitespace_error ||
2916                            state->ws_error_action != correct_ws_error) {
2917                                strbuf_add(&newlines, patch + 1, plen);
2918                        }
2919                        else {
2920                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2921                        }
2922                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2923                                      (first == '+' ? 0 : LINE_COMMON));
2924                        if (first == '+' &&
2925                            (ws_rule & WS_BLANK_AT_EOF) &&
2926                            ws_blank_line(patch + 1, plen, ws_rule))
2927                                added_blank_line = 1;
2928                        break;
2929                case '@': case '\\':
2930                        /* Ignore it, we already handled it */
2931                        break;
2932                default:
2933                        if (state->apply_verbosity > verbosity_normal)
2934                                error(_("invalid start of line: '%c'"), first);
2935                        applied_pos = -1;
2936                        goto out;
2937                }
2938                if (added_blank_line) {
2939                        if (!new_blank_lines_at_end)
2940                                found_new_blank_lines_at_end = hunk_linenr;
2941                        new_blank_lines_at_end++;
2942                }
2943                else if (is_blank_context)
2944                        ;
2945                else
2946                        new_blank_lines_at_end = 0;
2947                patch += len;
2948                size -= len;
2949                hunk_linenr++;
2950        }
2951        if (inaccurate_eof &&
2952            old > oldlines && old[-1] == '\n' &&
2953            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2954                old--;
2955                strbuf_setlen(&newlines, newlines.len - 1);
2956        }
2957
2958        leading = frag->leading;
2959        trailing = frag->trailing;
2960
2961        /*
2962         * A hunk to change lines at the beginning would begin with
2963         * @@ -1,L +N,M @@
2964         * but we need to be careful.  -U0 that inserts before the second
2965         * line also has this pattern.
2966         *
2967         * And a hunk to add to an empty file would begin with
2968         * @@ -0,0 +N,M @@
2969         *
2970         * In other words, a hunk that is (frag->oldpos <= 1) with or
2971         * without leading context must match at the beginning.
2972         */
2973        match_beginning = (!frag->oldpos ||
2974                           (frag->oldpos == 1 && !state->unidiff_zero));
2975
2976        /*
2977         * A hunk without trailing lines must match at the end.
2978         * However, we simply cannot tell if a hunk must match end
2979         * from the lack of trailing lines if the patch was generated
2980         * with unidiff without any context.
2981         */
2982        match_end = !state->unidiff_zero && !trailing;
2983
2984        pos = frag->newpos ? (frag->newpos - 1) : 0;
2985        preimage.buf = oldlines;
2986        preimage.len = old - oldlines;
2987        postimage.buf = newlines.buf;
2988        postimage.len = newlines.len;
2989        preimage.line = preimage.line_allocated;
2990        postimage.line = postimage.line_allocated;
2991
2992        for (;;) {
2993
2994                applied_pos = find_pos(state, img, &preimage, &postimage, pos,
2995                                       ws_rule, match_beginning, match_end);
2996
2997                if (applied_pos >= 0)
2998                        break;
2999
3000                /* Am I at my context limits? */
3001                if ((leading <= state->p_context) && (trailing <= state->p_context))
3002                        break;
3003                if (match_beginning || match_end) {
3004                        match_beginning = match_end = 0;
3005                        continue;
3006                }
3007
3008                /*
3009                 * Reduce the number of context lines; reduce both
3010                 * leading and trailing if they are equal otherwise
3011                 * just reduce the larger context.
3012                 */
3013                if (leading >= trailing) {
3014                        remove_first_line(&preimage);
3015                        remove_first_line(&postimage);
3016                        pos--;
3017                        leading--;
3018                }
3019                if (trailing > leading) {
3020                        remove_last_line(&preimage);
3021                        remove_last_line(&postimage);
3022                        trailing--;
3023                }
3024        }
3025
3026        if (applied_pos >= 0) {
3027                if (new_blank_lines_at_end &&
3028                    preimage.nr + applied_pos >= img->nr &&
3029                    (ws_rule & WS_BLANK_AT_EOF) &&
3030                    state->ws_error_action != nowarn_ws_error) {
3031                        record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3032                                        found_new_blank_lines_at_end);
3033                        if (state->ws_error_action == correct_ws_error) {
3034                                while (new_blank_lines_at_end--)
3035                                        remove_last_line(&postimage);
3036                        }
3037                        /*
3038                         * We would want to prevent write_out_results()
3039                         * from taking place in apply_patch() that follows
3040                         * the callchain led us here, which is:
3041                         * apply_patch->check_patch_list->check_patch->
3042                         * apply_data->apply_fragments->apply_one_fragment
3043                         */
3044                        if (state->ws_error_action == die_on_ws_error)
3045                                state->apply = 0;
3046                }
3047
3048                if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3049                        int offset = applied_pos - pos;
3050                        if (state->apply_in_reverse)
3051                                offset = 0 - offset;
3052                        fprintf_ln(stderr,
3053                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
3054                                      "Hunk #%d succeeded at %d (offset %d lines).",
3055                                      offset),
3056                                   nth_fragment, applied_pos + 1, offset);
3057                }
3058
3059                /*
3060                 * Warn if it was necessary to reduce the number
3061                 * of context lines.
3062                 */
3063                if ((leading != frag->leading ||
3064                     trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3065                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3066                                             " to apply fragment at %d"),
3067                                   leading, trailing, applied_pos+1);
3068                update_image(state, img, applied_pos, &preimage, &postimage);
3069        } else {
3070                if (state->apply_verbosity > verbosity_normal)
3071                        error(_("while searching for:\n%.*s"),
3072                              (int)(old - oldlines), oldlines);
3073        }
3074
3075out:
3076        free(oldlines);
3077        strbuf_release(&newlines);
3078        free(preimage.line_allocated);
3079        free(postimage.line_allocated);
3080
3081        return (applied_pos < 0);
3082}
3083
3084static int apply_binary_fragment(struct apply_state *state,
3085                                 struct image *img,
3086                                 struct patch *patch)
3087{
3088        struct fragment *fragment = patch->fragments;
3089        unsigned long len;
3090        void *dst;
3091
3092        if (!fragment)
3093                return error(_("missing binary patch data for '%s'"),
3094                             patch->new_name ?
3095                             patch->new_name :
3096                             patch->old_name);
3097
3098        /* Binary patch is irreversible without the optional second hunk */
3099        if (state->apply_in_reverse) {
3100                if (!fragment->next)
3101                        return error(_("cannot reverse-apply a binary patch "
3102                                       "without the reverse hunk to '%s'"),
3103                                     patch->new_name
3104                                     ? patch->new_name : patch->old_name);
3105                fragment = fragment->next;
3106        }
3107        switch (fragment->binary_patch_method) {
3108        case BINARY_DELTA_DEFLATED:
3109                dst = patch_delta(img->buf, img->len, fragment->patch,
3110                                  fragment->size, &len);
3111                if (!dst)
3112                        return -1;
3113                clear_image(img);
3114                img->buf = dst;
3115                img->len = len;
3116                return 0;
3117        case BINARY_LITERAL_DEFLATED:
3118                clear_image(img);
3119                img->len = fragment->size;
3120                img->buf = xmemdupz(fragment->patch, img->len);
3121                return 0;
3122        }
3123        return -1;
3124}
3125
3126/*
3127 * Replace "img" with the result of applying the binary patch.
3128 * The binary patch data itself in patch->fragment is still kept
3129 * but the preimage prepared by the caller in "img" is freed here
3130 * or in the helper function apply_binary_fragment() this calls.
3131 */
3132static int apply_binary(struct apply_state *state,
3133                        struct image *img,
3134                        struct patch *patch)
3135{
3136        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3137        struct object_id oid;
3138
3139        /*
3140         * For safety, we require patch index line to contain
3141         * full 40-byte textual SHA1 for old and new, at least for now.
3142         */
3143        if (strlen(patch->old_sha1_prefix) != 40 ||
3144            strlen(patch->new_sha1_prefix) != 40 ||
3145            get_oid_hex(patch->old_sha1_prefix, &oid) ||
3146            get_oid_hex(patch->new_sha1_prefix, &oid))
3147                return error(_("cannot apply binary patch to '%s' "
3148                               "without full index line"), name);
3149
3150        if (patch->old_name) {
3151                /*
3152                 * See if the old one matches what the patch
3153                 * applies to.
