apply.con commit Allow the test suite to pass in a directory whose name contains spaces (567c53d)
   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 "blob.h"
  12#include "delta.h"
  13#include "diff.h"
  14#include "dir.h"
  15#include "xdiff-interface.h"
  16#include "ll-merge.h"
  17#include "lockfile.h"
  18#include "parse-options.h"
  19#include "quote.h"
  20#include "rerere.h"
  21#include "apply.h"
  22
  23static void git_apply_config(void)
  24{
  25        git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
  26        git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
  27        git_config(git_default_config, NULL);
  28}
  29
  30static int parse_whitespace_option(struct apply_state *state, const char *option)
  31{
  32        if (!option) {
  33                state->ws_error_action = warn_on_ws_error;
  34                return 0;
  35        }
  36        if (!strcmp(option, "warn")) {
  37                state->ws_error_action = warn_on_ws_error;
  38                return 0;
  39        }
  40        if (!strcmp(option, "nowarn")) {
  41                state->ws_error_action = nowarn_ws_error;
  42                return 0;
  43        }
  44        if (!strcmp(option, "error")) {
  45                state->ws_error_action = die_on_ws_error;
  46                return 0;
  47        }
  48        if (!strcmp(option, "error-all")) {
  49                state->ws_error_action = die_on_ws_error;
  50                state->squelch_whitespace_errors = 0;
  51                return 0;
  52        }
  53        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
  54                state->ws_error_action = correct_ws_error;
  55                return 0;
  56        }
  57        return error(_("unrecognized whitespace option '%s'"), option);
  58}
  59
  60static int parse_ignorewhitespace_option(struct apply_state *state,
  61                                                 const char *option)
  62{
  63        if (!option || !strcmp(option, "no") ||
  64            !strcmp(option, "false") || !strcmp(option, "never") ||
  65            !strcmp(option, "none")) {
  66                state->ws_ignore_action = ignore_ws_none;
  67                return 0;
  68        }
  69        if (!strcmp(option, "change")) {
  70                state->ws_ignore_action = ignore_ws_change;
  71                return 0;
  72        }
  73        return error(_("unrecognized whitespace ignore option '%s'"), option);
  74}
  75
  76int init_apply_state(struct apply_state *state,
  77                     const char *prefix,
  78                     struct lock_file *lock_file)
  79{
  80        memset(state, 0, sizeof(*state));
  81        state->prefix = prefix;
  82        state->prefix_length = state->prefix ? strlen(state->prefix) : 0;
  83        state->lock_file = lock_file;
  84        state->newfd = -1;
  85        state->apply = 1;
  86        state->line_termination = '\n';
  87        state->p_value = 1;
  88        state->p_context = UINT_MAX;
  89        state->squelch_whitespace_errors = 5;
  90        state->ws_error_action = warn_on_ws_error;
  91        state->ws_ignore_action = ignore_ws_none;
  92        state->linenr = 1;
  93        string_list_init(&state->fn_table, 0);
  94        string_list_init(&state->limit_by_name, 0);
  95        string_list_init(&state->symlink_changes, 0);
  96        strbuf_init(&state->root, 0);
  97
  98        git_apply_config();
  99        if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
 100                return -1;
 101        if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
 102                return -1;
 103        return 0;
 104}
 105
 106void clear_apply_state(struct apply_state *state)
 107{
 108        string_list_clear(&state->limit_by_name, 0);
 109        string_list_clear(&state->symlink_changes, 0);
 110        strbuf_release(&state->root);
 111
 112        /* &state->fn_table is cleared at the end of apply_patch() */
 113}
 114
 115static void mute_routine(const char *msg, va_list params)
 116{
 117        /* do nothing */
 118}
 119
 120int check_apply_state(struct apply_state *state, int force_apply)
 121{
 122        int is_not_gitdir = !startup_info->have_repository;
 123
 124        if (state->apply_with_reject && state->threeway)
 125                return error(_("--reject and --3way cannot be used together."));
 126        if (state->cached && state->threeway)
 127                return error(_("--cached and --3way cannot be used together."));
 128        if (state->threeway) {
 129                if (is_not_gitdir)
 130                        return error(_("--3way outside a repository"));
 131                state->check_index = 1;
 132        }
 133        if (state->apply_with_reject) {
 134                state->apply = 1;
 135                if (state->apply_verbosity == verbosity_normal)
 136                        state->apply_verbosity = verbosity_verbose;
 137        }
 138        if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
 139                state->apply = 0;
 140        if (state->check_index && is_not_gitdir)
 141                return error(_("--index outside a repository"));
 142        if (state->cached) {
 143                if (is_not_gitdir)
 144                        return error(_("--cached outside a repository"));
 145                state->check_index = 1;
 146        }
 147        if (state->check_index)
 148                state->unsafe_paths = 0;
 149        if (!state->lock_file)
 150                return error("BUG: state->lock_file should not be NULL");
 151
 152        if (state->apply_verbosity <= verbosity_silent) {
 153                state->saved_error_routine = get_error_routine();
 154                state->saved_warn_routine = get_warn_routine();
 155                set_error_routine(mute_routine);
 156                set_warn_routine(mute_routine);
 157        }
 158
 159        return 0;
 160}
 161
 162static void set_default_whitespace_mode(struct apply_state *state)
 163{
 164        if (!state->whitespace_option && !apply_default_whitespace)
 165                state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
 166}
 167
 168/*
 169 * This represents one "hunk" from a patch, starting with
 170 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 171 * patch text is pointed at by patch, and its byte length
 172 * is stored in size.  leading and trailing are the number
 173 * of context lines.
 174 */
 175struct fragment {
 176        unsigned long leading, trailing;
 177        unsigned long oldpos, oldlines;
 178        unsigned long newpos, newlines;
 179        /*
 180         * 'patch' is usually borrowed from buf in apply_patch(),
 181         * but some codepaths store an allocated buffer.
 182         */
 183        const char *patch;
 184        unsigned free_patch:1,
 185                rejected:1;
 186        int size;
 187        int linenr;
 188        struct fragment *next;
 189};
 190
 191/*
 192 * When dealing with a binary patch, we reuse "leading" field
 193 * to store the type of the binary hunk, either deflated "delta"
 194 * or deflated "literal".
 195 */
 196#define binary_patch_method leading
 197#define BINARY_DELTA_DEFLATED   1
 198#define BINARY_LITERAL_DEFLATED 2
 199
 200/*
 201 * This represents a "patch" to a file, both metainfo changes
 202 * such as creation/deletion, filemode and content changes represented
 203 * as a series of fragments.
 204 */
 205struct patch {
 206        char *new_name, *old_name, *def_name;
 207        unsigned int old_mode, new_mode;
 208        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 209        int rejected;
 210        unsigned ws_rule;
 211        int lines_added, lines_deleted;
 212        int score;
 213        int extension_linenr; /* first line specifying delete/new/rename/copy */
 214        unsigned int is_toplevel_relative:1;
 215        unsigned int inaccurate_eof:1;
 216        unsigned int is_binary:1;
 217        unsigned int is_copy:1;
 218        unsigned int is_rename:1;
 219        unsigned int recount:1;
 220        unsigned int conflicted_threeway:1;
 221        unsigned int direct_to_threeway:1;
 222        struct fragment *fragments;
 223        char *result;
 224        size_t resultsize;
 225        char old_sha1_prefix[41];
 226        char new_sha1_prefix[41];
 227        struct patch *next;
 228
 229        /* three-way fallback result */
 230        struct object_id threeway_stage[3];
 231};
 232
 233static void free_fragment_list(struct fragment *list)
 234{
 235        while (list) {
 236                struct fragment *next = list->next;
 237                if (list->free_patch)
 238                        free((char *)list->patch);
 239                free(list);
 240                list = next;
 241        }
 242}
 243
 244static void free_patch(struct patch *patch)
 245{
 246        free_fragment_list(patch->fragments);
 247        free(patch->def_name);
 248        free(patch->old_name);
 249        free(patch->new_name);
 250        free(patch->result);
 251        free(patch);
 252}
 253
 254static void free_patch_list(struct patch *list)
 255{
 256        while (list) {
 257                struct patch *next = list->next;
 258                free_patch(list);
 259                list = next;
 260        }
 261}
 262
 263/*
 264 * A line in a file, len-bytes long (includes the terminating LF,
 265 * except for an incomplete line at the end if the file ends with
 266 * one), and its contents hashes to 'hash'.
 267 */
 268struct line {
 269        size_t len;
 270        unsigned hash : 24;
 271        unsigned flag : 8;
 272#define LINE_COMMON     1
 273#define LINE_PATCHED    2
 274};
 275
 276/*
 277 * This represents a "file", which is an array of "lines".
 278 */
 279struct image {
 280        char *buf;
 281        size_t len;
 282        size_t nr;
 283        size_t alloc;
 284        struct line *line_allocated;
 285        struct line *line;
 286};
 287
 288static uint32_t hash_line(const char *cp, size_t len)
 289{
 290        size_t i;
 291        uint32_t h;
 292        for (i = 0, h = 0; i < len; i++) {
 293                if (!isspace(cp[i])) {
 294                        h = h * 3 + (cp[i] & 0xff);
 295                }
 296        }
 297        return h;
 298}
 299
 300/*
 301 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 302 * whitespace difference. Returns 1 if they match, 0 otherwise
 303 */
 304static int fuzzy_matchlines(const char *s1, size_t n1,
 305                            const char *s2, size_t n2)
 306{
 307        const char *last1 = s1 + n1 - 1;
 308        const char *last2 = s2 + n2 - 1;
 309        int result = 0;
 310
 311        /* ignore line endings */
 312        while ((*last1 == '\r') || (*last1 == '\n'))
 313                last1--;
 314        while ((*last2 == '\r') || (*last2 == '\n'))
 315                last2--;
 316
 317        /* skip leading whitespaces, if both begin with whitespace */
 318        if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
 319                while (isspace(*s1) && (s1 <= last1))
 320                        s1++;
 321                while (isspace(*s2) && (s2 <= last2))
 322                        s2++;
 323        }
 324        /* early return if both lines are empty */
 325        if ((s1 > last1) && (s2 > last2))
 326                return 1;
 327        while (!result) {
 328                result = *s1++ - *s2++;
 329                /*
 330                 * Skip whitespace inside. We check for whitespace on
 331                 * both buffers because we don't want "a b" to match
 332                 * "ab"
 333                 */
 334                if (isspace(*s1) && isspace(*s2)) {
 335                        while (isspace(*s1) && s1 <= last1)
 336                                s1++;
 337                        while (isspace(*s2) && s2 <= last2)
 338                                s2++;
 339                }
 340                /*
 341                 * If we reached the end on one side only,
 342                 * lines don't match
 343                 */
 344                if (
 345                    ((s2 > last2) && (s1 <= last1)) ||
 346                    ((s1 > last1) && (s2 <= last2)))
 347                        return 0;
 348                if ((s1 > last1) && (s2 > last2))
 349                        break;
 350        }
 351
 352        return !result;
 353}
 354
 355static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 356{
 357        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 358        img->line_allocated[img->nr].len = len;
 359        img->line_allocated[img->nr].hash = hash_line(bol, len);
 360        img->line_allocated[img->nr].flag = flag;
 361        img->nr++;
 362}
 363
 364/*
 365 * "buf" has the file contents to be patched (read from various sources).
 366 * attach it to "image" and add line-based index to it.
 367 * "image" now owns the "buf".
 368 */
 369static void prepare_image(struct image *image, char *buf, size_t len,
 370                          int prepare_linetable)
 371{
 372        const char *cp, *ep;
 373
 374        memset(image, 0, sizeof(*image));
 375        image->buf = buf;
 376        image->len = len;
 377
 378        if (!prepare_linetable)
 379                return;
 380
 381        ep = image->buf + image->len;
 382        cp = image->buf;
 383        while (cp < ep) {
 384                const char *next;
 385                for (next = cp; next < ep && *next != '\n'; next++)
 386                        ;
 387                if (next < ep)
 388                        next++;
 389                add_line_info(image, cp, next - cp, 0);
 390                cp = next;
 391        }
 392        image->line = image->line_allocated;
 393}
 394
 395static void clear_image(struct image *image)
 396{
 397        free(image->buf);
 398        free(image->line_allocated);
 399        memset(image, 0, sizeof(*image));
 400}
 401
 402/* fmt must contain _one_ %s and no other substitution */
 403static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 404{
 405        struct strbuf sb = STRBUF_INIT;
 406
 407        if (patch->old_name && patch->new_name &&
 408            strcmp(patch->old_name, patch->new_name)) {
 409                quote_c_style(patch->old_name, &sb, NULL, 0);
 410                strbuf_addstr(&sb, " => ");
 411                quote_c_style(patch->new_name, &sb, NULL, 0);
 412        } else {
 413                const char *n = patch->new_name;
 414                if (!n)
 415                        n = patch->old_name;
 416                quote_c_style(n, &sb, NULL, 0);
 417        }
 418        fprintf(output, fmt, sb.buf);
 419        fputc('\n', output);
 420        strbuf_release(&sb);
 421}
 422
 423#define SLOP (16)
 424
 425static int read_patch_file(struct strbuf *sb, int fd)
 426{
 427        if (strbuf_read(sb, fd, 0) < 0)
 428                return error_errno("git apply: failed to read");
 429
 430        /*
 431         * Make sure that we have some slop in the buffer
 432         * so that we can do speculative "memcmp" etc, and
 433         * see to it that it is NUL-filled.