3154                 */
3155                hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3156                if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3157                        return error(_("the patch applies to '%s' (%s), "
3158                                       "which does not match the "
3159                                       "current contents."),
3160                                     name, oid_to_hex(&oid));
3161        }
3162        else {
3163                /* Otherwise, the old one must be empty. */
3164                if (img->len)
3165                        return error(_("the patch applies to an empty "
3166                                       "'%s' but it is not empty"), name);
3167        }
3168
3169        get_oid_hex(patch->new_sha1_prefix, &oid);
3170        if (is_null_oid(&oid)) {
3171                clear_image(img);
3172                return 0; /* deletion patch */
3173        }
3174
3175        if (has_sha1_file(oid.hash)) {
3176                /* We already have the postimage */
3177                enum object_type type;
3178                unsigned long size;
3179                char *result;
3180
3181                result = read_sha1_file(oid.hash, &type, &size);
3182                if (!result)
3183                        return error(_("the necessary postimage %s for "
3184                                       "'%s' cannot be read"),
3185                                     patch->new_sha1_prefix, name);
3186                clear_image(img);
3187                img->buf = result;
3188                img->len = size;
3189        } else {
3190                /*
3191                 * We have verified buf matches the preimage;
3192                 * apply the patch data to it, which is stored
3193                 * in the patch->fragments->{patch,size}.
3194                 */
3195                if (apply_binary_fragment(state, img, patch))
3196                        return error(_("binary patch does not apply to '%s'"),
3197                                     name);
3198
3199                /* verify that the result matches */
3200                hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3201                if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3202                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3203                                name, patch->new_sha1_prefix, oid_to_hex(&oid));
3204        }
3205
3206        return 0;
3207}
3208
3209static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3210{
3211        struct fragment *frag = patch->fragments;
3212        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3213        unsigned ws_rule = patch->ws_rule;
3214        unsigned inaccurate_eof = patch->inaccurate_eof;
3215        int nth = 0;
3216
3217        if (patch->is_binary)
3218                return apply_binary(state, img, patch);
3219
3220        while (frag) {
3221                nth++;
3222                if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3223                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
3224                        if (!state->apply_with_reject)
3225                                return -1;
3226                        frag->rejected = 1;
3227                }
3228                frag = frag->next;
3229        }
3230        return 0;
3231}
3232
3233static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3234{
3235        if (S_ISGITLINK(mode)) {
3236                strbuf_grow(buf, 100);
3237                strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3238        } else {
3239                enum object_type type;
3240                unsigned long sz;
3241                char *result;
3242
3243                result = read_sha1_file(oid->hash, &type, &sz);
3244                if (!result)
3245                        return -1;
3246                /* XXX read_sha1_file NUL-terminates */
3247                strbuf_attach(buf, result, sz, sz + 1);
3248        }
3249        return 0;
3250}
3251
3252static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3253{
3254        if (!ce)
3255                return 0;
3256        return read_blob_object(buf, &ce->oid, ce->ce_mode);
3257}
3258
3259static struct patch *in_fn_table(struct apply_state *state, const char *name)
3260{
3261        struct string_list_item *item;
3262
3263        if (name == NULL)
3264                return NULL;
3265
3266        item = string_list_lookup(&state->fn_table, name);
3267        if (item != NULL)
3268                return (struct patch *)item->util;
3269
3270        return NULL;
3271}
3272
3273/*
3274 * item->util in the filename table records the status of the path.
3275 * Usually it points at a patch (whose result records the contents
3276 * of it after applying it), but it could be PATH_WAS_DELETED for a
3277 * path that a previously applied patch has already removed, or
3278 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3279 *
3280 * The latter is needed to deal with a case where two paths A and B
3281 * are swapped by first renaming A to B and then renaming B to A;
3282 * moving A to B should not be prevented due to presence of B as we
3283 * will remove it in a later patch.
3284 */
3285#define PATH_TO_BE_DELETED ((struct patch *) -2)
3286#define PATH_WAS_DELETED ((struct patch *) -1)
3287
3288static int to_be_deleted(struct patch *patch)
3289{
3290        return patch == PATH_TO_BE_DELETED;
3291}
3292
3293static int was_deleted(struct patch *patch)
3294{
3295        return patch == PATH_WAS_DELETED;
3296}
3297
3298static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3299{
3300        struct string_list_item *item;
3301
3302        /*
3303         * Always add new_name unless patch is a deletion
3304         * This should cover the cases for normal diffs,
3305         * file creations and copies
3306         */
3307        if (patch->new_name != NULL) {
3308                item = string_list_insert(&state->fn_table, patch->new_name);
3309                item->util = patch;
3310        }
3311
3312        /*
3313         * store a failure on rename/deletion cases because
3314         * later chunks shouldn't patch old names
3315         */
3316        if ((patch->new_name == NULL) || (patch->is_rename)) {
3317                item = string_list_insert(&state->fn_table, patch->old_name);
3318                item->util = PATH_WAS_DELETED;
3319        }
3320}
3321
3322static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3323{
3324        /*
3325         * store information about incoming file deletion
3326         */
3327        while (patch) {
3328                if ((patch->new_name == NULL) || (patch->is_rename)) {
3329                        struct string_list_item *item;
3330                        item = string_list_insert(&state->fn_table, patch->old_name);
3331                        item->util = PATH_TO_BE_DELETED;
3332                }
3333                patch = patch->next;
3334        }
3335}
3336
3337static int checkout_target(struct index_state *istate,
3338                           struct cache_entry *ce, struct stat *st)
3339{
3340        struct checkout costate = CHECKOUT_INIT;
3341
3342        costate.refresh_cache = 1;
3343        costate.istate = istate;
3344        if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3345                return error(_("cannot checkout %s"), ce->name);
3346        return 0;
3347}
3348
3349static struct patch *previous_patch(struct apply_state *state,
3350                                    struct patch *patch,
3351                                    int *gone)
3352{
3353        struct patch *previous;
3354
3355        *gone = 0;
3356        if (patch->is_copy || patch->is_rename)
3357                return NULL; /* "git" patches do not depend on the order */
3358
3359        previous = in_fn_table(state, patch->old_name);
3360        if (!previous)
3361                return NULL;
3362
3363        if (to_be_deleted(previous))
3364                return NULL; /* the deletion hasn't happened yet */
3365
3366        if (was_deleted(previous))
3367                *gone = 1;
3368
3369        return previous;
3370}
3371
3372static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3373{
3374        if (S_ISGITLINK(ce->ce_mode)) {
3375                if (!S_ISDIR(st->st_mode))
3376                        return -1;
3377                return 0;
3378        }
3379        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3380}
3381
3382#define SUBMODULE_PATCH_WITHOUT_INDEX 1
3383
3384static int load_patch_target(struct apply_state *state,
3385                             struct strbuf *buf,
3386                             const struct cache_entry *ce,
3387                             struct stat *st,
3388                             struct patch *patch,
3389                             const char *name,
3390                             unsigned expected_mode)
3391{
3392        if (state->cached || state->check_index) {
3393                if (read_file_or_gitlink(ce, buf))
3394                        return error(_("failed to read %s"), name);
3395        } else if (name) {
3396                if (S_ISGITLINK(expected_mode)) {
3397                        if (ce)
3398                                return read_file_or_gitlink(ce, buf);
3399                        else
3400                                return SUBMODULE_PATCH_WITHOUT_INDEX;
3401                } else if (has_symlink_leading_path(name, strlen(name))) {
3402                        return error(_("reading from '%s' beyond a symbolic link"), name);
3403                } else {
3404                        if (read_old_data(st, patch, name, buf))
3405                                return error(_("failed to read %s"), name);
3406                }
3407        }
3408        return 0;
3409}
3410
3411/*
3412 * We are about to apply "patch"; populate the "image" with the
3413 * current version we have, from the working tree or from the index,
3414 * depending on the situation e.g. --cached/--index.  If we are
3415 * applying a non-git patch that incrementally updates the tree,
3416 * we read from the result of a previous diff.
3417 */
3418static int load_preimage(struct apply_state *state,
3419                         struct image *image,
3420                         struct patch *patch, struct stat *st,
3421                         const struct cache_entry *ce)
3422{
3423        struct strbuf buf = STRBUF_INIT;
3424        size_t len;
3425        char *img;
3426        struct patch *previous;
3427        int status;
3428
3429        previous = previous_patch(state, patch, &status);
3430        if (status)
3431                return error(_("path %s has been renamed/deleted"),
3432                             patch->old_name);
3433        if (previous) {
3434                /* We have a patched copy in memory; use that. */
3435                strbuf_add(&buf, previous->result, previous->resultsize);
3436        } else {
3437                status = load_patch_target(state, &buf, ce, st, patch,
3438                                           patch->old_name, patch->old_mode);
3439                if (status < 0)
3440                        return status;
3441                else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3442                        /*
3443                         * There is no way to apply subproject
3444                         * patch without looking at the index.
3445                         * NEEDSWORK: shouldn't this be flagged
3446                         * as an error???
3447                         */
3448                        free_fragment_list(patch->fragments);
3449                        patch->fragments = NULL;
3450                } else if (status) {
3451                        return error(_("failed to read %s"), patch->old_name);
3452                }
3453        }
3454
3455        img = strbuf_detach(&buf, &len);
3456        prepare_image(image, img, len, !patch->is_binary);
3457        return 0;
3458}
3459
3460static int three_way_merge(struct image *image,
3461                           char *path,
3462                           const struct object_id *base,
3463                           const struct object_id *ours,
3464                           const struct object_id *theirs)
3465{
3466        mmfile_t base_file, our_file, their_file;
3467        mmbuffer_t result = { NULL };
3468        int status;
3469
3470        read_mmblob(&base_file, base);
3471        read_mmblob(&our_file, ours);
3472        read_mmblob(&their_file, theirs);
3473        status = ll_merge(&result, path,
3474                          &base_file, "base",
3475                          &our_file, "ours",
3476                          &their_file, "theirs", NULL);
3477        free(base_file.ptr);
3478        free(our_file.ptr);
3479        free(their_file.ptr);
3480        if (status < 0 || !result.ptr) {
3481                free(result.ptr);
3482                return -1;
3483        }
3484        clear_image(image);
3485        image->buf = result.ptr;
3486        image->len = result.size;
3487
3488        return status;
3489}
3490
3491/*
3492 * When directly falling back to add/add three-way merge, we read from
3493 * the current contents of the new_name.  In no cases other than that
3494 * this function will be called.
3495 */
3496static int load_current(struct apply_state *state,
3497                        struct image *image,
3498                        struct patch *patch)
3499{
3500        struct strbuf buf = STRBUF_INIT;
3501        int status, pos;
3502        size_t len;
3503        char *img;
3504        struct stat st;
3505        struct cache_entry *ce;
3506        char *name = patch->new_name;
3507        unsigned mode = patch->new_mode;
3508
3509        if (!patch->is_new)
3510                die("BUG: patch to %s is not a creation", patch->old_name);
3511
3512        pos = cache_name_pos(name, strlen(name));
3513        if (pos < 0)
3514                return error(_("%s: does not exist in index"), name);
3515        ce = active_cache[pos];
3516        if (lstat(name, &st)) {
3517                if (errno != ENOENT)
3518                        return error_errno("%s", name);
3519                if (checkout_target(&the_index, ce, &st))
3520                        return -1;
3521        }
3522        if (verify_index_match(ce, &st))
3523                return error(_("%s: does not match index"), name);
3524
3525        status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3526        if (status < 0)
3527                return status;
3528        else if (status)
3529                return -1;
3530        img = strbuf_detach(&buf, &len);
3531        prepare_image(image, img, len, !patch->is_binary);
3532        return 0;
3533}
3534
3535static int try_threeway(struct apply_state *state,
3536                        struct image *image,
3537                        struct patch *patch,
3538                        struct stat *st,
3539                        const struct cache_entry *ce)
3540{
3541        struct object_id pre_oid, post_oid, our_oid;
3542        struct strbuf buf = STRBUF_INIT;
3543        size_t len;
3544        int status;
3545        char *img;
3546        struct image tmp_image;
3547
3548        /* No point falling back to 3-way merge in these cases */
3549        if (patch->is_delete ||
3550            S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3551                return -1;
3552
3553        /* Preimage the patch was prepared for */
3554        if (patch->is_new)
3555                write_sha1_file("", 0, blob_type, pre_oid.hash);
3556        else if (get_oid(patch->old_sha1_prefix, &pre_oid) ||
3557                 read_blob_object(&buf, &pre_oid, patch->old_mode))
3558                return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3559
3560        if (state->apply_verbosity > verbosity_silent)
3561                fprintf(stderr, _("Falling back to three-way merge...\n"));
3562
3563        img = strbuf_detach(&buf, &len);
3564        prepare_image(&tmp_image, img, len, 1);
3565        /* Apply the patch to get the post image */
3566        if (apply_fragments(state, &tmp_image, patch) < 0) {
3567                clear_image(&tmp_image);
3568                return -1;
3569        }
3570        /* post_oid is theirs */
3571        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3572        clear_image(&tmp_image);
3573
3574        /* our_oid is ours */
3575        if (patch->is_new) {
3576                if (load_current(state, &tmp_image, patch))
3577                        return error(_("cannot read the current contents of '%s'"),
3578                                     patch->new_name);
3579        } else {
3580                if (load_preimage(state, &tmp_image, patch, st, ce))
3581                        return error(_("cannot read the current contents of '%s'"),
3582                                     patch->old_name);
3583        }
3584        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3585        clear_image(&tmp_image);
3586
3587        /* in-core three-way merge between post and our using pre as base */
3588        status = three_way_merge(image, patch->new_name,
3589                                 &pre_oid, &our_oid, &post_oid);
3590        if (status < 0) {
3591                if (state->apply_verbosity > verbosity_silent)
3592                        fprintf(stderr,
3593                                _("Failed to fall back on three-way merge...\n"));
3594                return status;
3595        }
3596
3597        if (status) {
3598                patch->conflicted_threeway = 1;
3599                if (patch->is_new)
3600                        oidclr(&patch->threeway_stage[0]);
3601                else
3602                        oidcpy(&patch->threeway_stage[0], &pre_oid);
3603                oidcpy(&patch->threeway_stage[1], &our_oid);
3604                oidcpy(&patch->threeway_stage[2], &post_oid);
3605                if (state->apply_verbosity > verbosity_silent)
3606                        fprintf(stderr,
3607                                _("Applied patch to '%s' with conflicts.\n"),
3608                                patch->new_name);
3609        } else {
3610                if (state->apply_verbosity > verbosity_silent)
3611                        fprintf(stderr,
3612                                _("Applied patch to '%s' cleanly.\n"),
3613                                patch->new_name);
3614        }
3615        return 0;
3616}
3617
3618static int apply_data(struct apply_state *state, struct patch *patch,
3619                      struct stat *st, const struct cache_entry *ce)
3620{
3621        struct image image;
3622
3623        if (load_preimage(state, &image, patch, st, ce) < 0)
3624                return -1;
3625
3626        if (patch->direct_to_threeway ||
3627            apply_fragments(state, &image, patch) < 0) {
3628                /* Note: with --reject, apply_fragments() returns 0 */
3629                if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3630                        return -1;
3631        }
3632        patch->result = image.buf;
3633        patch->resultsize = image.len;
3634        add_to_fn_table(state, patch);
3635        free(image.line_allocated);
3636
3637        if (0 < patch->is_delete && patch->resultsize)
3638                return error(_("removal patch leaves file contents"));
3639
3640        return 0;
3641}
3642
3643/*
3644 * If "patch" that we are looking at modifies or deletes what we have,
3645 * we would want it not to lose any local modification we have, either
3646 * in the working tree or in the index.
3647 *
3648 * This also decides if a non-git patch is a creation patch or a
3649 * modification to an existing empty file.  We do not check the state
3650 * of the current tree for a creation patch in this function; the caller
3651 * check_patch() separately makes sure (and errors out otherwise) that
3652 * the path the patch creates does not exist in the current tree.