 434         */
 435        strbuf_grow(sb, SLOP);
 436        memset(sb->buf + sb->len, 0, SLOP);
 437        return 0;
 438}
 439
 440static unsigned long linelen(const char *buffer, unsigned long size)
 441{
 442        unsigned long len = 0;
 443        while (size--) {
 444                len++;
 445                if (*buffer++ == '\n')
 446                        break;
 447        }
 448        return len;
 449}
 450
 451static int is_dev_null(const char *str)
 452{
 453        return skip_prefix(str, "/dev/null", &str) && isspace(*str);
 454}
 455
 456#define TERM_SPACE      1
 457#define TERM_TAB        2
 458
 459static int name_terminate(int c, int terminate)
 460{
 461        if (c == ' ' && !(terminate & TERM_SPACE))
 462                return 0;
 463        if (c == '\t' && !(terminate & TERM_TAB))
 464                return 0;
 465
 466        return 1;
 467}
 468
 469/* remove double slashes to make --index work with such filenames */
 470static char *squash_slash(char *name)
 471{
 472        int i = 0, j = 0;
 473
 474        if (!name)
 475                return NULL;
 476
 477        while (name[i]) {
 478                if ((name[j++] = name[i++]) == '/')
 479                        while (name[i] == '/')
 480                                i++;
 481        }
 482        name[j] = '\0';
 483        return name;
 484}
 485
 486static char *find_name_gnu(struct apply_state *state,
 487                           const char *line,
 488                           const char *def,
 489                           int p_value)
 490{
 491        struct strbuf name = STRBUF_INIT;
 492        char *cp;
 493
 494        /*
 495         * Proposed "new-style" GNU patch/diff format; see
 496         * http://marc.info/?l=git&m=112927316408690&w=2
 497         */
 498        if (unquote_c_style(&name, line, NULL)) {
 499                strbuf_release(&name);
 500                return NULL;
 501        }
 502
 503        for (cp = name.buf; p_value; p_value--) {
 504                cp = strchr(cp, '/');
 505                if (!cp) {
 506                        strbuf_release(&name);
 507                        return NULL;
 508                }
 509                cp++;
 510        }
 511
 512        strbuf_remove(&name, 0, cp - name.buf);
 513        if (state->root.len)
 514                strbuf_insert(&name, 0, state->root.buf, state->root.len);
 515        return squash_slash(strbuf_detach(&name, NULL));
 516}
 517
 518static size_t sane_tz_len(const char *line, size_t len)
 519{
 520        const char *tz, *p;
 521
 522        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 523                return 0;
 524        tz = line + len - strlen(" +0500");
 525
 526        if (tz[1] != '+' && tz[1] != '-')
 527                return 0;
 528
 529        for (p = tz + 2; p != line + len; p++)
 530                if (!isdigit(*p))
 531                        return 0;
 532
 533        return line + len - tz;
 534}
 535
 536static size_t tz_with_colon_len(const char *line, size_t len)
 537{
 538        const char *tz, *p;
 539
 540        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 541                return 0;
 542        tz = line + len - strlen(" +08:00");
 543
 544        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 545                return 0;
 546        p = tz + 2;
 547        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 548            !isdigit(*p++) || !isdigit(*p++))
 549                return 0;
 550
 551        return line + len - tz;
 552}
 553
 554static size_t date_len(const char *line, size_t len)
 555{
 556        const char *date, *p;
 557
 558        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 559                return 0;
 560        p = date = line + len - strlen("72-02-05");
 561
 562        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 563            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 564            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 565                return 0;
 566
 567        if (date - line >= strlen("19") &&
 568            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 569                date -= strlen("19");
 570
 571        return line + len - date;
 572}
 573
 574static size_t short_time_len(const char *line, size_t len)
 575{
 576        const char *time, *p;
 577
 578        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 579                return 0;
 580        p = time = line + len - strlen(" 07:01:32");
 581
 582        /* Permit 1-digit hours? */
 583        if (*p++ != ' ' ||
 584            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 585            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 586            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 587                return 0;
 588
 589        return line + len - time;
 590}
 591
 592static size_t fractional_time_len(const char *line, size_t len)
 593{
 594        const char *p;
 595        size_t n;
 596
 597        /* Expected format: 19:41:17.620000023 */
 598        if (!len || !isdigit(line[len - 1]))
 599                return 0;
 600        p = line + len - 1;
 601
 602        /* Fractional seconds. */
 603        while (p > line && isdigit(*p))
 604                p--;
 605        if (*p != '.')
 606                return 0;
 607
 608        /* Hours, minutes, and whole seconds. */
 609        n = short_time_len(line, p - line);
 610        if (!n)
 611                return 0;
 612
 613        return line + len - p + n;
 614}
 615
 616static size_t trailing_spaces_len(const char *line, size_t len)
 617{
 618        const char *p;
 619
 620        /* Expected format: ' ' x (1 or more)  */
 621        if (!len || line[len - 1] != ' ')
 622                return 0;
 623
 624        p = line + len;
 625        while (p != line) {
 626                p--;
 627                if (*p != ' ')
 628                        return line + len - (p + 1);
 629        }
 630
 631        /* All spaces! */
 632        return len;
 633}
 634
 635static size_t diff_timestamp_len(const char *line, size_t len)
 636{
 637        const char *end = line + len;
 638        size_t n;
 639
 640        /*
 641         * Posix: 2010-07-05 19:41:17
 642         * GNU: 2010-07-05 19:41:17.620000023 -0500
 643         */
 644
 645        if (!isdigit(end[-1]))
 646                return 0;
 647
 648        n = sane_tz_len(line, end - line);
 649        if (!n)
 650                n = tz_with_colon_len(line, end - line);
 651        end -= n;
 652
 653        n = short_time_len(line, end - line);
 654        if (!n)
 655                n = fractional_time_len(line, end - line);
 656        end -= n;
 657
 658        n = date_len(line, end - line);
 659        if (!n) /* No date.  Too bad. */
 660                return 0;
 661        end -= n;
 662
 663        if (end == line)        /* No space before date. */
 664                return 0;
 665        if (end[-1] == '\t') {  /* Success! */
 666                end--;
 667                return line + len - end;
 668        }
 669        if (end[-1] != ' ')     /* No space before date. */
 670                return 0;
 671
 672        /* Whitespace damage. */
 673        end -= trailing_spaces_len(line, end - line);
 674        return line + len - end;
 675}
 676
 677static char *find_name_common(struct apply_state *state,
 678                              const char *line,
 679                              const char *def,
 680                              int p_value,
 681                              const char *end,
 682                              int terminate)
 683{
 684        int len;
 685        const char *start = NULL;
 686
 687        if (p_value == 0)
 688                start = line;
 689        while (line != end) {
 690                char c = *line;
 691
 692                if (!end && isspace(c)) {
 693                        if (c == '\n')
 694                                break;
 695                        if (name_terminate(c, terminate))
 696                                break;
 697                }
 698                line++;
 699                if (c == '/' && !--p_value)
 700                        start = line;
 701        }
 702        if (!start)
 703                return squash_slash(xstrdup_or_null(def));
 704        len = line - start;
 705        if (!len)
 706                return squash_slash(xstrdup_or_null(def));
 707
 708        /*
 709         * Generally we prefer the shorter name, especially
 710         * if the other one is just a variation of that with
 711         * something else tacked on to the end (ie "file.orig"
 712         * or "file~").
 713         */
 714        if (def) {
 715                int deflen = strlen(def);
 716                if (deflen < len && !strncmp(start, def, deflen))
 717                        return squash_slash(xstrdup(def));
 718        }
 719
 720        if (state->root.len) {
 721                char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
 722                return squash_slash(ret);
 723        }
 724
 725        return squash_slash(xmemdupz(start, len));
 726}
 727
 728static char *find_name(struct apply_state *state,
 729                       const char *line,
 730                       char *def,
 731                       int p_value,
 732                       int terminate)
 733{
 734        if (*line == '"') {
 735                char *name = find_name_gnu(state, line, def, p_value);
 736                if (name)
 737                        return name;
 738        }
 739
 740        return find_name_common(state, line, def, p_value, NULL, terminate);
 741}
 742
 743static char *find_name_traditional(struct apply_state *state,
 744                                   const char *line,
 745                                   char *def,
 746                                   int p_value)
 747{
 748        size_t len;
 749        size_t date_len;
 750
 751        if (*line == '"') {
 752                char *name = find_name_gnu(state, line, def, p_value);
 753                if (name)
 754                        return name;
 755        }
 756
 757        len = strchrnul(line, '\n') - line;
 758        date_len = diff_timestamp_len(line, len);
 759        if (!date_len)
 760                return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
 761        len -= date_len;
 762
 763        return find_name_common(state, line, def, p_value, line + len, 0);
 764}
 765
 766static int count_slashes(const char *cp)
 767{
 768        int cnt = 0;
 769        char ch;
 770
 771        while ((ch = *cp++))
 772                if (ch == '/')
 773                        cnt++;
 774        return cnt;
 775}
 776
 777/*
 778 * Given the string after "--- " or "+++ ", guess the appropriate
 779 * p_value for the given patch.
 780 */
 781static int guess_p_value(struct apply_state *state, const char *nameline)
 782{
 783        char *name, *cp;
 784        int val = -1;
 785
 786        if (is_dev_null(nameline))
 787                return -1;
 788        name = find_name_traditional(state, nameline, NULL, 0);
 789        if (!name)
 790                return -1;
 791        cp = strchr(name, '/');
 792        if (!cp)
 793                val = 0;
 794        else if (state->prefix) {
 795                /*
 796                 * Does it begin with "a/$our-prefix" and such?  Then this is
 797                 * very likely to apply to our directory.
 798                 */
 799                if (!strncmp(name, state->prefix, state->prefix_length))
 800                        val = count_slashes(state->prefix);
 801                else {
 802                        cp++;
 803                        if (!strncmp(cp, state->prefix, state->prefix_length))
 804                                val = count_slashes(state->prefix) + 1;
 805                }
 806        }
 807        free(name);
 808        return val;
 809}
 810
 811/*
 812 * Does the ---/+++ line have the POSIX timestamp after the last HT?
 813 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 814 * this such a timestamp?
 815 */
 816static int has_epoch_timestamp(const char *nameline)
 817{
 818        /*
 819         * We are only interested in epoch timestamp; any non-zero
 820         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 821         * For the same reason, the date must be either 1969-12-31 or
 822         * 1970-01-01, and the seconds part must be "00".
 823         */
 824        const char stamp_regexp[] =
 825                "^(1969-12-31|1970-01-01)"
 826                " "
 827                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 828                " "
 829                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 830        const char *timestamp = NULL, *cp, *colon;
 831        static regex_t *stamp;
 832        regmatch_t m[10];
 833        int zoneoffset;
 834        int hourminute;
 835        int status;
 836
 837        for (cp = nameline; *cp != '\n'; cp++) {
 838                if (*cp == '\t')
 839                        timestamp = cp + 1;
 840        }
 841        if (!timestamp)
 842                return 0;
 843        if (!stamp) {
 844                stamp = xmalloc(sizeof(*stamp));
 845                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 846                        warning(_("Cannot prepare timestamp regexp %s"),
 847                                stamp_regexp);
 848                        return 0;
 849                }
 850        }
 851
 852        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 853        if (status) {
 854                if (status != REG_NOMATCH)
 855                        warning(_("regexec returned %d for input: %s"),
 856                                status, timestamp);
 857                return 0;
 858        }
 859
 860        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 861        if (*colon == ':')
 862                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 863        else
 864                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 865        if (timestamp[m[3].rm_so] == '-')
 866                zoneoffset = -zoneoffset;
 867
 868        /*
 869         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 870         * (west of GMT) or 1970-01-01 (east of GMT)
 871         */
 872        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 873            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 874                return 0;
 875
 876        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 877                      strtol(timestamp + 14, NULL, 10) -
 878                      zoneoffset);
 879
 880        return ((zoneoffset < 0 && hourminute == 1440) ||
 881                (0 <= zoneoffset && !hourminute));
 882}
 883
 884/*
 885 * Get the name etc info from the ---/+++ lines of a traditional patch header
 886 *
 887 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 888 * files, we can happily check the index for a match, but for creating a
 889 * new file we should try to match whatever "patch" does. I have no idea.
 890 */
 891static int parse_traditional_patch(struct apply_state *state,
 892                                   const char *first,
 893                                   const char *second,
 894                                   struct patch *patch)
 895{
 896        char *name;
 897
 898        first += 4;     /* skip "--- " */
 899        second += 4;    /* skip "+++ " */
 900        if (!state->p_value_known) {
 901                int p, q;
 902                p = guess_p_value(state, first);
 903                q = guess_p_value(state, second);
 904                if (p < 0) p = q;
 905                if (0 <= p && p == q) {
 906                        state->p_value = p;
 907                        state->p_value_known = 1;
 908                }
 909        }
 910        if (is_dev_null(first)) {
 911                patch->is_new = 1;
 912                patch->is_delete = 0;
 913                name = find_name_traditional(state, second, NULL, state->p_value);
 914                patch->new_name = name;
 915        } else if (is_dev_null(second)) {
 916                patch->is_new = 0;
 917                patch->is_delete = 1;
 918                name = find_name_traditional(state, first, NULL, state->p_value);
 919                patch->old_name = name;
 920        } else {
 921                char *first_name;
 922                first_name = find_name_traditional(state, first, NULL, state->p_value);
 923                name = find_name_traditional(state, second, first_name, state->p_value);
 924                free(first_name);
 925                if (has_epoch_timestamp(first)) {
 926                        patch->is_new = 1;
 927                        patch->is_delete = 0;
 928                        patch->new_name = name;
 929                } else if (has_epoch_timestamp(second)) {
 930                        patch->is_new = 0;
 931                        patch->is_delete = 1;
 932                        patch->old_name = name;
 933                } else {
 934                        patch->old_name = name;
 935                        patch->new_name = xstrdup_or_null(name);
 936                }
 937        }
 938        if (!name)
 939                return error(_("unable to find filename in patch at line %d"), state->linenr);
 940
 941        return 0;
 942}
 943
 944static int gitdiff_hdrend(struct apply_state *state,
 945                          const char *line,
 946                          struct patch *patch)
 947{
 948        return 1;
 949}
 950
 951/*
 952 * We're anal about diff header consistency, to make
 953 * sure that we don't end up having strange ambiguous
 954 * patches floating around.
 955 *
 956 * As a result, gitdiff_{old|new}name() will check
 957 * their names against any previous information, just
 958 * to make sure..
 959 */
 960#define DIFF_OLD_NAME 0
 961#define DIFF_NEW_NAME 1
 962
 963static int gitdiff_verify_name(struct apply_state *state,
 964                               const char *line,
 965                               int isnull,
 966                               char **name,
 967                               int side)
 968{
 969        if (!*name && !isnull) {
 970                *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
 971                return 0;
 972        }
 973
 974        if (*name) {
 975                char *another;
 976                if (isnull)
 977                        return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
 978                                     *name, state->linenr);
 979                another = find_name(state, line, NULL, state->p_value, TERM_TAB);
 980                if (!another || strcmp(another, *name)) {
 981                        free(another);
 982                        return error((side == DIFF_NEW_NAME) ?
 983                            _("git apply: bad git-diff - inconsistent new filename on line %d") :
 984                            _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
 985                }
 986                free(another);
 987        } else {
 988                if (!starts_with(line, "/dev/null\n"))
 989                        return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
 990        }
 991
 992        return 0;
 993}
 994
 995static int gitdiff_oldname(struct apply_state *state,
 996                           const char *line,
 997                           struct patch *patch)
 998{
 999        return gitdiff_verify_name(state, line,
1000                                   patch->is_new, &patch->old_name,
1001                                   DIFF_OLD_NAME);
1002}
1003
1004static int gitdiff_newname(struct apply_state *state,
1005                           const char *line,
1006                           struct patch *patch)
1007{
1008        return gitdiff_verify_name(state, line,
1009                                   patch->is_delete, &patch->new_name,
1010                                   DIFF_NEW_NAME);
1011}
1012
1013static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
1014{
1015        char *end;
1016        *mode = strtoul(line, &end, 8);
1017        if (end == line || !isspace(*end))
1018                return error(_("invalid mode on line %d: %s"), linenr, line);
1019        return 0;
1020}
1021
1022static int gitdiff_oldmode(struct apply_state *state,
1023                           const char *line,
1024                           struct patch *patch)
1025{
1026        return parse_mode_line(line, state->linenr, &patch->old_mode);
1027}
1028
1029static int gitdiff_newmode(struct apply_state *state,
1030                           const char *line,
1031                           struct patch *patch)
1032{
1033        return parse_mode_line(line, state->linenr, &patch->new_mode);
1034}
1035
1036static int gitdiff_delete(struct apply_state *state,
1037                          const char *line,
1038                          struct patch *patch)
1039{
1040        patch->is_delete = 1;
1041        free(patch->old_name);
1042        patch->old_name = xstrdup_or_null(patch->def_name);
1043        return gitdiff_oldmode(state, line, patch);
1044}
1045
1046static int gitdiff_newfile(struct apply_state *state,
1047                           const char *line,
1048                           struct patch *patch)
1049{
1050        patch->is_new = 1;
1051        free(patch->new_name);
1052        patch->new_name = xstrdup_or_null(patch->def_name);
1053        return gitdiff_newmode(state, line, patch);
1054}
1055
1056static int gitdiff_copysrc(struct apply_state *state,
1057                           const char *line,
1058                           struct patch *patch)
1059{
1060        patch->is_copy = 1;
1061        free(patch->old_name);
1062        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1063        return 0;
1064}
1065
1066static int gitdiff_copydst(struct apply_state *state,
1067                           const char *line,
1068                           struct patch *patch)
1069{
1070        patch->is_copy = 1;
1071        free(patch->new_name);
1072        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1073        return 0;
1074}
1075
1076static int gitdiff_renamesrc(struct apply_state *state,
1077                             const char *line,
1078                             struct patch *patch)
1079{
1080        patch->is_rename = 1;
1081        free(patch->old_name);
1082        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1083        return 0;
1084}
1085
1086static int gitdiff_renamedst(struct apply_state *state,
1087                             const char *line,
1088                             struct patch *patch)
1089{
1090        patch->is_rename = 1;
1091        free(patch->new_name);
1092        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1093        return 0;
1094}
1095
1096static int gitdiff_similarity(struct apply_state *state,
1097                              const char *line,
1098                              struct patch *patch)
1099{
1100        unsigned long val = strtoul(line, NULL, 10);
1101        if (val <= 100)
1102                patch->score = val;
1103        return 0;
1104}
1105
1106static int gitdiff_dissimilarity(struct apply_state *state,
1107                                 const char *line,
1108                                 struct patch *patch)
1109{
1110        unsigned long val = strtoul(line, NULL, 10);
1111        if (val <= 100)
1112                patch->score = val;
1113        return 0;
1114}
1115
1116static int gitdiff_index(struct apply_state *state,
1117                         const char *line,
1118                         struct patch *patch)
1119{
1120        /*
1121         * index line is N hexadecimal, "..", N hexadecimal,
1122         * and optional space with octal mode.