3653 */
3654static int check_preimage(struct apply_state *state,
3655                          struct patch *patch,
3656                          struct cache_entry **ce,
3657                          struct stat *st)
3658{
3659        const char *old_name = patch->old_name;
3660        struct patch *previous = NULL;
3661        int stat_ret = 0, status;
3662        unsigned st_mode = 0;
3663
3664        if (!old_name)
3665                return 0;
3666
3667        assert(patch->is_new <= 0);
3668        previous = previous_patch(state, patch, &status);
3669
3670        if (status)
3671                return error(_("path %s has been renamed/deleted"), old_name);
3672        if (previous) {
3673                st_mode = previous->new_mode;
3674        } else if (!state->cached) {
3675                stat_ret = lstat(old_name, st);
3676                if (stat_ret && errno != ENOENT)
3677                        return error_errno("%s", old_name);
3678        }
3679
3680        if (state->check_index && !previous) {
3681                int pos = cache_name_pos(old_name, strlen(old_name));
3682                if (pos < 0) {
3683                        if (patch->is_new < 0)
3684                                goto is_new;
3685                        return error(_("%s: does not exist in index"), old_name);
3686                }
3687                *ce = active_cache[pos];
3688                if (stat_ret < 0) {
3689                        if (checkout_target(&the_index, *ce, st))
3690                                return -1;
3691                }
3692                if (!state->cached && verify_index_match(*ce, st))
3693                        return error(_("%s: does not match index"), old_name);
3694                if (state->cached)
3695                        st_mode = (*ce)->ce_mode;
3696        } else if (stat_ret < 0) {
3697                if (patch->is_new < 0)
3698                        goto is_new;
3699                return error_errno("%s", old_name);
3700        }
3701
3702        if (!state->cached && !previous)
3703                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3704
3705        if (patch->is_new < 0)
3706                patch->is_new = 0;
3707        if (!patch->old_mode)
3708                patch->old_mode = st_mode;
3709        if ((st_mode ^ patch->old_mode) & S_IFMT)
3710                return error(_("%s: wrong type"), old_name);
3711        if (st_mode != patch->old_mode)
3712                warning(_("%s has type %o, expected %o"),
3713                        old_name, st_mode, patch->old_mode);
3714        if (!patch->new_mode && !patch->is_delete)
3715                patch->new_mode = st_mode;
3716        return 0;
3717
3718 is_new:
3719        patch->is_new = 1;
3720        patch->is_delete = 0;
3721        FREE_AND_NULL(patch->old_name);
3722        return 0;
3723}
3724
3725
3726#define EXISTS_IN_INDEX 1
3727#define EXISTS_IN_WORKTREE 2
3728
3729static int check_to_create(struct apply_state *state,
3730                           const char *new_name,
3731                           int ok_if_exists)
3732{
3733        struct stat nst;
3734
3735        if (state->check_index &&
3736            cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3737            !ok_if_exists)
3738                return EXISTS_IN_INDEX;
3739        if (state->cached)
3740                return 0;
3741
3742        if (!lstat(new_name, &nst)) {
3743                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3744                        return 0;
3745                /*
3746                 * A leading component of new_name might be a symlink
3747                 * that is going to be removed with this patch, but
3748                 * still pointing at somewhere that has the path.
3749                 * In such a case, path "new_name" does not exist as
3750                 * far as git is concerned.
3751                 */
3752                if (has_symlink_leading_path(new_name, strlen(new_name)))
3753                        return 0;
3754
3755                return EXISTS_IN_WORKTREE;
3756        } else if (!is_missing_file_error(errno)) {
3757                return error_errno("%s", new_name);
3758        }
3759        return 0;
3760}
3761
3762static uintptr_t register_symlink_changes(struct apply_state *state,
3763                                          const char *path,
3764                                          uintptr_t what)
3765{
3766        struct string_list_item *ent;
3767
3768        ent = string_list_lookup(&state->symlink_changes, path);
3769        if (!ent) {
3770                ent = string_list_insert(&state->symlink_changes, path);
3771                ent->util = (void *)0;
3772        }
3773        ent->util = (void *)(what | ((uintptr_t)ent->util));
3774        return (uintptr_t)ent->util;
3775}
3776
3777static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3778{
3779        struct string_list_item *ent;
3780
3781        ent = string_list_lookup(&state->symlink_changes, path);
3782        if (!ent)
3783                return 0;
3784        return (uintptr_t)ent->util;
3785}
3786
3787static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3788{
3789        for ( ; patch; patch = patch->next) {
3790                if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3791                    (patch->is_rename || patch->is_delete))
3792                        /* the symlink at patch->old_name is removed */
3793                        register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3794
3795                if (patch->new_name && S_ISLNK(patch->new_mode))
3796                        /* the symlink at patch->new_name is created or remains */
3797                        register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3798        }
3799}
3800
3801static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3802{
3803        do {
3804                unsigned int change;
3805
3806                while (--name->len && name->buf[name->len] != '/')
3807                        ; /* scan backwards */
3808                if (!name->len)
3809                        break;
3810                name->buf[name->len] = '\0';
3811                change = check_symlink_changes(state, name->buf);
3812                if (change & APPLY_SYMLINK_IN_RESULT)
3813                        return 1;
3814                if (change & APPLY_SYMLINK_GOES_AWAY)
3815                        /*
3816                         * This cannot be "return 0", because we may
3817                         * see a new one created at a higher level.
3818                         */
3819                        continue;
3820
3821                /* otherwise, check the preimage */
3822                if (state->check_index) {
3823                        struct cache_entry *ce;
3824
3825                        ce = cache_file_exists(name->buf, name->len, ignore_case);
3826                        if (ce && S_ISLNK(ce->ce_mode))
3827                                return 1;
3828                } else {
3829                        struct stat st;
3830                        if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3831                                return 1;
3832                }
3833        } while (1);
3834        return 0;
3835}
3836
3837static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3838{
3839        int ret;
3840        struct strbuf name = STRBUF_INIT;
3841
3842        assert(*name_ != '\0');
3843        strbuf_addstr(&name, name_);
3844        ret = path_is_beyond_symlink_1(state, &name);
3845        strbuf_release(&name);
3846
3847        return ret;
3848}
3849
3850static int check_unsafe_path(struct patch *patch)
3851{
3852        const char *old_name = NULL;
3853        const char *new_name = NULL;
3854        if (patch->is_delete)
3855                old_name = patch->old_name;
3856        else if (!patch->is_new && !patch->is_copy)
3857                old_name = patch->old_name;
3858        if (!patch->is_delete)
3859                new_name = patch->new_name;
3860
3861        if (old_name && !verify_path(old_name))
3862                return error(_("invalid path '%s'"), old_name);
3863        if (new_name && !verify_path(new_name))
3864                return error(_("invalid path '%s'"), new_name);
3865        return 0;
3866}
3867
3868/*
3869 * Check and apply the patch in-core; leave the result in patch->result
3870 * for the caller to write it out to the final destination.
3871 */
3872static int check_patch(struct apply_state *state, struct patch *patch)
3873{
3874        struct stat st;
3875        const char *old_name = patch->old_name;
3876        const char *new_name = patch->new_name;
3877        const char *name = old_name ? old_name : new_name;
3878        struct cache_entry *ce = NULL;
3879        struct patch *tpatch;
3880        int ok_if_exists;
3881        int status;
3882
3883        patch->rejected = 1; /* we will drop this after we succeed */
3884
3885        status = check_preimage(state, patch, &ce, &st);
3886        if (status)
3887                return status;
3888        old_name = patch->old_name;
3889
3890        /*
3891         * A type-change diff is always split into a patch to delete
3892         * old, immediately followed by a patch to create new (see
3893         * diff.c::run_diff()); in such a case it is Ok that the entry
3894         * to be deleted by the previous patch is still in the working
3895         * tree and in the index.
3896         *
3897         * A patch to swap-rename between A and B would first rename A
3898         * to B and then rename B to A.  While applying the first one,
3899         * the presence of B should not stop A from getting renamed to
3900         * B; ask to_be_deleted() about the later rename.  Removal of
3901         * B and rename from A to B is handled the same way by asking
3902         * was_deleted().
3903         */
3904        if ((tpatch = in_fn_table(state, new_name)) &&
3905            (was_deleted(tpatch) || to_be_deleted(tpatch)))
3906                ok_if_exists = 1;
3907        else
3908                ok_if_exists = 0;
3909
3910        if (new_name &&
3911            ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3912                int err = check_to_create(state, new_name, ok_if_exists);
3913
3914                if (err && state->threeway) {
3915                        patch->direct_to_threeway = 1;
3916                } else switch (err) {
3917                case 0:
3918                        break; /* happy */
3919                case EXISTS_IN_INDEX:
3920                        return error(_("%s: already exists in index"), new_name);
3921                        break;
3922                case EXISTS_IN_WORKTREE:
3923                        return error(_("%s: already exists in working directory"),
3924                                     new_name);
3925                default:
3926                        return err;
3927                }
3928
3929                if (!patch->new_mode) {
3930                        if (0 < patch->is_new)
3931                                patch->new_mode = S_IFREG | 0644;
3932                        else
3933                                patch->new_mode = patch->old_mode;
3934                }
3935        }
3936
3937        if (new_name && old_name) {
3938                int same = !strcmp(old_name, new_name);
3939                if (!patch->new_mode)
3940                        patch->new_mode = patch->old_mode;
3941                if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3942                        if (same)
3943                                return error(_("new mode (%o) of %s does not "
3944                                               "match old mode (%o)"),
3945                                        patch->new_mode, new_name,
3946                                        patch->old_mode);
3947                        else
3948                                return error(_("new mode (%o) of %s does not "
3949                                               "match old mode (%o) of %s"),
3950                                        patch->new_mode, new_name,
3951                                        patch->old_mode, old_name);
3952                }
3953        }
3954
3955        if (!state->unsafe_paths && check_unsafe_path(patch))
3956                return -128;
3957
3958        /*
3959         * An attempt to read from or delete a path that is beyond a
3960         * symbolic link will be prevented by load_patch_target() that
3961         * is called at the beginning of apply_data() so we do not
3962         * have to worry about a patch marked with "is_delete" bit
3963         * here.  We however need to make sure that the patch result
3964         * is not deposited to a path that is beyond a symbolic link
3965         * here.
3966         */
3967        if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3968                return error(_("affected file '%s' is beyond a symbolic link"),
3969                             patch->new_name);
3970
3971        if (apply_data(state, patch, &st, ce) < 0)
3972                return error(_("%s: patch does not apply"), name);
3973        patch->rejected = 0;
3974        return 0;
3975}
3976
3977static int check_patch_list(struct apply_state *state, struct patch *patch)
3978{
3979        int err = 0;
3980
3981        prepare_symlink_changes(state, patch);
3982        prepare_fn_table(state, patch);
3983        while (patch) {
3984                int res;
3985                if (state->apply_verbosity > verbosity_normal)
3986                        say_patch_name(stderr,
3987                                       _("Checking patch %s..."), patch);
3988                res = check_patch(state, patch);
3989                if (res == -128)
3990                        return -128;
3991                err |= res;
3992                patch = patch->next;
3993        }
3994        return err;
3995}
3996
3997static int read_apply_cache(struct apply_state *state)
3998{
3999        if (state->index_file)
4000                return read_cache_from(state->index_file);
4001        else
4002                return read_cache();
4003}
4004
4005/* This function tries to read the object name from the current index */
4006static int get_current_oid(struct apply_state *state, const char *path,
4007                           struct object_id *oid)
4008{
4009        int pos;
4010
4011        if (read_apply_cache(state) < 0)
4012                return -1;
4013        pos = cache_name_pos(path, strlen(path));
4014        if (pos < 0)
4015                return -1;
4016        oidcpy(oid, &active_cache[pos]->oid);
4017        return 0;
4018}
4019
4020static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4021{
4022        /*
4023         * A usable gitlink patch has only one fragment (hunk) that looks like:
4024         * @@ -1 +1 @@
4025         * -Subproject commit <old sha1>
4026         * +Subproject commit <new sha1>
4027         * or
4028         * @@ -1 +0,0 @@
4029         * -Subproject commit <old sha1>
4030         * for a removal patch.
4031         */
4032        struct fragment *hunk = p->fragments;
4033        static const char heading[] = "-Subproject commit ";
4034        char *preimage;
4035
4036        if (/* does the patch have only one hunk? */
4037            hunk && !hunk->next &&
4038            /* is its preimage one line? */
4039            hunk->oldpos == 1 && hunk->oldlines == 1 &&
4040            /* does preimage begin with the heading? */
4041            (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4042            starts_with(++preimage, heading) &&
4043            /* does it record full SHA-1? */
4044            !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4045            preimage[sizeof(heading) + GIT_SHA1_HEXSZ - 1] == '\n' &&
4046            /* does the abbreviated name on the index line agree with it? */
4047            starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
4048                return 0; /* it all looks fine */
4049
4050        /* we may have full object name on the index line */
4051        return get_oid_hex(p->old_sha1_prefix, oid);
4052}
4053
4054/* Build an index that contains the just the files needed for a 3way merge */
4055static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4056{
4057        struct patch *patch;
4058        struct index_state result = { NULL };
4059        static struct lock_file lock;
4060        int res;
4061
4062        /* Once we start supporting the reverse patch, it may be
4063         * worth showing the new sha1 prefix, but until then...
4064         */
4065        for (patch = list; patch; patch = patch->next) {
4066                struct object_id oid;
4067                struct cache_entry *ce;
4068                const char *name;
4069
4070                name = patch->old_name ? patch->old_name : patch->new_name;
4071                if (0 < patch->is_new)
4072                        continue;
4073
4074                if (S_ISGITLINK(patch->old_mode)) {
4075                        if (!preimage_oid_in_gitlink_patch(patch, &oid))
4076                                ; /* ok, the textual part looks sane */
4077                        else
4078                                return error(_("sha1 information is lacking or "
4079                                               "useless for submodule %s"), name);
4080                } else if (!get_oid_blob(patch->old_sha1_prefix, &oid)) {
4081                        ; /* ok */
4082                } else if (!patch->lines_added && !patch->lines_deleted) {
4083                        /* mode-only change: update the current */
4084                        if (get_current_oid(state, patch->old_name, &oid))
4085                                return error(_("mode change for %s, which is not "
4086                                               "in current HEAD"), name);
4087                } else
4088                        return error(_("sha1 information is lacking or useless "
4089                                       "(%s)."), name);
4090
4091                ce = make_cache_entry(patch->old_mode, oid.hash, name, 0, 0);
4092                if (!ce)
4093                        return error(_("make_cache_entry failed for path '%s'"),
4094                                     name);
4095                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4096                        free(ce);
4097                        return error(_("could not add %s to temporary index"),
4098                                     name);
4099                }
4100        }
4101
4102        hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4103        res = write_locked_index(&result, &lock, COMMIT_LOCK);
4104        discard_index(&result);
4105
4106        if (res)
4107                return error(_("could not write temporary index to %s"),
4108                             state->fake_ancestor);
4109
4110        return 0;
4111}
4112
4113static void stat_patch_list(struct apply_state *state, struct patch *patch)
4114{
4115        int files, adds, dels;
4116
4117        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4118                files++;
4119                adds += patch->lines_added;
4120                dels += patch->lines_deleted;
4121                show_stats(state, patch);
4122        }
4123
4124        print_stat_summary(stdout, files, adds, dels);
4125}
4126
4127static void numstat_patch_list(struct apply_state *state,
4128                               struct patch *patch)
4129{
4130        for ( ; patch; patch = patch->next) {
4131                const char *name;
4132                name = patch->new_name ? patch->new_name : patch->old_name;
4133                if (patch->is_binary)
4134                        printf("-\t-\t");
4135                else
4136                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4137                write_name_quoted(name, stdout, state->line_termination);
4138        }
4139}
4140
4141static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4142{
4143        if (mode)
4144                printf(" %s mode %06o %s\n", newdelete, mode, name);
4145        else
4146                printf(" %s %s\n", newdelete, name);
4147}
4148
4149static void show_mode_change(struct patch *p, int show_name)
4150{
4151        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4152                if (show_name)
4153                        printf(" mode change %06o => %06o %s\n",
4154                               p->old_mode, p->new_mode, p->new_name);
4155                else
4156                        printf(" mode change %06o => %06o\n",
4157                               p->old_mode, p->new_mode);
4158        }
4159}
4160
4161static void show_rename_copy(struct patch *p)
4162{
4163        const char *renamecopy = p->is_rename ? "rename" : "copy";
4164        const char *old, *new;
4165
4166        /* Find common prefix */
4167        old = p->old_name;
4168        new = p->new_name;
4169        while (1) {
4170                const char *slash_old, *slash_new;
4171                slash_old = strchr(old, '/');
4172                slash_new = strchr(new, '/');
4173                if (!slash_old ||
4174                    !slash_new ||
4175                    slash_old - old != slash_new - new ||
4176                    memcmp(old, new, slash_new - new))
4177                        break;
4178                old = slash_old + 1;
4179                new = slash_new + 1;
4180        }
4181        /* p->old_name thru old is the common prefix, and old and new
4182         * through the end of names are renames
4183         */
4184        if (old != p->old_name)
4185                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4186                       (int)(old - p->old_name), p->old_name,
4187                       old, new, p->score);
4188        else
4189                printf(" %s %s => %s (%d%%)\n", renamecopy,
4190                       p->old_name, p->new_name, p->score);
4191        show_mode_change(p, 0);
4192}
4193
4194static void summary_patch_list(struct patch *patch)
4195{
4196        struct patch *p;
4197
4198        for (p = patch; p; p = p->next) {
4199                if (p->is_new)
4200                        show_file_mode_name("create", p->new_mode, p->new_name);
4201                else if (p->is_delete)
4202                        show_file_mode_name("delete", p->old_mode, p->old_name);
4203                else {
4204                        if (p->is_rename || p->is_copy)
4205                                show_rename_copy(p);
4206                        else {
4207                                if (p->score) {
4208                                        printf(" rewrite %s (%d%%)\n",
4209                                               p->new_name, p->score);
4210                                        show_mode_change(p, 0);
4211                                }
4212                                else
4213                                        show_mode_change(p, 1);
4214                        }
4215                }
4216        }
4217}
4218
4219static void patch_stats(struct apply_state *state, struct patch *patch)
4220{
4221        int lines = patch->lines_added + patch->lines_deleted;
4222
4223        if (lines > state->max_change)
4224                state->max_change = lines;
4225        if (patch->old_name) {
4226                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4227                if (!len)