1123         */
1124        const char *ptr, *eol;
1125        int len;
1126
1127        ptr = strchr(line, '.');
1128        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1129                return 0;
1130        len = ptr - line;
1131        memcpy(patch->old_sha1_prefix, line, len);
1132        patch->old_sha1_prefix[len] = 0;
1133
1134        line = ptr + 2;
1135        ptr = strchr(line, ' ');
1136        eol = strchrnul(line, '\n');
1137
1138        if (!ptr || eol < ptr)
1139                ptr = eol;
1140        len = ptr - line;
1141
1142        if (40 < len)
1143                return 0;
1144        memcpy(patch->new_sha1_prefix, line, len);
1145        patch->new_sha1_prefix[len] = 0;
1146        if (*ptr == ' ')
1147                return gitdiff_oldmode(state, ptr + 1, patch);
1148        return 0;
1149}
1150
1151/*
1152 * This is normal for a diff that doesn't change anything: we'll fall through
1153 * into the next diff. Tell the parser to break out.
1154 */
1155static int gitdiff_unrecognized(struct apply_state *state,
1156                                const char *line,
1157                                struct patch *patch)
1158{
1159        return 1;
1160}
1161
1162/*
1163 * Skip p_value leading components from "line"; as we do not accept
1164 * absolute paths, return NULL in that case.
1165 */
1166static const char *skip_tree_prefix(struct apply_state *state,
1167                                    const char *line,
1168                                    int llen)
1169{
1170        int nslash;
1171        int i;
1172
1173        if (!state->p_value)
1174                return (llen && line[0] == '/') ? NULL : line;
1175
1176        nslash = state->p_value;
1177        for (i = 0; i < llen; i++) {
1178                int ch = line[i];
1179                if (ch == '/' && --nslash <= 0)
1180                        return (i == 0) ? NULL : &line[i + 1];
1181        }
1182        return NULL;
1183}
1184
1185/*
1186 * This is to extract the same name that appears on "diff --git"
1187 * line.  We do not find and return anything if it is a rename
1188 * patch, and it is OK because we will find the name elsewhere.
1189 * We need to reliably find name only when it is mode-change only,
1190 * creation or deletion of an empty file.  In any of these cases,
1191 * both sides are the same name under a/ and b/ respectively.
1192 */
1193static char *git_header_name(struct apply_state *state,
1194                             const char *line,
1195                             int llen)
1196{
1197        const char *name;
1198        const char *second = NULL;
1199        size_t len, line_len;
1200
1201        line += strlen("diff --git ");
1202        llen -= strlen("diff --git ");
1203
1204        if (*line == '"') {
1205                const char *cp;
1206                struct strbuf first = STRBUF_INIT;
1207                struct strbuf sp = STRBUF_INIT;
1208
1209                if (unquote_c_style(&first, line, &second))
1210                        goto free_and_fail1;
1211
1212                /* strip the a/b prefix including trailing slash */
1213                cp = skip_tree_prefix(state, first.buf, first.len);
1214                if (!cp)
1215                        goto free_and_fail1;
1216                strbuf_remove(&first, 0, cp - first.buf);
1217
1218                /*
1219                 * second points at one past closing dq of name.
1220                 * find the second name.
1221                 */
1222                while ((second < line + llen) && isspace(*second))
1223                        second++;
1224
1225                if (line + llen <= second)
1226                        goto free_and_fail1;
1227                if (*second == '"') {
1228                        if (unquote_c_style(&sp, second, NULL))
1229                                goto free_and_fail1;
1230                        cp = skip_tree_prefix(state, sp.buf, sp.len);
1231                        if (!cp)
1232                                goto free_and_fail1;
1233                        /* They must match, otherwise ignore */
1234                        if (strcmp(cp, first.buf))
1235                                goto free_and_fail1;
1236                        strbuf_release(&sp);
1237                        return strbuf_detach(&first, NULL);
1238                }
1239
1240                /* unquoted second */
1241                cp = skip_tree_prefix(state, second, line + llen - second);
1242                if (!cp)
1243                        goto free_and_fail1;
1244                if (line + llen - cp != first.len ||
1245                    memcmp(first.buf, cp, first.len))
1246                        goto free_and_fail1;
1247                return strbuf_detach(&first, NULL);
1248
1249        free_and_fail1:
1250                strbuf_release(&first);
1251                strbuf_release(&sp);
1252                return NULL;
1253        }
1254
1255        /* unquoted first name */
1256        name = skip_tree_prefix(state, line, llen);
1257        if (!name)
1258                return NULL;
1259
1260        /*
1261         * since the first name is unquoted, a dq if exists must be
1262         * the beginning of the second name.
1263         */
1264        for (second = name; second < line + llen; second++) {
1265                if (*second == '"') {
1266                        struct strbuf sp = STRBUF_INIT;
1267                        const char *np;
1268
1269                        if (unquote_c_style(&sp, second, NULL))
1270                                goto free_and_fail2;
1271
1272                        np = skip_tree_prefix(state, sp.buf, sp.len);
1273                        if (!np)
1274                                goto free_and_fail2;
1275
1276                        len = sp.buf + sp.len - np;
1277                        if (len < second - name &&
1278                            !strncmp(np, name, len) &&
1279                            isspace(name[len])) {
1280                                /* Good */
1281                                strbuf_remove(&sp, 0, np - sp.buf);
1282                                return strbuf_detach(&sp, NULL);
1283                        }
1284
1285                free_and_fail2:
1286                        strbuf_release(&sp);
1287                        return NULL;
1288                }
1289        }
1290
1291        /*
1292         * Accept a name only if it shows up twice, exactly the same
1293         * form.
1294         */
1295        second = strchr(name, '\n');
1296        if (!second)
1297                return NULL;
1298        line_len = second - name;
1299        for (len = 0 ; ; len++) {
1300                switch (name[len]) {
1301                default:
1302                        continue;
1303                case '\n':
1304                        return NULL;
1305                case '\t': case ' ':
1306                        /*
1307                         * Is this the separator between the preimage
1308                         * and the postimage pathname?  Again, we are
1309                         * only interested in the case where there is
1310                         * no rename, as this is only to set def_name
1311                         * and a rename patch has the names elsewhere
1312                         * in an unambiguous form.
1313                         */
1314                        if (!name[len + 1])
1315                                return NULL; /* no postimage name */
1316                        second = skip_tree_prefix(state, name + len + 1,
1317                                                  line_len - (len + 1));
1318                        if (!second)
1319                                return NULL;
1320                        /*
1321                         * Does len bytes starting at "name" and "second"
1322                         * (that are separated by one HT or SP we just
1323                         * found) exactly match?
1324                         */
1325                        if (second[len] == '\n' && !strncmp(name, second, len))
1326                                return xmemdupz(name, len);
1327                }
1328        }
1329}
1330
1331static int check_header_line(struct apply_state *state, struct patch *patch)
1332{
1333        int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1334                         (patch->is_rename == 1) + (patch->is_copy == 1);
1335        if (extensions > 1)
1336                return error(_("inconsistent header lines %d and %d"),
1337                             patch->extension_linenr, state->linenr);
1338        if (extensions && !patch->extension_linenr)
1339                patch->extension_linenr = state->linenr;
1340        return 0;
1341}
1342
1343/* Verify that we recognize the lines following a git header */
1344static int parse_git_header(struct apply_state *state,
1345                            const char *line,
1346                            int len,
1347                            unsigned int size,
1348                            struct patch *patch)
1349{
1350        unsigned long offset;
1351
1352        /* A git diff has explicit new/delete information, so we don't guess */
1353        patch->is_new = 0;
1354        patch->is_delete = 0;
1355
1356        /*
1357         * Some things may not have the old name in the
1358         * rest of the headers anywhere (pure mode changes,
1359         * or removing or adding empty files), so we get
1360         * the default name from the header.
1361         */
1362        patch->def_name = git_header_name(state, line, len);
1363        if (patch->def_name && state->root.len) {
1364                char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1365                free(patch->def_name);
1366                patch->def_name = s;
1367        }
1368
1369        line += len;
1370        size -= len;
1371        state->linenr++;
1372        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1373                static const struct opentry {
1374                        const char *str;
1375                        int (*fn)(struct apply_state *, const char *, struct patch *);
1376                } optable[] = {
1377                        { "@@ -", gitdiff_hdrend },
1378                        { "--- ", gitdiff_oldname },
1379                        { "+++ ", gitdiff_newname },
1380                        { "old mode ", gitdiff_oldmode },
1381                        { "new mode ", gitdiff_newmode },
1382                        { "deleted file mode ", gitdiff_delete },
1383                        { "new file mode ", gitdiff_newfile },
1384                        { "copy from ", gitdiff_copysrc },
1385                        { "copy to ", gitdiff_copydst },
1386                        { "rename old ", gitdiff_renamesrc },
1387                        { "rename new ", gitdiff_renamedst },
1388                        { "rename from ", gitdiff_renamesrc },
1389                        { "rename to ", gitdiff_renamedst },
1390                        { "similarity index ", gitdiff_similarity },
1391                        { "dissimilarity index ", gitdiff_dissimilarity },
1392                        { "index ", gitdiff_index },
1393                        { "", gitdiff_unrecognized },
1394                };
1395                int i;
1396
1397                len = linelen(line, size);
1398                if (!len || line[len-1] != '\n')
1399                        break;
1400                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1401                        const struct opentry *p = optable + i;
1402                        int oplen = strlen(p->str);
1403                        int res;
1404                        if (len < oplen || memcmp(p->str, line, oplen))
1405                                continue;
1406                        res = p->fn(state, line + oplen, patch);
1407                        if (res < 0)
1408                                return -1;
1409                        if (check_header_line(state, patch))
1410                                return -1;
1411                        if (res > 0)
1412                                return offset;
1413                        break;
1414                }
1415        }
1416
1417        return offset;
1418}
1419
1420static int parse_num(const char *line, unsigned long *p)
1421{
1422        char *ptr;
1423
1424        if (!isdigit(*line))
1425                return 0;
1426        *p = strtoul(line, &ptr, 10);
1427        return ptr - line;
1428}
1429
1430static int parse_range(const char *line, int len, int offset, const char *expect,
1431                       unsigned long *p1, unsigned long *p2)
1432{
1433        int digits, ex;
1434
1435        if (offset < 0 || offset >= len)
1436                return -1;
1437        line += offset;
1438        len -= offset;
1439
1440        digits = parse_num(line, p1);
1441        if (!digits)
1442                return -1;
1443
1444        offset += digits;
1445        line += digits;
1446        len -= digits;
1447
1448        *p2 = 1;
1449        if (*line == ',') {
1450                digits = parse_num(line+1, p2);
1451                if (!digits)
1452                        return -1;
1453
1454                offset += digits+1;
1455                line += digits+1;
1456                len -= digits+1;
1457        }
1458
1459        ex = strlen(expect);
1460        if (ex > len)
1461                return -1;
1462        if (memcmp(line, expect, ex))
1463                return -1;
1464
1465        return offset + ex;
1466}
1467
1468static void recount_diff(const char *line, int size, struct fragment *fragment)
1469{
1470        int oldlines = 0, newlines = 0, ret = 0;
1471
1472        if (size < 1) {
1473                warning("recount: ignore empty hunk");
1474                return;
1475        }
1476
1477        for (;;) {
1478                int len = linelen(line, size);
1479                size -= len;
1480                line += len;
1481
1482                if (size < 1)
1483                        break;
1484
1485                switch (*line) {
1486                case ' ': case '\n':
1487                        newlines++;
1488                        /* fall through */
1489                case '-':
1490                        oldlines++;
1491                        continue;
1492                case '+':
1493                        newlines++;
1494                        continue;
1495                case '\\':
1496                        continue;
1497                case '@':
1498                        ret = size < 3 || !starts_with(line, "@@ ");
1499                        break;
1500                case 'd':
1501                        ret = size < 5 || !starts_with(line, "diff ");
1502                        break;
1503                default:
1504                        ret = -1;
1505                        break;
1506                }
1507                if (ret) {
1508                        warning(_("recount: unexpected line: %.*s"),
1509                                (int)linelen(line, size), line);
1510                        return;
1511                }
1512                break;
1513        }
1514        fragment->oldlines = oldlines;
1515        fragment->newlines = newlines;
1516}
1517
1518/*
1519 * Parse a unified diff fragment header of the
1520 * form "@@ -a,b +c,d @@"
1521 */
1522static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1523{
1524        int offset;
1525
1526        if (!len || line[len-1] != '\n')
1527                return -1;
1528
1529        /* Figure out the number of lines in a fragment */
1530        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1531        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1532
1533        return offset;
1534}
1535
1536/*
1537 * Find file diff header
1538 *
1539 * Returns:
1540 *  -1 if no header was found
1541 *  -128 in case of error
1542 *   the size of the header in bytes (called "offset") otherwise
1543 */
1544static int find_header(struct apply_state *state,
1545                       const char *line,
1546                       unsigned long size,
1547                       int *hdrsize,
1548                       struct patch *patch)
1549{
1550        unsigned long offset, len;
1551
1552        patch->is_toplevel_relative = 0;
1553        patch->is_rename = patch->is_copy = 0;
1554        patch->is_new = patch->is_delete = -1;
1555        patch->old_mode = patch->new_mode = 0;
1556        patch->old_name = patch->new_name = NULL;
1557        for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1558                unsigned long nextlen;
1559
1560                len = linelen(line, size);
1561                if (!len)
1562                        break;
1563
1564                /* Testing this early allows us to take a few shortcuts.. */
1565                if (len < 6)
1566                        continue;
1567
1568                /*
1569                 * Make sure we don't find any unconnected patch fragments.
1570                 * That's a sign that we didn't find a header, and that a
1571                 * patch has become corrupted/broken up.