4228                        len = strlen(patch->old_name);
4229                if (len > state->max_len)
4230                        state->max_len = len;
4231        }
4232        if (patch->new_name) {
4233                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4234                if (!len)
4235                        len = strlen(patch->new_name);
4236                if (len > state->max_len)
4237                        state->max_len = len;
4238        }
4239}
4240
4241static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4242{
4243        if (state->update_index) {
4244                if (remove_file_from_cache(patch->old_name) < 0)
4245                        return error(_("unable to remove %s from index"), patch->old_name);
4246        }
4247        if (!state->cached) {
4248                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4249                        remove_path(patch->old_name);
4250                }
4251        }
4252        return 0;
4253}
4254
4255static int add_index_file(struct apply_state *state,
4256                          const char *path,
4257                          unsigned mode,
4258                          void *buf,
4259                          unsigned long size)
4260{
4261        struct stat st;
4262        struct cache_entry *ce;
4263        int namelen = strlen(path);
4264        unsigned ce_size = cache_entry_size(namelen);
4265
4266        if (!state->update_index)
4267                return 0;
4268
4269        ce = xcalloc(1, ce_size);
4270        memcpy(ce->name, path, namelen);
4271        ce->ce_mode = create_ce_mode(mode);
4272        ce->ce_flags = create_ce_flags(0);
4273        ce->ce_namelen = namelen;
4274        if (S_ISGITLINK(mode)) {
4275                const char *s;
4276
4277                if (!skip_prefix(buf, "Subproject commit ", &s) ||
4278                    get_oid_hex(s, &ce->oid)) {
4279                        free(ce);
4280                       return error(_("corrupt patch for submodule %s"), path);
4281                }
4282        } else {
4283                if (!state->cached) {
4284                        if (lstat(path, &st) < 0) {
4285                                free(ce);
4286                                return error_errno(_("unable to stat newly "
4287                                                     "created file '%s'"),
4288                                                   path);
4289                        }
4290                        fill_stat_cache_info(ce, &st);
4291                }
4292                if (write_sha1_file(buf, size, blob_type, ce->oid.hash) < 0) {
4293                        free(ce);
4294                        return error(_("unable to create backing store "
4295                                       "for newly created file %s"), path);
4296                }
4297        }
4298        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4299                free(ce);
4300                return error(_("unable to add cache entry for %s"), path);
4301        }
4302
4303        return 0;
4304}
4305
4306/*
4307 * Returns:
4308 *  -1 if an unrecoverable error happened
4309 *   0 if everything went well
4310 *   1 if a recoverable error happened
4311 */
4312static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4313{
4314        int fd, res;
4315        struct strbuf nbuf = STRBUF_INIT;
4316
4317        if (S_ISGITLINK(mode)) {
4318                struct stat st;
4319                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4320                        return 0;
4321                return !!mkdir(path, 0777);
4322        }
4323
4324        if (has_symlinks && S_ISLNK(mode))
4325                /* Although buf:size is counted string, it also is NUL
4326                 * terminated.
4327                 */
4328                return !!symlink(buf, path);
4329
4330        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4331        if (fd < 0)
4332                return 1;
4333
4334        if (convert_to_working_tree(path, buf, size, &nbuf)) {
4335                size = nbuf.len;
4336                buf  = nbuf.buf;
4337        }
4338
4339        res = write_in_full(fd, buf, size) < 0;
4340        if (res)
4341                error_errno(_("failed to write to '%s'"), path);
4342        strbuf_release(&nbuf);
4343
4344        if (close(fd) < 0 && !res)
4345                return error_errno(_("closing file '%s'"), path);
4346
4347        return res ? -1 : 0;
4348}
4349
4350/*
4351 * We optimistically assume that the directories exist,
4352 * which is true 99% of the time anyway. If they don't,
4353 * we create them and try again.
4354 *
4355 * Returns:
4356 *   -1 on error
4357 *   0 otherwise
4358 */
4359static int create_one_file(struct apply_state *state,
4360                           char *path,
4361                           unsigned mode,
4362                           const char *buf,
4363                           unsigned long size)
4364{
4365        int res;
4366
4367        if (state->cached)
4368                return 0;
4369
4370        res = try_create_file(path, mode, buf, size);
4371        if (res < 0)
4372                return -1;
4373        if (!res)
4374                return 0;
4375
4376        if (errno == ENOENT) {
4377                if (safe_create_leading_directories(path))
4378                        return 0;
4379                res = try_create_file(path, mode, buf, size);
4380                if (res < 0)
4381                        return -1;
4382                if (!res)
4383                        return 0;
4384        }
4385
4386        if (errno == EEXIST || errno == EACCES) {
4387                /* We may be trying to create a file where a directory
4388                 * used to be.
4389                 */
4390                struct stat st;
4391                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4392                        errno = EEXIST;
4393        }
4394
4395        if (errno == EEXIST) {
4396                unsigned int nr = getpid();
4397
4398                for (;;) {
4399                        char newpath[PATH_MAX];
4400                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4401                        res = try_create_file(newpath, mode, buf, size);
4402                        if (res < 0)
4403                                return -1;
4404                        if (!res) {
4405                                if (!rename(newpath, path))
4406                                        return 0;
4407                                unlink_or_warn(newpath);
4408                                break;
4409                        }
4410                        if (errno != EEXIST)
4411                                break;
4412                        ++nr;
4413                }
4414        }
4415        return error_errno(_("unable to write file '%s' mode %o"),
4416                           path, mode);
4417}
4418
4419static int add_conflicted_stages_file(struct apply_state *state,
4420                                       struct patch *patch)
4421{
4422        int stage, namelen;
4423        unsigned ce_size, mode;
4424        struct cache_entry *ce;
4425
4426        if (!state->update_index)
4427                return 0;
4428        namelen = strlen(patch->new_name);
4429        ce_size = cache_entry_size(namelen);
4430        mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4431
4432        remove_file_from_cache(patch->new_name);
4433        for (stage = 1; stage < 4; stage++) {
4434                if (is_null_oid(&patch->threeway_stage[stage - 1]))
4435                        continue;
4436                ce = xcalloc(1, ce_size);
4437                memcpy(ce->name, patch->new_name, namelen);
4438                ce->ce_mode = create_ce_mode(mode);
4439                ce->ce_flags = create_ce_flags(stage);
4440                ce->ce_namelen = namelen;
4441                oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4442                if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4443                        free(ce);
4444                        return error(_("unable to add cache entry for %s"),
4445                                     patch->new_name);
4446                }
4447        }
4448
4449        return 0;
4450}
4451
4452static int create_file(struct apply_state *state, struct patch *patch)
4453{
4454        char *path = patch->new_name;
4455        unsigned mode = patch->new_mode;
4456        unsigned long size = patch->resultsize;
4457        char *buf = patch->result;
4458
4459        if (!mode)
4460                mode = S_IFREG | 0644;
4461        if (create_one_file(state, path, mode, buf, size))
4462                return -1;
4463
4464        if (patch->conflicted_threeway)
4465                return add_conflicted_stages_file(state, patch);
4466        else
4467                return add_index_file(state, path, mode, buf, size);
4468}
4469
4470/* phase zero is to remove, phase one is to create */
4471static int write_out_one_result(struct apply_state *state,
4472                                struct patch *patch,
4473                                int phase)
4474{
4475        if (patch->is_delete > 0) {
4476                if (phase == 0)
4477                        return remove_file(state, patch, 1);
4478                return 0;
4479        }
4480        if (patch->is_new > 0 || patch->is_copy) {
4481                if (phase == 1)
4482                        return create_file(state, patch);
4483                return 0;
4484        }
4485        /*
4486         * Rename or modification boils down to the same
4487         * thing: remove the old, write the new
4488         */
4489        if (phase == 0)
4490                return remove_file(state, patch, patch->is_rename);
4491        if (phase == 1)
4492                return create_file(state, patch);
4493        return 0;
4494}
4495
4496static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4497{
4498        FILE *rej;
4499        char namebuf[PATH_MAX];
4500        struct fragment *frag;
4501        int cnt = 0;
4502        struct strbuf sb = STRBUF_INIT;
4503
4504        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4505                if (!frag->rejected)
4506                        continue;
4507                cnt++;
4508        }
4509
4510        if (!cnt) {
4511                if (state->apply_verbosity > verbosity_normal)
4512                        say_patch_name(stderr,
4513                                       _("Applied patch %s cleanly."), patch);
4514                return 0;
4515        }
4516
4517        /* This should not happen, because a removal patch that leaves
4518         * contents are marked "rejected" at the patch level.