1572                 */
1573                if (!memcmp("@@ -", line, 4)) {
1574                        struct fragment dummy;
1575                        if (parse_fragment_header(line, len, &dummy) < 0)
1576                                continue;
1577                        error(_("patch fragment without header at line %d: %.*s"),
1578                                     state->linenr, (int)len-1, line);
1579                        return -128;
1580                }
1581
1582                if (size < len + 6)
1583                        break;
1584
1585                /*
1586                 * Git patch? It might not have a real patch, just a rename
1587                 * or mode change, so we handle that specially
1588                 */
1589                if (!memcmp("diff --git ", line, 11)) {
1590                        int git_hdr_len = parse_git_header(state, line, len, size, patch);
1591                        if (git_hdr_len < 0)
1592                                return -128;
1593                        if (git_hdr_len <= len)
1594                                continue;
1595                        if (!patch->old_name && !patch->new_name) {
1596                                if (!patch->def_name) {
1597                                        error(Q_("git diff header lacks filename information when removing "
1598                                                        "%d leading pathname component (line %d)",
1599                                                        "git diff header lacks filename information when removing "
1600                                                        "%d leading pathname components (line %d)",
1601                                                        state->p_value),
1602                                                     state->p_value, state->linenr);
1603                                        return -128;
1604                                }
1605                                patch->old_name = xstrdup(patch->def_name);
1606                                patch->new_name = xstrdup(patch->def_name);
1607                        }
1608                        if ((!patch->new_name && !patch->is_delete) ||
1609                            (!patch->old_name && !patch->is_new)) {
1610                                error(_("git diff header lacks filename information "
1611                                             "(line %d)"), state->linenr);
1612                                return -128;
1613                        }
1614                        patch->is_toplevel_relative = 1;
1615                        *hdrsize = git_hdr_len;
1616                        return offset;
1617                }
1618
1619                /* --- followed by +++ ? */
1620                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1621                        continue;
1622
1623                /*
1624                 * We only accept unified patches, so we want it to
1625                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1626                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1627                 */
1628                nextlen = linelen(line + len, size - len);
1629                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1630                        continue;
1631
1632                /* Ok, we'll consider it a patch */
1633                if (parse_traditional_patch(state, line, line+len, patch))
1634                        return -128;
1635                *hdrsize = len + nextlen;
1636                state->linenr += 2;
1637                return offset;
1638        }
1639        return -1;
1640}
1641
1642static void record_ws_error(struct apply_state *state,
1643                            unsigned result,
1644                            const char *line,
1645                            int len,
1646                            int linenr)
1647{
1648        char *err;
1649
1650        if (!result)
1651                return;
1652
1653        state->whitespace_error++;
1654        if (state->squelch_whitespace_errors &&
1655            state->squelch_whitespace_errors < state->whitespace_error)
1656                return;
1657
1658        err = whitespace_error_string(result);
1659        if (state->apply_verbosity > verbosity_silent)
1660                fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1661                        state->patch_input_file, linenr, err, len, line);
1662        free(err);
1663}
1664
1665static void check_whitespace(struct apply_state *state,
1666                             const char *line,
1667                             int len,
1668                             unsigned ws_rule)
1669{
1670        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1671
1672        record_ws_error(state, result, line + 1, len - 2, state->linenr);
1673}
1674
1675/*
1676 * Parse a unified diff. Note that this really needs to parse each
1677 * fragment separately, since the only way to know the difference
1678 * between a "---" that is part of a patch, and a "---" that starts
1679 * the next patch is to look at the line counts..
1680 */
1681static int parse_fragment(struct apply_state *state,
1682                          const char *line,
1683                          unsigned long size,
1684                          struct patch *patch,
1685                          struct fragment *fragment)
1686{
1687        int added, deleted;
1688        int len = linelen(line, size), offset;
1689        unsigned long oldlines, newlines;
1690        unsigned long leading, trailing;
1691
1692        offset = parse_fragment_header(line, len, fragment);
1693        if (offset < 0)
1694                return -1;
1695        if (offset > 0 && patch->recount)
1696                recount_diff(line + offset, size - offset, fragment);
1697        oldlines = fragment->oldlines;
1698        newlines = fragment->newlines;
1699        leading = 0;
1700        trailing = 0;
1701
1702        /* Parse the thing.. */
1703        line += len;
1704        size -= len;
1705        state->linenr++;
1706        added = deleted = 0;
1707        for (offset = len;
1708             0 < size;
1709             offset += len, size -= len, line += len, state->linenr++) {
1710                if (!oldlines && !newlines)
1711                        break;
1712                len = linelen(line, size);
1713                if (!len || line[len-1] != '\n')
1714                        return -1;
1715                switch (*line) {
1716                default:
1717                        return -1;
1718                case '\n': /* newer GNU diff, an empty context line */
1719                case ' ':
1720                        oldlines--;
1721                        newlines--;
1722                        if (!deleted && !added)
1723                                leading++;
1724                        trailing++;
1725                        if (!state->apply_in_reverse &&
1726                            state->ws_error_action == correct_ws_error)
1727                                check_whitespace(state, line, len, patch->ws_rule);
1728                        break;
1729                case '-':
1730                        if (state->apply_in_reverse &&
1731                            state->ws_error_action != nowarn_ws_error)
1732                                check_whitespace(state, line, len, patch->ws_rule);
1733                        deleted++;
1734                        oldlines--;
1735                        trailing = 0;
1736                        break;
1737                case '+':
1738                        if (!state->apply_in_reverse &&
1739                            state->ws_error_action != nowarn_ws_error)
1740                                check_whitespace(state, line, len, patch->ws_rule);
1741                        added++;
1742                        newlines--;
1743                        trailing = 0;
1744                        break;
1745
1746                /*
1747                 * We allow "\ No newline at end of file". Depending
1748                 * on locale settings when the patch was produced we
1749                 * don't know what this line looks like. The only
1750                 * thing we do know is that it begins with "\ ".
1751                 * Checking for 12 is just for sanity check -- any
1752                 * l10n of "\ No newline..." is at least that long.
1753                 */
1754                case '\\':
1755                        if (len < 12 || memcmp(line, "\\ ", 2))
1756                                return -1;
1757                        break;
1758                }
1759        }
1760        if (oldlines || newlines)
1761                return -1;
1762        if (!deleted && !added)
1763                return -1;
1764
1765        fragment->leading = leading;
1766        fragment->trailing = trailing;
1767
1768        /*
1769         * If a fragment ends with an incomplete line, we failed to include
1770         * it in the above loop because we hit oldlines == newlines == 0
1771         * before seeing it.
1772         */
1773        if (12 < size && !memcmp(line, "\\ ", 2))
1774                offset += linelen(line, size);
1775
1776        patch->lines_added += added;
1777        patch->lines_deleted += deleted;
1778
1779        if (0 < patch->is_new && oldlines)
1780                return error(_("new file depends on old contents"));
1781        if (0 < patch->is_delete && newlines)
1782                return error(_("deleted file still has contents"));
1783        return offset;
1784}
1785
1786/*
1787 * We have seen "diff --git a/... b/..." header (or a traditional patch
1788 * header).  Read hunks that belong to this patch into fragments and hang
1789 * them to the given patch structure.
1790 *
1791 * The (fragment->patch, fragment->size) pair points into the memory given
1792 * by the caller, not a copy, when we return.
1793 *
1794 * Returns:
1795 *   -1 in case of error,
1796 *   the number of bytes in the patch otherwise.
1797 */
1798static int parse_single_patch(struct apply_state *state,
1799                              const char *line,
1800                              unsigned long size,
1801                              struct patch *patch)
1802{
1803        unsigned long offset = 0;
1804        unsigned long oldlines = 0, newlines = 0, context = 0;
1805        struct fragment **fragp = &patch->fragments;
1806
1807        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1808                struct fragment *fragment;
1809                int len;
1810
1811                fragment = xcalloc(1, sizeof(*fragment));
1812                fragment->linenr = state->linenr;
1813                len = parse_fragment(state, line, size, patch, fragment);
1814                if (len <= 0) {
1815                        free(fragment);
1816                        return error(_("corrupt patch at line %d"), state->linenr);
1817                }
1818                fragment->patch = line;
1819                fragment->size = len;
1820                oldlines += fragment->oldlines;
1821                newlines += fragment->newlines;
1822                context += fragment->leading + fragment->trailing;
1823
1824                *fragp = fragment;
1825                fragp = &fragment->next;
1826
1827                offset += len;
1828                line += len;
1829                size -= len;
1830        }
1831
1832        /*
1833         * If something was removed (i.e. we have old-lines) it cannot
1834         * be creation, and if something was added it cannot be
1835         * deletion.  However, the reverse is not true; --unified=0
1836         * patches that only add are not necessarily creation even
1837         * though they do not have any old lines, and ones that only
1838         * delete are not necessarily deletion.
1839         *
1840         * Unfortunately, a real creation/deletion patch do _not_ have
1841         * any context line by definition, so we cannot safely tell it
1842         * apart with --unified=0 insanity.  At least if the patch has
1843         * more than one hunk it is not creation or deletion.
1844         */
1845        if (patch->is_new < 0 &&
1846            (oldlines || (patch->fragments && patch->fragments->next)))
1847                patch->is_new = 0;
1848        if (patch->is_delete < 0 &&
1849            (newlines || (patch->fragments && patch->fragments->next)))
1850                patch->is_delete = 0;
1851
1852        if (0 < patch->is_new && oldlines)
1853                return error(_("new file %s depends on old contents"), patch->new_name);
1854        if (0 < patch->is_delete && newlines)
1855                return error(_("deleted file %s still has contents"), patch->old_name);
1856        if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1857                fprintf_ln(stderr,
1858                           _("** warning: "
1859                             "file %s becomes empty but is not deleted"),
1860                           patch->new_name);
1861
1862        return offset;
1863}
1864
1865static inline int metadata_changes(struct patch *patch)
1866{
1867        return  patch->is_rename > 0 ||
1868                patch->is_copy > 0 ||
1869                patch->is_new > 0 ||
1870                patch->is_delete ||
1871                (patch->old_mode && patch->new_mode &&
1872                 patch->old_mode != patch->new_mode);
1873}
1874
1875static char *inflate_it(const void *data, unsigned long size,
1876                        unsigned long inflated_size)
1877{
1878        git_zstream stream;
1879        void *out;
1880        int st;
1881
1882        memset(&stream, 0, sizeof(stream));
1883
1884        stream.next_in = (unsigned char *)data;
1885        stream.avail_in = size;
1886        stream.next_out = out = xmalloc(inflated_size);
1887        stream.avail_out = inflated_size;
1888        git_inflate_init(&stream);
1889        st = git_inflate(&stream, Z_FINISH);
1890        git_inflate_end(&stream);
1891        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1892                free(out);
1893                return NULL;
1894        }
1895        return out;
1896}
1897
1898/*
1899 * Read a binary hunk and return a new fragment; fragment->patch
1900 * points at an allocated memory that the caller must free, so
1901 * it is marked as "->free_patch = 1".
1902 */
1903static struct fragment *parse_binary_hunk(struct apply_state *state,
1904                                          char **buf_p,
1905                                          unsigned long *sz_p,
1906                                          int *status_p,
1907                                          int *used_p)
1908{
1909        /*
1910         * Expect a line that begins with binary patch method ("literal"
1911         * or "delta"), followed by the length of data before deflating.
1912         * a sequence of 'length-byte' followed by base-85 encoded data
1913         * should follow, terminated by a newline.
1914         *
1915         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1916         * and we would limit the patch line to 66 characters,
1917         * so one line can fit up to 13 groups that would decode
1918         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1919         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1920         */
1921        int llen, used;
1922        unsigned long size = *sz_p;
1923        char *buffer = *buf_p;
1924        int patch_method;
1925        unsigned long origlen;
1926        char *data = NULL;
1927        int hunk_size = 0;
1928        struct fragment *frag;
1929
1930        llen = linelen(buffer, size);
1931        used = llen;
1932
1933        *status_p = 0;
1934
1935        if (starts_with(buffer, "delta ")) {
1936                patch_method = BINARY_DELTA_DEFLATED;
1937                origlen = strtoul(buffer + 6, NULL, 10);
1938        }
1939        else if (starts_with(buffer, "literal ")) {
1940                patch_method = BINARY_LITERAL_DEFLATED;
1941                origlen = strtoul(buffer + 8, NULL, 10);
1942        }
1943        else
1944                return NULL;
1945
1946        state->linenr++;
1947        buffer += llen;
1948        while (1) {
1949                int byte_length, max_byte_length, newsize;
1950                llen = linelen(buffer, size);
1951                used += llen;
1952                state->linenr++;
1953                if (llen == 1) {
1954                        /* consume the blank line */
1955                        buffer++;
1956                        size--;
1957                        break;
1958                }
1959                /*
1960                 * Minimum line is "A00000\n" which is 7-byte long,
1961                 * and the line length must be multiple of 5 plus 2.
1962                 */
1963                if ((llen < 7) || (llen-2) % 5)
1964                        goto corrupt;
1965                max_byte_length = (llen - 2) / 5 * 4;
1966                byte_length = *buffer;
1967                if ('A' <= byte_length && byte_length <= 'Z')
1968                        byte_length = byte_length - 'A' + 1;
1969                else if ('a' <= byte_length && byte_length <= 'z')
1970                        byte_length = byte_length - 'a' + 27;
1971                else
1972                        goto corrupt;
1973                /* if the input length was not multiple of 4, we would
1974                 * have filler at the end but the filler should never
1975                 * exceed 3 bytes
1976                 */
1977                if (max_byte_length < byte_length ||
1978                    byte_length <= max_byte_length - 4)
1979                        goto corrupt;
1980                newsize = hunk_size + byte_length;
1981                data = xrealloc(data, newsize);
1982                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1983                        goto corrupt;
1984                hunk_size = newsize;
1985                buffer += llen;
1986                size -= llen;
1987        }
1988
1989        frag = xcalloc(1, sizeof(*frag));
1990        frag->patch = inflate_it(data, hunk_size, origlen);
1991        frag->free_patch = 1;
1992        if (!frag->patch)
1993                goto corrupt;
1994        free(data);
1995        frag->size = origlen;
1996        *buf_p = buffer;
1997        *sz_p = size;
1998        *used_p = used;
1999        frag->binary_patch_method = patch_method;
2000        return frag;
2001
2002 corrupt:
2003        free(data);
2004        *status_p = -1;
2005        error(_("corrupt binary patch at line %d: %.*s"),
2006              state->linenr-1, llen-1, buffer);
2007        return NULL;
2008}
2009
2010/*
2011 * Returns:
2012 *   -1 in case of error,
2013 *   the length of the parsed binary patch otherwise
2014 */
2015static int parse_binary(struct apply_state *state,
2016                        char *buffer,
2017                        unsigned long size,
2018                        struct patch *patch)
2019{
2020        /*
2021         * We have read "GIT binary patch\n"; what follows is a line
2022         * that says the patch method (currently, either "literal" or
2023         * "delta") and the length of data before deflating; a
2024         * sequence of 'length-byte' followed by base-85 encoded data
2025         * follows.
2026         *
2027         * When a binary patch is reversible, there is another binary
2028         * hunk in the same format, starting with patch method (either
2029         * "literal" or "delta") with the length of data, and a sequence
2030         * of length-byte + base-85 encoded data, terminated with another
2031         * empty line.  This data, when applied to the postimage, produces
2032         * the preimage.
2033         */
2034        struct fragment *forward;
2035        struct fragment *reverse;
2036        int status;
2037        int used, used_1;
2038
2039        forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2040        if (!forward && !status)
2041                /* there has to be one hunk (forward hunk) */
2042                return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2043        if (status)
2044                /* otherwise we already gave an error message */
2045                return status;
2046
2047        reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2048        if (reverse)
2049                used += used_1;
2050        else if (status) {
2051                /*
2052                 * Not having reverse hunk is not an error, but having
2053                 * a corrupt reverse hunk is.
2054                 */
2055                free((void*) forward->patch);
2056                free(forward);
2057                return status;
2058        }
2059        forward->next = reverse;
2060        patch->fragments = forward;
2061        patch->is_binary = 1;
2062        return used;
2063}
2064
2065static void prefix_one(struct apply_state *state, char **name)
2066{
2067        char *old_name = *name;
2068        if (!old_name)
2069                return;
2070        *name = prefix_filename(state->prefix, *name);
2071        free(old_name);
2072}
2073
2074static void prefix_patch(struct apply_state *state, struct patch *p)
2075{
2076        if (!state->prefix || p->is_toplevel_relative)
2077                return;
2078        prefix_one(state, &p->new_name);
2079        prefix_one(state, &p->old_name);
2080}
2081
2082/*
2083 * include/exclude
2084 */
2085
2086static void add_name_limit(struct apply_state *state,
2087                           const char *name,
2088                           int exclude)
2089{
2090        struct string_list_item *it;
2091
2092        it = string_list_append(&state->limit_by_name, name);
2093        it->util = exclude ? NULL : (void *) 1;
2094}
2095
2096static int use_patch(struct apply_state *state, struct patch *p)
2097{
2098        const char *pathname = p->new_name ? p->new_name : p->old_name;
2099        int i;
2100
2101        /* Paths outside are not touched regardless of "--include" */
2102        if (0 < state->prefix_length) {
2103                int pathlen = strlen(pathname);
2104                if (pathlen <= state->prefix_length ||
2105                    memcmp(state->prefix, pathname, state->prefix_length))
2106                        return 0;
2107        }
2108
2109        /* See if it matches any of exclude/include rule */
2110        for (i = 0; i < state->limit_by_name.nr; i++) {
2111                struct string_list_item *it = &state->limit_by_name.items[i];
2112                if (!wildmatch(it->string, pathname, 0, NULL))
2113                        return (it->util != NULL);
2114        }
2115
2116        /*
2117         * If we had any include, a path that does not match any rule is
2118         * not used.  Otherwise, we saw bunch of exclude rules (or none)
2119         * and such a path is used.
2120         */
2121        return !state->has_include;
2122}
2123
2124/*
2125 * Read the patch text in "buffer" that extends for "size" bytes; stop
2126 * reading after seeing a single patch (i.e. changes to a single file).
2127 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2128 *
2129 * Returns:
2130 *   -1 if no header was found or parse_binary() failed,
2131 *   -128 on another error,
2132 *   the number of bytes consumed otherwise,
2133 *     so that the caller can call us again for the next patch.
2134 */
2135static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2136{
2137        int hdrsize, patchsize;
2138        int offset = find_header(state, buffer, size, &hdrsize, patch);
2139
2140        if (offset < 0)
2141                return offset;
2142
2143        prefix_patch(state, patch);
2144
2145        if (!use_patch(state, patch))
2146                patch->ws_rule = 0;
2147        else
2148                patch->ws_rule = whitespace_rule(patch->new_name
2149                                                 ? patch->new_name
2150                                                 : patch->old_name);
2151
2152        patchsize = parse_single_patch(state,
2153                                       buffer + offset + hdrsize,
2154                                       size - offset - hdrsize,
2155                                       patch);
2156
2157        if (patchsize < 0)
2158                return -128;
2159
2160        if (!patchsize) {
2161                static const char git_binary[] = "GIT binary patch\n";
2162                int hd = hdrsize + offset;
2163                unsigned long llen = linelen(buffer + hd, size - hd);
2164
2165                if (llen == sizeof(git_binary) - 1 &&
2166                    !memcmp(git_binary, buffer + hd, llen)) {
2167                        int used;
2168                        state->linenr++;
2169                        used = parse_binary(state, buffer + hd + llen,
2170                                            size - hd - llen, patch);
2171                        if (used < 0)
2172                                return -1;
2173                        if (used)
2174                                patchsize = used + llen;
2175                        else
2176                                patchsize = 0;
2177                }
2178                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2179                        static const char *binhdr[] = {
2180                                "Binary files ",
2181                                "Files ",
2182                                NULL,
2183                        };
2184                        int i;
2185                        for (i = 0; binhdr[i]; i++) {
2186                                int len = strlen(binhdr[i]);
2187                                if (len < size - hd &&
2188                                    !memcmp(binhdr[i], buffer + hd, len)) {
2189                                        state->linenr++;
2190                                        patch->is_binary = 1;
2191                                        patchsize = llen;
2192                                        break;
2193                                }
2194                        }
2195                }
2196
2197                /* Empty patch cannot be applied if it is a text patch
2198                 * without metadata change.  A binary patch appears
2199                 * empty to us here.
2200                 */
2201                if ((state->apply || state->check) &&
2202                    (!patch->is_binary && !metadata_changes(patch))) {
2203                        error(_("patch with only garbage at line %d"), state->linenr);
2204                        return -128;
2205                }
2206        }
2207
2208        return offset + hdrsize + patchsize;
2209}
2210
2211static void reverse_patches(struct patch *p)
2212{
2213        for (; p; p = p->next) {
2214                struct fragment *frag = p->fragments;
2215
2216                SWAP(p->new_name, p->old_name);
2217                SWAP(p->new_mode, p->old_mode);
2218                SWAP(p->is_new, p->is_delete);
2219                SWAP(p->lines_added, p->lines_deleted);
2220                SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2221
2222                for (; frag; frag = frag->next) {
2223                        SWAP(frag->newpos, frag->oldpos);
2224                        SWAP(frag->newlines, frag->oldlines);
2225                }
2226        }
2227}
2228
2229static const char pluses[] =
2230"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2231static const char minuses[]=
2232"----------------------------------------------------------------------";
2233
2234static void show_stats(struct apply_state *state, struct patch *patch)
2235{
2236        struct strbuf qname = STRBUF_INIT;
2237        char *cp = patch->new_name ? patch->new_name : patch->old_name;
2238        int max, add, del;
2239
2240        quote_c_style(cp, &qname, NULL, 0);
2241
2242        /*
2243         * "scale" the filename
2244         */
2245        max = state->max_len;
2246        if (max > 50)
2247                max = 50;
2248
2249        if (qname.len > max) {
2250                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2251                if (!cp)
2252                        cp = qname.buf + qname.len + 3 - max;
2253                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2254        }
2255
2256        if (patch->is_binary) {
2257                printf(" %-*s |  Bin\n", max, qname.buf);
2258                strbuf_release(&qname);
2259                return;
2260        }
2261
2262        printf(" %-*s |", max, qname.buf);
2263        strbuf_release(&qname);
2264
2265        /*
2266         * scale the add/delete
2267         */
2268        max = max + state->max_change > 70 ? 70 - max : state->max_change;
2269        add = patch->lines_added;
2270        del = patch->lines_deleted;
2271
2272        if (state->max_change > 0) {
2273                int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2274                add = (add * max + state->max_change / 2) / state->max_change;
2275                del = total - add;
2276        }
2277        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2278                add, pluses, del, minuses);
2279}
2280
2281static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2282{
2283        switch (st->st_mode & S_IFMT) {
2284        case S_IFLNK:
2285                if (strbuf_readlink(buf, path, st->st_size) < 0)
2286                        return error(_("unable to read symlink %s"), path);
2287                return 0;
2288        case S_IFREG:
2289                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2290                        return error(_("unable to open or read %s"), path);
2291                convert_to_git(path, buf->buf, buf->len, buf, 0);
2292                return 0;
2293        default:
2294                return -1;
2295        }
2296}
2297
2298/*
2299 * Update the preimage, and the common lines in postimage,
2300 * from buffer buf of length len. If postlen is 0 the postimage
2301 * is updated in place, otherwise it's updated on a new buffer
2302 * of length postlen
2303 */
2304
2305static void update_pre_post_images(struct image *preimage,
2306                                   struct image *postimage,
2307                                   char *buf,
2308                                   size_t len, size_t postlen)
2309{
2310        int i, ctx, reduced;
2311        char *new, *old, *fixed;
2312        struct image fixed_preimage;
2313
2314        /*
2315         * Update the preimage with whitespace fixes.  Note that we
2316         * are not losing preimage->buf -- apply_one_fragment() will
2317         * free "oldlines".
2318         */
2319        prepare_image(&fixed_preimage, buf, len, 1);
2320        assert(postlen
2321               ? fixed_preimage.nr == preimage->nr
2322               : fixed_preimage.nr <= preimage->nr);
2323        for (i = 0; i < fixed_preimage.nr; i++)
2324                fixed_preimage.line[i].flag = preimage->line[i].flag;
2325        free(preimage->line_allocated);
2326        *preimage = fixed_preimage;
2327
2328        /*
2329         * Adjust the common context lines in postimage. This can be
2330         * done in-place when we are shrinking it with whitespace
2331         * fixing, but needs a new buffer when ignoring whitespace or
2332         * expanding leading tabs to spaces.
2333         *
2334         * We trust the caller to tell us if the update can be done
2335         * in place (postlen==0) or not.
2336         */
2337        old = postimage->buf;
2338        if (postlen)
2339                new = postimage->buf = xmalloc(postlen);
2340        else
2341                new = old;
2342        fixed = preimage->buf;
2343
2344        for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2345                size_t l_len = postimage->line[i].len;
2346                if (!(postimage->line[i].flag & LINE_COMMON)) {
2347                        /* an added line -- no counterparts in preimage */
2348                        memmove(new, old, l_len);
2349                        old += l_len;
2350                        new += l_len;
2351                        continue;
2352                }
2353
2354                /* a common context -- skip it in the original postimage */
2355                old += l_len;
2356
2357                /* and find the corresponding one in the fixed preimage */
2358                while (ctx < preimage->nr &&
2359                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2360                        fixed += preimage->line[ctx].len;
2361                        ctx++;
2362                }
2363
2364                /*
2365                 * preimage is expected to run out, if the caller
2366                 * fixed addition of trailing blank lines.
2367                 */
2368                if (preimage->nr <= ctx) {
2369                        reduced++;
2370                        continue;
2371                }
2372
2373                /* and copy it in, while fixing the line length */
2374                l_len = preimage->line[ctx].len;
2375                memcpy(new, fixed, l_len);
2376                new += l_len;
2377                fixed += l_len;
2378                postimage->line[i].len = l_len;
2379                ctx++;
2380        }
2381
2382        if (postlen
2383            ? postlen < new - postimage->buf
2384            : postimage->len < new - postimage->buf)
2385                die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2386                    (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2387
2388        /* Fix the length of the whole thing */
2389        postimage->len = new - postimage->buf;
2390        postimage->nr -= reduced;
2391}
2392
2393static int line_by_line_fuzzy_match(struct image *img,
2394                                    struct image *preimage,
2395                                    struct image *postimage,
2396                                    unsigned long try,
2397                                    int try_lno,
2398                                    int preimage_limit)
2399{
2400        int i;
2401        size_t imgoff = 0;
2402        size_t preoff = 0;
2403        size_t postlen = postimage->len;
2404        size_t extra_chars;
2405        char *buf;
2406        char *preimage_eof;
2407        char *preimage_end;
2408        struct strbuf fixed;
2409        char *fixed_buf;
2410        size_t fixed_len;
2411
2412        for (i = 0; i < preimage_limit; i++) {
2413                size_t prelen = preimage->line[i].len;
2414                size_t imglen = img->line[try_lno+i].len;
2415
2416                if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2417                                      preimage->buf + preoff, prelen))
2418                        return 0;
2419                if (preimage->line[i].flag & LINE_COMMON)
2420                        postlen += imglen - prelen;
2421                imgoff += imglen;
2422                preoff += prelen;
2423        }
2424
2425        /*
2426         * Ok, the preimage matches with whitespace fuzz.
2427         *
2428         * imgoff now holds the true length of the target that
2429         * matches the preimage before the end of the file.
2430         *
2431         * Count the number of characters in the preimage that fall
2432         * beyond the end of the file and make sure that all of them
2433         * are whitespace characters. (This can only happen if
2434         * we are removing blank lines at the end of the file.)
2435         */
2436        buf = preimage_eof = preimage->buf + preoff;
2437        for ( ; i < preimage->nr; i++)
2438                preoff += preimage->line[i].len;
2439        preimage_end = preimage->buf + preoff;
2440        for ( ; buf < preimage_end; buf++)
2441                if (!isspace(*buf))
2442                        return 0;
2443
2444        /*
2445         * Update the preimage and the common postimage context
2446         * lines to use the same whitespace as the target.
2447         * If whitespace is missing in the target (i.e.
2448         * if the preimage extends beyond the end of the file),
2449         * use the whitespace from the preimage.
2450         */
2451        extra_chars = preimage_end - preimage_eof;
2452        strbuf_init(&fixed, imgoff + extra_chars);
2453        strbuf_add(&fixed, img->buf + try, imgoff);
2454        strbuf_add(&fixed, preimage_eof, extra_chars);
2455        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2456        update_pre_post_images(preimage, postimage,
2457                               fixed_buf, fixed_len, postlen);
2458        return 1;
2459}
2460
2461static int match_fragment(struct apply_state *state,
2462                          struct image *img,
2463                          struct image *preimage,
2464                          struct image *postimage,
2465                          unsigned long try,
2466                          int try_lno,
2467                          unsigned ws_rule,
2468                          int match_beginning, int match_end)
2469{
2470        int i;
2471        char *fixed_buf, *buf, *orig, *target;
2472        struct strbuf fixed;
2473        size_t fixed_len, postlen;
2474        int preimage_limit;
2475
2476        if (preimage->nr + try_lno <= img->nr) {
2477                /*
2478                 * The hunk falls within the boundaries of img.
2479                 */
2480                preimage_limit = preimage->nr;
2481                if (match_end && (preimage->nr + try_lno != img->nr))
2482                        return 0;
2483        } else if (state->ws_error_action == correct_ws_error &&
2484                   (ws_rule & WS_BLANK_AT_EOF)) {
2485                /*
2486                 * This hunk extends beyond the end of img, and we are
2487                 * removing blank lines at the end of the file.  This
2488                 * many lines from the beginning of the preimage must
2489                 * match with img, and the remainder of the preimage
2490                 * must be blank.
2491                 */
2492                preimage_limit = img->nr - try_lno;
2493        } else {
2494                /*
2495                 * The hunk extends beyond the end of the img and
2496                 * we are not removing blanks at the end, so we
2497                 * should reject the hunk at this position.
2498                 */
2499                return 0;
2500        }
2501
2502        if (match_beginning && try_lno)
2503                return 0;
2504
2505        /* Quick hash check */
2506        for (i = 0; i < preimage_limit; i++)
2507                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2508                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2509                        return 0;
2510
2511        if (preimage_limit == preimage->nr) {
2512                /*
2513                 * Do we have an exact match?  If we were told to match
2514                 * at the end, size must be exactly at try+fragsize,
2515                 * otherwise try+fragsize must be still within the preimage,
2516                 * and either case, the old piece should match the preimage
2517                 * exactly.
2518                 */
2519                if ((match_end
2520                     ? (try + preimage->len == img->len)
2521                     : (try + preimage->len <= img->len)) &&
2522                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2523                        return 1;
2524        } else {
2525                /*
2526                 * The preimage extends beyond the end of img, so
2527                 * there cannot be an exact match.