4519         */
4520        if (!patch->new_name)
4521                die(_("internal error"));
4522
4523        /* Say this even without --verbose */
4524        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4525                            "Applying patch %%s with %d rejects...",
4526                            cnt),
4527                    cnt);
4528        if (state->apply_verbosity > verbosity_silent)
4529                say_patch_name(stderr, sb.buf, patch);
4530        strbuf_release(&sb);
4531
4532        cnt = strlen(patch->new_name);
4533        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4534                cnt = ARRAY_SIZE(namebuf) - 5;
4535                warning(_("truncating .rej filename to %.*s.rej"),
4536                        cnt - 1, patch->new_name);
4537        }
4538        memcpy(namebuf, patch->new_name, cnt);
4539        memcpy(namebuf + cnt, ".rej", 5);
4540
4541        rej = fopen(namebuf, "w");
4542        if (!rej)
4543                return error_errno(_("cannot open %s"), namebuf);
4544
4545        /* Normal git tools never deal with .rej, so do not pretend
4546         * this is a git patch by saying --git or giving extended
4547         * headers.  While at it, maybe please "kompare" that wants
4548         * the trailing TAB and some garbage at the end of line ;-).
4549         */
4550        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4551                patch->new_name, patch->new_name);
4552        for (cnt = 1, frag = patch->fragments;
4553             frag;
4554             cnt++, frag = frag->next) {
4555                if (!frag->rejected) {
4556                        if (state->apply_verbosity > verbosity_silent)
4557                                fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4558                        continue;
4559                }
4560                if (state->apply_verbosity > verbosity_silent)
4561                        fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4562                fprintf(rej, "%.*s", frag->size, frag->patch);
4563                if (frag->patch[frag->size-1] != '\n')
4564                        fputc('\n', rej);
4565        }
4566        fclose(rej);
4567        return -1;
4568}
4569
4570/*
4571 * Returns:
4572 *  -1 if an error happened
4573 *   0 if the patch applied cleanly
4574 *   1 if the patch did not apply cleanly
4575 */
4576static int write_out_results(struct apply_state *state, struct patch *list)
4577{
4578        int phase;
4579        int errs = 0;
4580        struct patch *l;
4581        struct string_list cpath = STRING_LIST_INIT_DUP;
4582
4583        for (phase = 0; phase < 2; phase++) {
4584                l = list;
4585                while (l) {
4586                        if (l->rejected)
4587                                errs = 1;
4588                        else {
4589                                if (write_out_one_result(state, l, phase)) {
4590                                        string_list_clear(&cpath, 0);
4591                                        return -1;
4592                                }
4593                                if (phase == 1) {
4594                                        if (write_out_one_reject(state, l))
4595                                                errs = 1;
4596                                        if (l->conflicted_threeway) {
4597                                                string_list_append(&cpath, l->new_name);
4598                                                errs = 1;
4599                                        }
4600                                }
4601                        }
4602                        l = l->next;
4603                }
4604        }
4605
4606        if (cpath.nr) {
4607                struct string_list_item *item;
4608
4609                string_list_sort(&cpath);
4610                if (state->apply_verbosity > verbosity_silent) {
4611                        for_each_string_list_item(item, &cpath)
4612                                fprintf(stderr, "U %s\n", item->string);
4613                }
4614                string_list_clear(&cpath, 0);
4615
4616                rerere(0);
4617        }
4618
4619        return errs;
4620}
4621
4622/*
4623 * Try to apply a patch.
4624 *
4625 * Returns:
4626 *  -128 if a bad error happened (like patch unreadable)
4627 *  -1 if patch did not apply and user cannot deal with it
4628 *   0 if the patch applied
4629 *   1 if the patch did not apply but user might fix it
4630 */
4631static int apply_patch(struct apply_state *state,
4632                       int fd,
4633                       const char *filename,
4634                       int options)
4635{
4636        size_t offset;
4637        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4638        struct patch *list = NULL, **listp = &list;
4639        int skipped_patch = 0;
4640        int res = 0;
4641
4642        state->patch_input_file = filename;
4643        if (read_patch_file(&buf, fd) < 0)
4644                return -128;
4645        offset = 0;
4646        while (offset < buf.len) {
4647                struct patch *patch;
4648                int nr;
4649
4650                patch = xcalloc(1, sizeof(*patch));
4651                patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4652                patch->recount =  !!(options & APPLY_OPT_RECOUNT);
4653                nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4654                if (nr < 0) {
4655                        free_patch(patch);
4656                        if (nr == -128) {
4657                                res = -128;
4658                                goto end;
4659                        }
4660                        break;
4661                }
4662                if (state->apply_in_reverse)
4663                        reverse_patches(patch);
4664                if (use_patch(state, patch)) {
4665                        patch_stats(state, patch);
4666                        *listp = patch;
4667                        listp = &patch->next;
4668                }
4669                else {
4670                        if (state->apply_verbosity > verbosity_normal)
4671                                say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4672                        free_patch(patch);
4673                        skipped_patch++;
4674                }
4675                offset += nr;
4676        }
4677
4678        if (!list && !skipped_patch) {
4679                error(_("unrecognized input"));
4680                res = -128;
4681                goto end;
4682        }
4683
4684        if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4685                state->apply = 0;
4686
4687        state->update_index = state->check_index && state->apply;
4688        if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4689                if (state->index_file)
4690                        hold_lock_file_for_update(&state->lock_file,
4691                                                  state->index_file,
4692                                                  LOCK_DIE_ON_ERROR);
4693                else
4694                        hold_locked_index(&state->lock_file, LOCK_DIE_ON_ERROR);
4695        }
4696
4697        if (state->check_index && read_apply_cache(state) < 0) {
4698                error(_("unable to read index file"));
4699                res = -128;
4700                goto end;
4701        }
4702
4703        if (state->check || state->apply) {
4704                int r = check_patch_list(state, list);
4705                if (r == -128) {
4706                        res = -128;
4707                        goto end;
4708                }
4709                if (r < 0 && !state->apply_with_reject) {
4710                        res = -1;
4711                        goto end;
4712                }
4713        }
4714
4715        if (state->apply) {
4716                int write_res = write_out_results(state, list);
4717                if (write_res < 0) {
4718                        res = -128;
4719                        goto end;
4720                }
4721                if (write_res > 0) {
4722                        /* with --3way, we still need to write the index out */
4723                        res = state->apply_with_reject ? -1 : 1;
4724                        goto end;
4725                }
4726        }
4727
4728        if (state->fake_ancestor &&
4729            build_fake_ancestor(state, list)) {
4730                res = -128;
4731                goto end;
4732        }
4733
4734        if (state->diffstat && state->apply_verbosity > verbosity_silent)
4735                stat_patch_list(state, list);
4736
4737        if (state->numstat && state->apply_verbosity > verbosity_silent)
4738                numstat_patch_list(state, list);
4739
4740        if (state->summary && state->apply_verbosity > verbosity_silent)
4741                summary_patch_list(list);
4742
4743end:
4744        free_patch_list(list);
4745        strbuf_release(&buf);
4746        string_list_clear(&state->fn_table, 0);
4747        return res;
4748}
4749
4750static int apply_option_parse_exclude(const struct option *opt,
4751                                      const char *arg, int unset)
4752{
4753        struct apply_state *state = opt->value;
4754        add_name_limit(state, arg, 1);
4755        return 0;
4756}
4757
4758static int apply_option_parse_include(const struct option *opt,
4759                                      const char *arg, int unset)
4760{
4761        struct apply_state *state = opt->value;
4762        add_name_limit(state, arg, 0);
4763        state->has_include = 1;
4764        return 0;
4765}
4766
4767static int apply_option_parse_p(const struct option *opt,
4768                                const char *arg,
4769                                int unset)
4770{
4771        struct apply_state *state = opt->value;
4772        state->p_value = atoi(arg);
4773        state->p_value_known = 1;
4774        return 0;
4775}
4776
4777static int apply_option_parse_space_change(const struct option *opt,
4778                                           const char *arg, int unset)
4779{
4780        struct apply_state *state = opt->value;
4781        if (unset)
4782                state->ws_ignore_action = ignore_ws_none;
4783        else
4784                state->ws_ignore_action = ignore_ws_change;