2528                 *
2529                 * There must be one non-blank context line that match
2530                 * a line before the end of img.
2531                 */
2532                char *buf_end;
2533
2534                buf = preimage->buf;
2535                buf_end = buf;
2536                for (i = 0; i < preimage_limit; i++)
2537                        buf_end += preimage->line[i].len;
2538
2539                for ( ; buf < buf_end; buf++)
2540                        if (!isspace(*buf))
2541                                break;
2542                if (buf == buf_end)
2543                        return 0;
2544        }
2545
2546        /*
2547         * No exact match. If we are ignoring whitespace, run a line-by-line
2548         * fuzzy matching. We collect all the line length information because
2549         * we need it to adjust whitespace if we match.
2550         */
2551        if (state->ws_ignore_action == ignore_ws_change)
2552                return line_by_line_fuzzy_match(img, preimage, postimage,
2553                                                try, try_lno, preimage_limit);
2554
2555        if (state->ws_error_action != correct_ws_error)
2556                return 0;
2557
2558        /*
2559         * The hunk does not apply byte-by-byte, but the hash says
2560         * it might with whitespace fuzz. We weren't asked to
2561         * ignore whitespace, we were asked to correct whitespace
2562         * errors, so let's try matching after whitespace correction.
2563         *
2564         * While checking the preimage against the target, whitespace
2565         * errors in both fixed, we count how large the corresponding
2566         * postimage needs to be.  The postimage prepared by
2567         * apply_one_fragment() has whitespace errors fixed on added
2568         * lines already, but the common lines were propagated as-is,
2569         * which may become longer when their whitespace errors are
2570         * fixed.
2571         */
2572
2573        /* First count added lines in postimage */
2574        postlen = 0;
2575        for (i = 0; i < postimage->nr; i++) {
2576                if (!(postimage->line[i].flag & LINE_COMMON))
2577                        postlen += postimage->line[i].len;
2578        }
2579
2580        /*
2581         * The preimage may extend beyond the end of the file,
2582         * but in this loop we will only handle the part of the
2583         * preimage that falls within the file.
2584         */
2585        strbuf_init(&fixed, preimage->len + 1);
2586        orig = preimage->buf;
2587        target = img->buf + try;
2588        for (i = 0; i < preimage_limit; i++) {
2589                size_t oldlen = preimage->line[i].len;
2590                size_t tgtlen = img->line[try_lno + i].len;
2591                size_t fixstart = fixed.len;
2592                struct strbuf tgtfix;
2593                int match;
2594
2595                /* Try fixing the line in the preimage */
2596                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2597
2598                /* Try fixing the line in the target */
2599                strbuf_init(&tgtfix, tgtlen);
2600                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2601
2602                /*
2603                 * If they match, either the preimage was based on
2604                 * a version before our tree fixed whitespace breakage,
2605                 * or we are lacking a whitespace-fix patch the tree
2606                 * the preimage was based on already had (i.e. target
2607                 * has whitespace breakage, the preimage doesn't).
2608                 * In either case, we are fixing the whitespace breakages
2609                 * so we might as well take the fix together with their
2610                 * real change.
2611                 */
2612                match = (tgtfix.len == fixed.len - fixstart &&
2613                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2614                                             fixed.len - fixstart));
2615
2616                /* Add the length if this is common with the postimage */
2617                if (preimage->line[i].flag & LINE_COMMON)
2618                        postlen += tgtfix.len;
2619
2620                strbuf_release(&tgtfix);
2621                if (!match)
2622                        goto unmatch_exit;
2623
2624                orig += oldlen;
2625                target += tgtlen;
2626        }
2627
2628
2629        /*
2630         * Now handle the lines in the preimage that falls beyond the
2631         * end of the file (if any). They will only match if they are
2632         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2633         * false).
2634         */
2635        for ( ; i < preimage->nr; i++) {
2636                size_t fixstart = fixed.len; /* start of the fixed preimage */
2637                size_t oldlen = preimage->line[i].len;
2638                int j;
2639
2640                /* Try fixing the line in the preimage */
2641                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2642
2643                for (j = fixstart; j < fixed.len; j++)
2644                        if (!isspace(fixed.buf[j]))
2645                                goto unmatch_exit;
2646
2647                orig += oldlen;
2648        }
2649
2650        /*
2651         * Yes, the preimage is based on an older version that still
2652         * has whitespace breakages unfixed, and fixing them makes the
2653         * hunk match.  Update the context lines in the postimage.
2654         */
2655        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2656        if (postlen < postimage->len)
2657                postlen = 0;
2658        update_pre_post_images(preimage, postimage,
2659                               fixed_buf, fixed_len, postlen);
2660        return 1;
2661
2662 unmatch_exit:
2663        strbuf_release(&fixed);
2664        return 0;
2665}
2666
2667static int find_pos(struct apply_state *state,
2668                    struct image *img,
2669                    struct image *preimage,
2670                    struct image *postimage,
2671                    int line,
2672                    unsigned ws_rule,
2673                    int match_beginning, int match_end)
2674{
2675        int i;
2676        unsigned long backwards, forwards, try;
2677        int backwards_lno, forwards_lno, try_lno;
2678
2679        /*
2680         * If match_beginning or match_end is specified, there is no
2681         * point starting from a wrong line that will never match and
2682         * wander around and wait for a match at the specified end.
2683         */
2684        if (match_beginning)
2685                line = 0;
2686        else if (match_end)
2687                line = img->nr - preimage->nr;
2688
2689        /*
2690         * Because the comparison is unsigned, the following test
2691         * will also take care of a negative line number that can
2692         * result when match_end and preimage is larger than the target.
2693         */
2694        if ((size_t) line > img->nr)
2695                line = img->nr;
2696
2697        try = 0;
2698        for (i = 0; i < line; i++)
2699                try += img->line[i].len;
2700
2701        /*
2702         * There's probably some smart way to do this, but I'll leave
2703         * that to the smart and beautiful people. I'm simple and stupid.
2704         */
2705        backwards = try;
2706        backwards_lno = line;
2707        forwards = try;
2708        forwards_lno = line;
2709        try_lno = line;
2710
2711        for (i = 0; ; i++) {
2712                if (match_fragment(state, img, preimage, postimage,
2713                                   try, try_lno, ws_rule,
2714                                   match_beginning, match_end))
2715                        return try_lno;
2716
2717        again:
2718                if (backwards_lno == 0 && forwards_lno == img->nr)
2719                        break;
2720
2721                if (i & 1) {
2722                        if (backwards_lno == 0) {
2723                                i++;
2724                                goto again;
2725                        }
2726                        backwards_lno--;
2727                        backwards -= img->line[backwards_lno].len;
2728                        try = backwards;
2729                        try_lno = backwards_lno;
2730                } else {
2731                        if (forwards_lno == img->nr) {
2732                                i++;
2733                                goto again;
2734                        }
2735                        forwards += img->line[forwards_lno].len;
2736                        forwards_lno++;
2737                        try = forwards;
2738                        try_lno = forwards_lno;
2739                }
2740
2741        }
2742        return -1;
2743}
2744
2745static void remove_first_line(struct image *img)
2746{
2747        img->buf += img->line[0].len;
2748        img->len -= img->line[0].len;
2749        img->line++;
2750        img->nr--;
2751}
2752
2753static void remove_last_line(struct image *img)
2754{
2755        img->len -= img->line[--img->nr].len;
2756}
2757
2758/*
2759 * The change from "preimage" and "postimage" has been found to
2760 * apply at applied_pos (counts in line numbers) in "img".
2761 * Update "img" to remove "preimage" and replace it with "postimage".
2762 */
2763static void update_image(struct apply_state *state,
2764                         struct image *img,
2765                         int applied_pos,
2766                         struct image *preimage,
2767                         struct image *postimage)
2768{
2769        /*
2770         * remove the copy of preimage at offset in img
2771         * and replace it with postimage
2772         */
2773        int i, nr;
2774        size_t remove_count, insert_count, applied_at = 0;
2775        char *result;
2776        int preimage_limit;
2777
2778        /*
2779         * If we are removing blank lines at the end of img,
2780         * the preimage may extend beyond the end.
2781         * If that is the case, we must be careful only to
2782         * remove the part of the preimage that falls within
2783         * the boundaries of img. Initialize preimage_limit
2784         * to the number of lines in the preimage that falls
2785         * within the boundaries.
2786         */
2787        preimage_limit = preimage->nr;
2788        if (preimage_limit > img->nr - applied_pos)
2789                preimage_limit = img->nr - applied_pos;
2790
2791        for (i = 0; i < applied_pos; i++)
2792                applied_at += img->line[i].len;
2793
2794        remove_count = 0;
2795        for (i = 0; i < preimage_limit; i++)
2796                remove_count += img->line[applied_pos + i].len;
2797        insert_count = postimage->len;
2798
2799        /* Adjust the contents */
2800        result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2801        memcpy(result, img->buf, applied_at);
2802        memcpy(result + applied_at, postimage->buf, postimage->len);
2803        memcpy(result + applied_at + postimage->len,
2804               img->buf + (applied_at + remove_count),
2805               img->len - (applied_at + remove_count));
2806        free(img->buf);
2807        img->buf = result;
2808        img->len += insert_count - remove_count;
2809        result[img->len] = '\0';
2810
2811        /* Adjust the line table */
2812        nr = img->nr + postimage->nr - preimage_limit;
2813        if (preimage_limit < postimage->nr) {
2814                /*
2815                 * NOTE: this knows that we never call remove_first_line()
2816                 * on anything other than pre/post image.
2817                 */
2818                REALLOC_ARRAY(img->line, nr);
2819                img->line_allocated = img->line;
2820        }
2821        if (preimage_limit != postimage->nr)
2822                memmove(img->line + applied_pos + postimage->nr,
2823                        img->line + applied_pos + preimage_limit,
2824                        (img->nr - (applied_pos + preimage_limit)) *
2825                        sizeof(*img->line));
2826        memcpy(img->line + applied_pos,
2827               postimage->line,
2828               postimage->nr * sizeof(*img->line));
2829        if (!state->allow_overlap)
2830                for (i = 0; i < postimage->nr; i++)
2831                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2832        img->nr = nr;
2833}
2834
2835/*
2836 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2837 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2838 * replace the part of "img" with "postimage" text.
2839 */
2840static int apply_one_fragment(struct apply_state *state,
2841                              struct image *img, struct fragment *frag,
2842                              int inaccurate_eof, unsigned ws_rule,
2843                              int nth_fragment)
2844{
2845        int match_beginning, match_end;
2846        const char *patch = frag->patch;
2847        int size = frag->size;
2848        char *old, *oldlines;
2849        struct strbuf newlines;
2850        int new_blank_lines_at_end = 0;
2851        int found_new_blank_lines_at_end = 0;
2852        int hunk_linenr = frag->linenr;
2853        unsigned long leading, trailing;
2854        int pos, applied_pos;
2855        struct image preimage;
2856        struct image postimage;
2857
2858        memset(&preimage, 0, sizeof(preimage));
2859        memset(&postimage, 0, sizeof(postimage));
2860        oldlines = xmalloc(size);
2861        strbuf_init(&newlines, size);
2862
2863        old = oldlines;
2864        while (size > 0) {
2865                char first;
2866                int len = linelen(patch, size);
2867                int plen;
2868                int added_blank_line = 0;
2869                int is_blank_context = 0;
2870                size_t start;
2871
2872                if (!len)
2873                        break;
2874
2875                /*
2876                 * "plen" is how much of the line we should use for
2877                 * the actual patch data. Normally we just remove the
2878                 * first character on the line, but if the line is
2879                 * followed by "\ No newline", then we also remove the
2880                 * last one (which is the newline, of course).
2881                 */
2882                plen = len - 1;
2883                if (len < size && patch[len] == '\\')
2884                        plen--;
2885                first = *patch;
2886                if (state->apply_in_reverse) {
2887                        if (first == '-')
2888                                first = '+';
2889                        else if (first == '+')
2890                                first = '-';
2891                }
2892
2893                switch (first) {
2894                case '\n':
2895                        /* Newer GNU diff, empty context line */
2896                        if (plen < 0)
2897                                /* ... followed by '\No newline'; nothing */
2898                                break;
2899                        *old++ = '\n';
2900                        strbuf_addch(&newlines, '\n');
2901                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2902                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2903                        is_blank_context = 1;
2904                        break;
2905                case ' ':
2906                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2907                            ws_blank_line(patch + 1, plen, ws_rule))
2908                                is_blank_context = 1;
2909                case '-':
2910                        memcpy(old, patch + 1, plen);
2911                        add_line_info(&preimage, old, plen,
2912                                      (first == ' ' ? LINE_COMMON : 0));
2913                        old += plen;
2914                        if (first == '-')
2915                                break;
2916                /* Fall-through for ' ' */
2917                case '+':
2918                        /* --no-add does not add new lines */
2919                        if (first == '+' && state->no_add)
2920                                break;
2921
2922                        start = newlines.len;
2923                        if (first != '+' ||
2924                            !state->whitespace_error ||
2925                            state->ws_error_action != correct_ws_error) {
2926                                strbuf_add(&newlines, patch + 1, plen);
2927                        }
2928                        else {
2929                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2930                        }
2931                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2932                                      (first == '+' ? 0 : LINE_COMMON));
2933                        if (first == '+' &&
2934                            (ws_rule & WS_BLANK_AT_EOF) &&
2935                            ws_blank_line(patch + 1, plen, ws_rule))
2936                                added_blank_line = 1;
2937                        break;
2938                case '@': case '\\':
2939                        /* Ignore it, we already handled it */
2940                        break;
2941                default:
2942                        if (state->apply_verbosity > verbosity_normal)
2943                                error(_("invalid start of line: '%c'"), first);
2944                        applied_pos = -1;
2945                        goto out;
2946                }
2947                if (added_blank_line) {
2948                        if (!new_blank_lines_at_end)
2949                                found_new_blank_lines_at_end = hunk_linenr;
2950                        new_blank_lines_at_end++;
2951                }
2952                else if (is_blank_context)
2953                        ;
2954                else
2955                        new_blank_lines_at_end = 0;
2956                patch += len;
2957                size -= len;
2958                hunk_linenr++;
2959        }
2960        if (inaccurate_eof &&
2961            old > oldlines && old[-1] == '\n' &&
2962            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2963                old--;
2964                strbuf_setlen(&newlines, newlines.len - 1);
2965        }
2966
2967        leading = frag->leading;
2968        trailing = frag->trailing;
2969
2970        /*
2971         * A hunk to change lines at the beginning would begin with
2972         * @@ -1,L +N,M @@
2973         * but we need to be careful.  -U0 that inserts before the second
2974         * line also has this pattern.
2975         *
2976         * And a hunk to add to an empty file would begin with
2977         * @@ -0,0 +N,M @@
2978         *
2979         * In other words, a hunk that is (frag->oldpos <= 1) with or
2980         * without leading context must match at the beginning.
2981         */
2982        match_beginning = (!frag->oldpos ||
2983                           (frag->oldpos == 1 && !state->unidiff_zero));
2984
2985        /*
2986         * A hunk without trailing lines must match at the end.
2987         * However, we simply cannot tell if a hunk must match end
2988         * from the lack of trailing lines if the patch was generated
2989         * with unidiff without any context.
2990         */
2991        match_end = !state->unidiff_zero && !trailing;
2992
2993        pos = frag->newpos ? (frag->newpos - 1) : 0;
2994        preimage.buf = oldlines;
2995        preimage.len = old - oldlines;
2996        postimage.buf = newlines.buf;
2997        postimage.len = newlines.len;
2998        preimage.line = preimage.line_allocated;
2999        postimage.line = postimage.line_allocated;
3000
3001        for (;;) {
3002
3003                applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3004                                       ws_rule, match_beginning, match_end);
3005
3006                if (applied_pos >= 0)
3007                        break;
3008
3009                /* Am I at my context limits? */
3010                if ((leading <= state->p_context) && (trailing <= state->p_context))
3011                        break;
3012                if (match_beginning || match_end) {
3013                        match_beginning = match_end = 0;
3014                        continue;
3015                }
3016
3017                /*
3018                 * Reduce the number of context lines; reduce both
3019                 * leading and trailing if they are equal otherwise
3020                 * just reduce the larger context.
3021                 */
3022                if (leading >= trailing) {
3023                        remove_first_line(&preimage);
3024                        remove_first_line(&postimage);
3025                        pos--;
3026                        leading--;
3027                }
3028                if (trailing > leading) {
3029                        remove_last_line(&preimage);
3030                        remove_last_line(&postimage);
3031                        trailing--;
3032                }
3033        }
3034
3035        if (applied_pos >= 0) {
3036                if (new_blank_lines_at_end &&
3037                    preimage.nr + applied_pos >= img->nr &&
3038                    (ws_rule & WS_BLANK_AT_EOF) &&
3039                    state->ws_error_action != nowarn_ws_error) {
3040                        record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3041                                        found_new_blank_lines_at_end);
3042                        if (state->ws_error_action == correct_ws_error) {
3043                                while (new_blank_lines_at_end--)
3044                                        remove_last_line(&postimage);
3045                        }
3046                        /*
3047                         * We would want to prevent write_out_results()
3048                         * from taking place in apply_patch() that follows
3049                         * the callchain led us here, which is:
3050                         * apply_patch->check_patch_list->check_patch->
3051                         * apply_data->apply_fragments->apply_one_fragment
3052                         */
3053                        if (state->ws_error_action == die_on_ws_error)
3054                                state->apply = 0;
3055                }
3056
3057                if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3058                        int offset = applied_pos - pos;
3059                        if (state->apply_in_reverse)
3060                                offset = 0 - offset;
3061                        fprintf_ln(stderr,
3062                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
3063                                      "Hunk #%d succeeded at %d (offset %d lines).",
3064                                      offset),
3065                                   nth_fragment, applied_pos + 1, offset);
3066                }
3067
3068                /*
3069                 * Warn if it was necessary to reduce the number
3070                 * of context lines.
3071                 */
3072                if ((leading != frag->leading ||
3073                     trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3074                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3075                                             " to apply fragment at %d"),
3076                                   leading, trailing, applied_pos+1);
3077                update_image(state, img, applied_pos, &preimage, &postimage);
3078        } else {
3079                if (state->apply_verbosity > verbosity_normal)
3080                        error(_("while searching for:\n%.*s"),
3081                              (int)(old - oldlines), oldlines);
3082        }
3083
3084out:
3085        free(oldlines);
3086        strbuf_release(&newlines);
3087        free(preimage.line_allocated);
3088        free(postimage.line_allocated);
3089
3090        return (applied_pos < 0);
3091}
3092
3093static int apply_binary_fragment(struct apply_state *state,
3094                                 struct image *img,
3095                                 struct patch *patch)
3096{
3097        struct fragment *fragment = patch->fragments;
3098        unsigned long len;
3099        void *dst;
3100
3101        if (!fragment)
3102                return error(_("missing binary patch data for '%s'"),
3103                             patch->new_name ?
3104                             patch->new_name :
3105                             patch->old_name);
3106
3107        /* Binary patch is irreversible without the optional second hunk */
3108        if (state->apply_in_reverse) {
3109                if (!fragment->next)
3110                        return error(_("cannot reverse-apply a binary patch "
3111                                       "without the reverse hunk to '%s'"),
3112                                     patch->new_name
3113                                     ? patch->new_name : patch->old_name);
3114                fragment = fragment->next;
3115        }
3116        switch (fragment->binary_patch_method) {
3117        case BINARY_DELTA_DEFLATED:
3118                dst = patch_delta(img->buf, img->len, fragment->patch,
3119                                  fragment->size, &len);
3120                if (!dst)
3121                        return -1;
3122                clear_image(img);
3123                img->buf = dst;
3124                img->len = len;
3125                return 0;
3126        case BINARY_LITERAL_DEFLATED:
3127                clear_image(img);
3128                img->len = fragment->size;
3129                img->buf = xmemdupz(fragment->patch, img->len);
3130                return 0;
3131        }
3132        return -1;
3133}
3134
3135/*
3136 * Replace "img" with the result of applying the binary patch.
3137 * The binary patch data itself in patch->fragment is still kept
3138 * but the preimage prepared by the caller in "img" is freed here
3139 * or in the helper function apply_binary_fragment() this calls.
3140 */
3141static int apply_binary(struct apply_state *state,
3142                        struct image *img,
3143                        struct patch *patch)
3144{
3145        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3146        struct object_id oid;
3147
3148        /*
3149         * For safety, we require patch index line to contain
3150         * full 40-byte textual SHA1 for old and new, at least for now.
3151         */
3152        if (strlen(patch->old_sha1_prefix) != 40 ||
3153            strlen(patch->new_sha1_prefix) != 40 ||
3154            get_oid_hex(patch->old_sha1_prefix, &oid) ||
3155            get_oid_hex(patch->new_sha1_prefix, &oid))
3156                return error(_("cannot apply binary patch to '%s' "
3157                               "without full index line"), name);
3158
3159        if (patch->old_name) {
3160                /*
3161                 * See if the old one matches what the patch
3162                 * applies to.
3163                 */
3164                hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3165                if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3166                        return error(_("the patch applies to '%s' (%s), "
3167                                       "which does not match the "
3168                                       "current contents."),
3169                                     name, oid_to_hex(&oid));
3170        }
3171        else {
3172                /* Otherwise, the old one must be empty. */
3173                if (img->len)
3174                        return error(_("the patch applies to an empty "
3175                                       "'%s' but it is not empty"), name);
3176        }
3177
3178        get_oid_hex(patch->new_sha1_prefix, &oid);
3179        if (is_null_oid(&oid)) {
3180                clear_image(img);
3181                return 0; /* deletion patch */
3182        }
3183
3184        if (has_sha1_file(oid.hash)) {
3185                /* We already have the postimage */
3186                enum object_type type;
3187                unsigned long size;
3188                char *result;
3189
3190                result = read_sha1_file(oid.hash, &type, &size);
3191                if (!result)
3192                        return error(_("the necessary postimage %s for "
3193                                       "'%s' cannot be read"),
3194                                     patch->new_sha1_prefix, name);
3195                clear_image(img);
3196                img->buf = result;
3197                img->len = size;
3198        } else {
3199                /*
3200                 * We have verified buf matches the preimage;
3201                 * apply the patch data to it, which is stored
3202                 * in the patch->fragments->{patch,size}.
3203                 */
3204                if (apply_binary_fragment(state, img, patch))
3205                        return error(_("binary patch does not apply to '%s'"),
3206                                     name);
3207
3208                /* verify that the result matches */
3209                hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3210                if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3211                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3212                                name, patch->new_sha1_prefix, oid_to_hex(&oid));
3213        }
3214
3215        return 0;
3216}
3217
3218static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3219{
3220        struct fragment *frag = patch->fragments;
3221        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3222        unsigned ws_rule = patch->ws_rule;
3223        unsigned inaccurate_eof = patch->inaccurate_eof;
3224        int nth = 0;
3225
3226        if (patch->is_binary)
3227                return apply_binary(state, img, patch);
3228
3229        while (frag) {
3230                nth++;
3231                if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3232                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
3233                        if (!state->apply_with_reject)
3234                                return -1;
3235                        frag->rejected = 1;
3236                }
3237                frag = frag->next;
3238        }
3239        return 0;
3240}
3241
3242static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3243{
3244        if (S_ISGITLINK(mode)) {
3245                strbuf_grow(buf, 100);
3246                strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3247        } else {
3248                enum object_type type;
3249                unsigned long sz;
3250                char *result;
3251
3252                result = read_sha1_file(oid->hash, &type, &sz);
3253                if (!result)
3254                        return -1;
3255                /* XXX read_sha1_file NUL-terminates */
3256                strbuf_attach(buf, result, sz, sz + 1);
3257        }
3258        return 0;
3259}
3260
3261static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3262{
3263        if (!ce)
3264                return 0;
3265        return read_blob_object(buf, &ce->oid, ce->ce_mode);
3266}
3267
3268static struct patch *in_fn_table(struct apply_state *state, const char *name)
3269{
3270        struct string_list_item *item;
3271
3272        if (name == NULL)
3273                return NULL;
3274
3275        item = string_list_lookup(&state->fn_table, name);
3276        if (item != NULL)
3277                return (struct patch *)item->util;
3278
3279        return NULL;
3280}
3281
3282/*
3283 * item->util in the filename table records the status of the path.
3284 * Usually it points at a patch (whose result records the contents
3285 * of it after applying it), but it could be PATH_WAS_DELETED for a
3286 * path that a previously applied patch has already removed, or
3287 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3288 *
3289 * The latter is needed to deal with a case where two paths A and B
3290 * are swapped by first renaming A to B and then renaming B to A;
3291 * moving A to B should not be prevented due to presence of B as we
3292 * will remove it in a later patch.
3293 */
3294#define PATH_TO_BE_DELETED ((struct patch *) -2)
3295#define PATH_WAS_DELETED ((struct patch *) -1)
3296
3297static int to_be_deleted(struct patch *patch)
3298{
3299        return patch == PATH_TO_BE_DELETED;
3300}
3301
3302static int was_deleted(struct patch *patch)
3303{
3304        return patch == PATH_WAS_DELETED;
3305}
3306
3307static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3308{
3309        struct string_list_item *item;
3310
3311        /*
3312         * Always add new_name unless patch is a deletion
3313         * This should cover the cases for normal diffs,
3314         * file creations and copies
3315         */
3316        if (patch->new_name != NULL) {
3317                item = string_list_insert(&state->fn_table, patch->new_name);
3318                item->util = patch;
3319        }
3320
3321        /*
3322         * store a failure on rename/deletion cases because
3323         * later chunks shouldn't patch old names
3324         */
3325        if ((patch->new_name == NULL) || (patch->is_rename)) {
3326                item = string_list_insert(&state->fn_table, patch->old_name);
3327                item->util = PATH_WAS_DELETED;
3328        }
3329}
3330
3331static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3332{
3333        /*
3334         * store information about incoming file deletion
3335         */
3336        while (patch) {
3337                if ((patch->new_name == NULL) || (patch->is_rename)) {
3338                        struct string_list_item *item;
3339                        item = string_list_insert(&state->fn_table, patch->old_name);
3340                        item->util = PATH_TO_BE_DELETED;
3341                }
3342                patch = patch->next;
3343        }
3344}
3345
3346static int checkout_target(struct index_state *istate,
3347                           struct cache_entry *ce, struct stat *st)
3348{
3349        struct checkout costate = CHECKOUT_INIT;
3350
3351        costate.refresh_cache = 1;
3352        costate.istate = istate;
3353        if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3354                return error(_("cannot checkout %s"), ce->name);
3355        return 0;
3356}
3357
3358static struct patch *previous_patch(struct apply_state *state,
3359                                    struct patch *patch,
3360                                    int *gone)
3361{
3362        struct patch *previous;
3363
3364        *gone = 0;
3365        if (patch->is_copy || patch->is_rename)
3366                return NULL; /* "git" patches do not depend on the order */
3367
3368        previous = in_fn_table(state, patch->old_name);
3369        if (!previous)
3370                return NULL;
3371
3372        if (to_be_deleted(previous))
3373                return NULL; /* the deletion hasn't happened yet */
3374
3375        if (was_deleted(previous))
3376                *gone = 1;
3377
3378        return previous;
3379}
3380
3381static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3382{
3383        if (S_ISGITLINK(ce->ce_mode)) {
3384                if (!S_ISDIR(st->st_mode))
3385                        return -1;
3386                return 0;
3387        }
3388        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3389}
3390
3391#define SUBMODULE_PATCH_WITHOUT_INDEX 1
3392
3393static int load_patch_target(struct apply_state *state,
3394                             struct strbuf *buf,
3395                             const struct cache_entry *ce,
3396                             struct stat *st,
3397                             const char *name,
3398                             unsigned expected_mode)
3399{
3400        if (state->cached || state->check_index) {
3401                if (read_file_or_gitlink(ce, buf))
3402                        return error(_("failed to read %s"), name);
3403        } else if (name) {
3404                if (S_ISGITLINK(expected_mode)) {
3405                        if (ce)
3406                                return read_file_or_gitlink(ce, buf);
3407                        else
3408                                return SUBMODULE_PATCH_WITHOUT_INDEX;
3409                } else if (has_symlink_leading_path(name, strlen(name))) {
3410                        return error(_("reading from '%s' beyond a symbolic link"), name);
3411                } else {
3412                        if (read_old_data(st, name, buf))
3413                                return error(_("failed to read %s"), name);
3414                }
3415        }
3416        return 0;
3417}
3418
3419/*
3420 * We are about to apply "patch"; populate the "image" with the
3421 * current version we have, from the working tree or from the index,
3422 * depending on the situation e.g. --cached/--index.  If we are
3423 * applying a non-git patch that incrementally updates the tree,
3424 * we read from the result of a previous diff.
3425 */
3426static int load_preimage(struct apply_state *state,
3427                         struct image *image,
3428                         struct patch *patch, struct stat *st,
3429                         const struct cache_entry *ce)
3430{
3431        struct strbuf buf = STRBUF_INIT;
3432        size_t len;
3433        char *img;
3434        struct patch *previous;
3435        int status;
3436
3437        previous = previous_patch(state, patch, &status);
3438        if (status)
3439                return error(_("path %s has been renamed/deleted"),
3440                             patch->old_name);
3441        if (previous) {
3442                /* We have a patched copy in memory; use that. */
3443                strbuf_add(&buf, previous->result, previous->resultsize);
3444        } else {
3445                status = load_patch_target(state, &buf, ce, st,
3446                                           patch->old_name, patch->old_mode);
3447                if (status < 0)
3448                        return status;
3449                else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3450                        /*
3451                         * There is no way to apply subproject
3452                         * patch without looking at the index.
3453                         * NEEDSWORK: shouldn't this be flagged
3454                         * as an error???
3455                         */
3456                        free_fragment_list(patch->fragments);
3457                        patch->fragments = NULL;
3458                } else if (status) {
3459                        return error(_("failed to read %s"), patch->old_name);
3460                }
3461        }
3462
3463        img = strbuf_detach(&buf, &len);
3464        prepare_image(image, img, len, !patch->is_binary);
3465        return 0;
3466}
3467
3468static int three_way_merge(struct image *image,
3469                           char *path,
3470                           const struct object_id *base,
3471                           const struct object_id *ours,
3472                           const struct object_id *theirs)
3473{
3474        mmfile_t base_file, our_file, their_file;
3475        mmbuffer_t result = { NULL };
3476        int status;
3477
3478        read_mmblob(&base_file, base);
3479        read_mmblob(&our_file, ours);
3480        read_mmblob(&their_file, theirs);
3481        status = ll_merge(&result, path,
3482                          &base_file, "base",
3483                          &our_file, "ours",
3484                          &their_file, "theirs", NULL);
3485        free(base_file.ptr);
3486        free(our_file.ptr);
3487        free(their_file.ptr);
3488        if (status < 0 || !result.ptr) {
3489                free(result.ptr);
3490                return -1;
3491        }
3492        clear_image(image);
3493        image->buf = result.ptr;
3494        image->len = result.size;
3495
3496        return status;
3497}
3498
3499/*
3500 * When directly falling back to add/add three-way merge, we read from
3501 * the current contents of the new_name.  In no cases other than that
3502 * this function will be called.