4785        return 0;
4786}
4787
4788static int apply_option_parse_whitespace(const struct option *opt,
4789                                         const char *arg, int unset)
4790{
4791        struct apply_state *state = opt->value;
4792        state->whitespace_option = arg;
4793        if (parse_whitespace_option(state, arg))
4794                exit(1);
4795        return 0;
4796}
4797
4798static int apply_option_parse_directory(const struct option *opt,
4799                                        const char *arg, int unset)
4800{
4801        struct apply_state *state = opt->value;
4802        strbuf_reset(&state->root);
4803        strbuf_addstr(&state->root, arg);
4804        strbuf_complete(&state->root, '/');
4805        return 0;
4806}
4807
4808int apply_all_patches(struct apply_state *state,
4809                      int argc,
4810                      const char **argv,
4811                      int options)
4812{
4813        int i;
4814        int res;
4815        int errs = 0;
4816        int read_stdin = 1;
4817
4818        for (i = 0; i < argc; i++) {
4819                const char *arg = argv[i];
4820                char *to_free = NULL;
4821                int fd;
4822
4823                if (!strcmp(arg, "-")) {
4824                        res = apply_patch(state, 0, "<stdin>", options);
4825                        if (res < 0)
4826                                goto end;
4827                        errs |= res;
4828                        read_stdin = 0;
4829                        continue;
4830                } else
4831                        arg = to_free = prefix_filename(state->prefix, arg);
4832
4833                fd = open(arg, O_RDONLY);
4834                if (fd < 0) {
4835                        error(_("can't open patch '%s': %s"), arg, strerror(errno));
4836                        res = -128;
4837                        free(to_free);
4838                        goto end;
4839                }
4840                read_stdin = 0;
4841                set_default_whitespace_mode(state);
4842                res = apply_patch(state, fd, arg, options);
4843                close(fd);
4844                free(to_free);
4845                if (res < 0)
4846                        goto end;
4847                errs |= res;
4848        }
4849        set_default_whitespace_mode(state);
4850        if (read_stdin) {
4851                res = apply_patch(state, 0, "<stdin>", options);
4852                if (res < 0)
4853                        goto end;
4854                errs |= res;
4855        }
4856
4857        if (state->whitespace_error) {
4858                if (state->squelch_whitespace_errors &&
4859                    state->squelch_whitespace_errors < state->whitespace_error) {
4860                        int squelched =
4861                                state->whitespace_error - state->squelch_whitespace_errors;
4862                        warning(Q_("squelched %d whitespace error",
4863                                   "squelched %d whitespace errors",
4864                                   squelched),
4865                                squelched);
4866                }
4867                if (state->ws_error_action == die_on_ws_error) {
4868                        error(Q_("%d line adds whitespace errors.",
4869                                 "%d lines add whitespace errors.",
4870                                 state->whitespace_error),
4871                              state->whitespace_error);
4872                        res = -128;
4873                        goto end;
4874                }
4875                if (state->applied_after_fixing_ws && state->apply)
4876                        warning(Q_("%d line applied after"
4877                                   " fixing whitespace errors.",
4878                                   "%d lines applied after"
4879                                   " fixing whitespace errors.",
4880                                   state->applied_after_fixing_ws),
4881                                state->applied_after_fixing_ws);
4882                else if (state->whitespace_error)
4883                        warning(Q_("%d line adds whitespace errors.",
4884                                   "%d lines add whitespace errors.",
4885                                   state->whitespace_error),
4886                                state->whitespace_error);
4887        }
4888
4889        if (state->update_index) {
4890                res = write_locked_index(&the_index, &state->lock_file, COMMIT_LOCK);
4891                if (res) {
4892                        error(_("Unable to write new index file"));
4893                        res = -128;
4894                        goto end;
4895                }
4896        }
4897
4898        res = !!errs;
4899
4900end:
4901        rollback_lock_file(&state->lock_file);
4902
4903        if (state->apply_verbosity <= verbosity_silent) {
4904                set_error_routine(state->saved_error_routine);
4905                set_warn_routine(state->saved_warn_routine);
4906        }
4907
4908        if (res > -1)
4909                return res;
4910        return (res == -1 ? 1 : 128);
4911}
4912
4913int apply_parse_options(int argc, const char **argv,
4914                        struct apply_state *state,
4915                        int *force_apply, int *options,
4916                        const char * const *apply_usage)
4917{
4918        struct option builtin_apply_options[] = {
4919                { OPTION_CALLBACK, 0, "exclude", state, N_("path"),
4920                        N_("don't apply changes matching the given path"),
4921                        0, apply_option_parse_exclude },
4922                { OPTION_CALLBACK, 0, "include", state, N_("path"),
4923                        N_("apply changes matching the given path"),
4924                        0, apply_option_parse_include },
4925                { OPTION_CALLBACK, 'p', NULL, state, N_("num"),
4926                        N_("remove <num> leading slashes from traditional diff paths"),
4927                        0, apply_option_parse_p },
4928                OPT_BOOL(0, "no-add", &state->no_add,
4929                        N_("ignore additions made by the patch")),
4930                OPT_BOOL(0, "stat", &state->diffstat,
4931                        N_("instead of applying the patch, output diffstat for the input")),
4932                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4933                OPT_NOOP_NOARG(0, "binary"),
4934                OPT_BOOL(0, "numstat", &state->numstat,
4935                        N_("show number of added and deleted lines in decimal notation")),
4936                OPT_BOOL(0, "summary", &state->summary,
4937                        N_("instead of applying the patch, output a summary for the input")),
4938                OPT_BOOL(0, "check", &state->check,
4939                        N_("instead of applying the patch, see if the patch is applicable")),
4940                OPT_BOOL(0, "index", &state->check_index,
4941                        N_("make sure the patch is applicable to the current index")),
4942                OPT_BOOL(0, "cached", &state->cached,
4943                        N_("apply a patch without touching the working tree")),
4944                OPT_BOOL(0, "unsafe-paths", &state->unsafe_paths,
4945                        N_("accept a patch that touches outside the working area")),
4946                OPT_BOOL(0, "apply", force_apply,
4947                        N_("also apply the patch (use with --stat/--summary/--check)")),
4948                OPT_BOOL('3', "3way", &state->threeway,
4949                         N_( "attempt three-way merge if a patch does not apply")),
4950                OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
4951                        N_("build a temporary index based on embedded index information")),
4952                /* Think twice before adding "--nul" synonym to this */
4953                OPT_SET_INT('z', NULL, &state->line_termination,
4954                        N_("paths are separated with NUL character"), '\0'),
4955                OPT_INTEGER('C', NULL, &state->p_context,
4956                                N_("ensure at least <n> lines of context match")),
4957                { OPTION_CALLBACK, 0, "whitespace", state, N_("action"),
4958                        N_("detect new or modified lines that have whitespace errors"),
4959                        0, apply_option_parse_whitespace },
4960                { OPTION_CALLBACK, 0, "ignore-space-change", state, NULL,
4961                        N_("ignore changes in whitespace when finding context"),
4962                        PARSE_OPT_NOARG, apply_option_parse_space_change },
4963                { OPTION_CALLBACK, 0, "ignore-whitespace", state, NULL,
4964                        N_("ignore changes in whitespace when finding context"),
4965                        PARSE_OPT_NOARG, apply_option_parse_space_change },
4966                OPT_BOOL('R', "reverse", &state->apply_in_reverse,
4967                        N_("apply the patch in reverse")),
4968                OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
4969                        N_("don't expect at least one line of context")),
4970                OPT_BOOL(0, "reject", &state->apply_with_reject,
4971                        N_("leave the rejected hunks in corresponding *.rej files")),
4972                OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
4973                        N_("allow overlapping hunks")),
4974                OPT__VERBOSE(&state->apply_verbosity, N_("be verbose")),
4975                OPT_BIT(0, "inaccurate-eof", options,
4976                        N_("tolerate incorrectly detected missing new-line at the end of file"),
4977                        APPLY_OPT_INACCURATE_EOF),
4978                OPT_BIT(0, "recount", options,
4979                        N_("do not trust the line counts in the hunk headers"),
4980                        APPLY_OPT_RECOUNT),
4981                { OPTION_CALLBACK, 0, "directory", state, N_("root"),
4982                        N_("prepend <root> to all filenames"),
4983                        0, apply_option_parse_directory },
4984                OPT_END()
4985        };
4986
4987        return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);
4988}