3503 */
3504static int load_current(struct apply_state *state,
3505                        struct image *image,
3506                        struct patch *patch)
3507{
3508        struct strbuf buf = STRBUF_INIT;
3509        int status, pos;
3510        size_t len;
3511        char *img;
3512        struct stat st;
3513        struct cache_entry *ce;
3514        char *name = patch->new_name;
3515        unsigned mode = patch->new_mode;
3516
3517        if (!patch->is_new)
3518                die("BUG: patch to %s is not a creation", patch->old_name);
3519
3520        pos = cache_name_pos(name, strlen(name));
3521        if (pos < 0)
3522                return error(_("%s: does not exist in index"), name);
3523        ce = active_cache[pos];
3524        if (lstat(name, &st)) {
3525                if (errno != ENOENT)
3526                        return error_errno("%s", name);
3527                if (checkout_target(&the_index, ce, &st))
3528                        return -1;
3529        }
3530        if (verify_index_match(ce, &st))
3531                return error(_("%s: does not match index"), name);
3532
3533        status = load_patch_target(state, &buf, ce, &st, name, mode);
3534        if (status < 0)
3535                return status;
3536        else if (status)
3537                return -1;
3538        img = strbuf_detach(&buf, &len);
3539        prepare_image(image, img, len, !patch->is_binary);
3540        return 0;
3541}
3542
3543static int try_threeway(struct apply_state *state,
3544                        struct image *image,
3545                        struct patch *patch,
3546                        struct stat *st,
3547                        const struct cache_entry *ce)
3548{
3549        struct object_id pre_oid, post_oid, our_oid;
3550        struct strbuf buf = STRBUF_INIT;
3551        size_t len;
3552        int status;
3553        char *img;
3554        struct image tmp_image;
3555
3556        /* No point falling back to 3-way merge in these cases */
3557        if (patch->is_delete ||
3558            S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3559                return -1;
3560
3561        /* Preimage the patch was prepared for */
3562        if (patch->is_new)
3563                write_sha1_file("", 0, blob_type, pre_oid.hash);
3564        else if (get_sha1(patch->old_sha1_prefix, pre_oid.hash) ||
3565                 read_blob_object(&buf, &pre_oid, patch->old_mode))
3566                return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3567
3568        if (state->apply_verbosity > verbosity_silent)
3569                fprintf(stderr, _("Falling back to three-way merge...\n"));
3570
3571        img = strbuf_detach(&buf, &len);
3572        prepare_image(&tmp_image, img, len, 1);
3573        /* Apply the patch to get the post image */
3574        if (apply_fragments(state, &tmp_image, patch) < 0) {
3575                clear_image(&tmp_image);
3576                return -1;
3577        }
3578        /* post_oid is theirs */
3579        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3580        clear_image(&tmp_image);
3581
3582        /* our_oid is ours */
3583        if (patch->is_new) {
3584                if (load_current(state, &tmp_image, patch))
3585                        return error(_("cannot read the current contents of '%s'"),
3586                                     patch->new_name);
3587        } else {
3588                if (load_preimage(state, &tmp_image, patch, st, ce))
3589                        return error(_("cannot read the current contents of '%s'"),
3590                                     patch->old_name);
3591        }
3592        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3593        clear_image(&tmp_image);
3594
3595        /* in-core three-way merge between post and our using pre as base */
3596        status = three_way_merge(image, patch->new_name,
3597                                 &pre_oid, &our_oid, &post_oid);
3598        if (status < 0) {
3599                if (state->apply_verbosity > verbosity_silent)
3600                        fprintf(stderr,
3601                                _("Failed to fall back on three-way merge...\n"));
3602                return status;
3603        }
3604
3605        if (status) {
3606                patch->conflicted_threeway = 1;
3607                if (patch->is_new)
3608                        oidclr(&patch->threeway_stage[0]);
3609                else
3610                        oidcpy(&patch->threeway_stage[0], &pre_oid);
3611                oidcpy(&patch->threeway_stage[1], &our_oid);
3612                oidcpy(&patch->threeway_stage[2], &post_oid);
3613                if (state->apply_verbosity > verbosity_silent)
3614                        fprintf(stderr,
3615                                _("Applied patch to '%s' with conflicts.\n"),
3616                                patch->new_name);
3617        } else {
3618                if (state->apply_verbosity > verbosity_silent)
3619                        fprintf(stderr,
3620                                _("Applied patch to '%s' cleanly.\n"),
3621                                patch->new_name);
3622        }
3623        return 0;
3624}
3625
3626static int apply_data(struct apply_state *state, struct patch *patch,
3627                      struct stat *st, const struct cache_entry *ce)
3628{
3629        struct image image;
3630
3631        if (load_preimage(state, &image, patch, st, ce) < 0)
3632                return -1;
3633
3634        if (patch->direct_to_threeway ||
3635            apply_fragments(state, &image, patch) < 0) {
3636                /* Note: with --reject, apply_fragments() returns 0 */
3637                if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3638                        return -1;
3639        }
3640        patch->result = image.buf;
3641        patch->resultsize = image.len;
3642        add_to_fn_table(state, patch);
3643        free(image.line_allocated);
3644
3645        if (0 < patch->is_delete && patch->resultsize)
3646                return error(_("removal patch leaves file contents"));
3647
3648        return 0;
3649}
3650
3651/*
3652 * If "patch" that we are looking at modifies or deletes what we have,
3653 * we would want it not to lose any local modification we have, either
3654 * in the working tree or in the index.
3655 *
3656 * This also decides if a non-git patch is a creation patch or a
3657 * modification to an existing empty file.  We do not check the state
3658 * of the current tree for a creation patch in this function; the caller
3659 * check_patch() separately makes sure (and errors out otherwise) that
3660 * the path the patch creates does not exist in the current tree.
3661 */
3662static int check_preimage(struct apply_state *state,
3663                          struct patch *patch,
3664                          struct cache_entry **ce,
3665                          struct stat *st)
3666{
3667        const char *old_name = patch->old_name;
3668        struct patch *previous = NULL;
3669        int stat_ret = 0, status;
3670        unsigned st_mode = 0;
3671
3672        if (!old_name)
3673                return 0;
3674
3675        assert(patch->is_new <= 0);
3676        previous = previous_patch(state, patch, &status);
3677
3678        if (status)
3679                return error(_("path %s has been renamed/deleted"), old_name);
3680        if (previous) {
3681                st_mode = previous->new_mode;
3682        } else if (!state->cached) {
3683                stat_ret = lstat(old_name, st);
3684                if (stat_ret && errno != ENOENT)
3685                        return error_errno("%s", old_name);
3686        }
3687
3688        if (state->check_index && !previous) {
3689                int pos = cache_name_pos(old_name, strlen(old_name));
3690                if (pos < 0) {
3691                        if (patch->is_new < 0)
3692                                goto is_new;
3693                        return error(_("%s: does not exist in index"), old_name);
3694                }
3695                *ce = active_cache[pos];
3696                if (stat_ret < 0) {
3697                        if (checkout_target(&the_index, *ce, st))
3698                                return -1;
3699                }
3700                if (!state->cached && verify_index_match(*ce, st))
3701                        return error(_("%s: does not match index"), old_name);
3702                if (state->cached)
3703                        st_mode = (*ce)->ce_mode;
3704        } else if (stat_ret < 0) {
3705                if (patch->is_new < 0)
3706                        goto is_new;
3707                return error_errno("%s", old_name);
3708        }
3709
3710        if (!state->cached && !previous)
3711                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3712
3713        if (patch->is_new < 0)
3714                patch->is_new = 0;
3715        if (!patch->old_mode)
3716                patch->old_mode = st_mode;
3717        if ((st_mode ^ patch->old_mode) & S_IFMT)
3718                return error(_("%s: wrong type"), old_name);
3719        if (st_mode != patch->old_mode)
3720                warning(_("%s has type %o, expected %o"),
3721                        old_name, st_mode, patch->old_mode);
3722        if (!patch->new_mode && !patch->is_delete)
3723                patch->new_mode = st_mode;
3724        return 0;
3725
3726 is_new:
3727        patch->is_new = 1;
3728        patch->is_delete = 0;
3729        free(patch->old_name);
3730        patch->old_name = NULL;
3731        return 0;
3732}
3733
3734
3735#define EXISTS_IN_INDEX 1
3736#define EXISTS_IN_WORKTREE 2
3737
3738static int check_to_create(struct apply_state *state,
3739                           const char *new_name,
3740                           int ok_if_exists)
3741{
3742        struct stat nst;
3743
3744        if (state->check_index &&
3745            cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3746            !ok_if_exists)
3747                return EXISTS_IN_INDEX;
3748        if (state->cached)
3749                return 0;
3750
3751        if (!lstat(new_name, &nst)) {
3752                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3753                        return 0;
3754                /*
3755                 * A leading component of new_name might be a symlink
3756                 * that is going to be removed with this patch, but
3757                 * still pointing at somewhere that has the path.
3758                 * In such a case, path "new_name" does not exist as
3759                 * far as git is concerned.
3760                 */
3761                if (has_symlink_leading_path(new_name, strlen(new_name)))
3762                        return 0;
3763
3764                return EXISTS_IN_WORKTREE;
3765        } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3766                return error_errno("%s", new_name);
3767        }
3768        return 0;
3769}
3770
3771static uintptr_t register_symlink_changes(struct apply_state *state,
3772                                          const char *path,
3773                                          uintptr_t what)
3774{
3775        struct string_list_item *ent;
3776
3777        ent = string_list_lookup(&state->symlink_changes, path);
3778        if (!ent) {
3779                ent = string_list_insert(&state->symlink_changes, path);
3780                ent->util = (void *)0;
3781        }
3782        ent->util = (void *)(what | ((uintptr_t)ent->util));
3783        return (uintptr_t)ent->util;
3784}
3785
3786static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3787{
3788        struct string_list_item *ent;
3789
3790        ent = string_list_lookup(&state->symlink_changes, path);
3791        if (!ent)
3792                return 0;
3793        return (uintptr_t)ent->util;
3794}
3795
3796static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3797{
3798        for ( ; patch; patch = patch->next) {
3799                if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3800                    (patch->is_rename || patch->is_delete))
3801                        /* the symlink at patch->old_name is removed */
3802                        register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3803
3804                if (patch->new_name && S_ISLNK(patch->new_mode))
3805                        /* the symlink at patch->new_name is created or remains */
3806                        register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3807        }
3808}
3809
3810static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3811{
3812        do {
3813                unsigned int change;
3814
3815                while (--name->len && name->buf[name->len] != '/')
3816                        ; /* scan backwards */
3817                if (!name->len)
3818                        break;
3819                name->buf[name->len] = '\0';
3820                change = check_symlink_changes(state, name->buf);
3821                if (change & APPLY_SYMLINK_IN_RESULT)
3822                        return 1;
3823                if (change & APPLY_SYMLINK_GOES_AWAY)
3824                        /*
3825                         * This cannot be "return 0", because we may
3826                         * see a new one created at a higher level.
3827                         */
3828                        continue;
3829
3830                /* otherwise, check the preimage */
3831                if (state->check_index) {
3832                        struct cache_entry *ce;
3833
3834                        ce = cache_file_exists(name->buf, name->len, ignore_case);
3835                        if (ce && S_ISLNK(ce->ce_mode))
3836                                return 1;
3837                } else {
3838                        struct stat st;
3839                        if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3840                                return 1;
3841                }
3842        } while (1);
3843        return 0;
3844}
3845
3846static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3847{
3848        int ret;
3849        struct strbuf name = STRBUF_INIT;
3850
3851        assert(*name_ != '\0');
3852        strbuf_addstr(&name, name_);
3853        ret = path_is_beyond_symlink_1(state, &name);
3854        strbuf_release(&name);
3855
3856        return ret;
3857}
3858
3859static int check_unsafe_path(struct patch *patch)
3860{
3861        const char *old_name = NULL;
3862        const char *new_name = NULL;
3863        if (patch->is_delete)
3864                old_name = patch->old_name;
3865        else if (!patch->is_new && !patch->is_copy)
3866                old_name = patch->old_name;
3867        if (!patch->is_delete)
3868                new_name = patch->new_name;
3869
3870        if (old_name && !verify_path(old_name))
3871                return error(_("invalid path '%s'"), old_name);
3872        if (new_name && !verify_path(new_name))
3873                return error(_("invalid path '%s'"), new_name);
3874        return 0;
3875}
3876
3877/*
3878 * Check and apply the patch in-core; leave the result in patch->result
3879 * for the caller to write it out to the final destination.
3880 */
3881static int check_patch(struct apply_state *state, struct patch *patch)
3882{
3883        struct stat st;
3884        const char *old_name = patch->old_name;
3885        const char *new_name = patch->new_name;
3886        const char *name = old_name ? old_name : new_name;
3887        struct cache_entry *ce = NULL;
3888        struct patch *tpatch;
3889        int ok_if_exists;
3890        int status;
3891
3892        patch->rejected = 1; /* we will drop this after we succeed */
3893
3894        status = check_preimage(state, patch, &ce, &st);
3895        if (status)
3896                return status;
3897        old_name = patch->old_name;
3898
3899        /*
3900         * A type-change diff is always split into a patch to delete
3901         * old, immediately followed by a patch to create new (see
3902         * diff.c::run_diff()); in such a case it is Ok that the entry
3903         * to be deleted by the previous patch is still in the working
3904         * tree and in the index.
3905         *
3906         * A patch to swap-rename between A and B would first rename A
3907         * to B and then rename B to A.  While applying the first one,
3908         * the presence of B should not stop A from getting renamed to
3909         * B; ask to_be_deleted() about the later rename.  Removal of
3910         * B and rename from A to B is handled the same way by asking
3911         * was_deleted().
3912         */
3913        if ((tpatch = in_fn_table(state, new_name)) &&
3914            (was_deleted(tpatch) || to_be_deleted(tpatch)))
3915                ok_if_exists = 1;
3916        else
3917                ok_if_exists = 0;
3918
3919        if (new_name &&
3920            ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3921                int err = check_to_create(state, new_name, ok_if_exists);
3922
3923                if (err && state->threeway) {
3924                        patch->direct_to_threeway = 1;
3925                } else switch (err) {
3926                case 0:
3927                        break; /* happy */
3928                case EXISTS_IN_INDEX:
3929                        return error(_("%s: already exists in index"), new_name);
3930                        break;
3931                case EXISTS_IN_WORKTREE:
3932                        return error(_("%s: already exists in working directory"),
3933                                     new_name);
3934                default:
3935                        return err;
3936                }
3937
3938                if (!patch->new_mode) {
3939                        if (0 < patch->is_new)
3940                                patch->new_mode = S_IFREG | 0644;
3941                        else
3942                                patch->new_mode = patch->old_mode;
3943                }
3944        }
3945
3946        if (new_name && old_name) {
3947                int same = !strcmp(old_name, new_name);
3948                if (!patch->new_mode)
3949                        patch->new_mode = patch->old_mode;
3950                if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3951                        if (same)
3952                                return error(_("new mode (%o) of %s does not "
3953                                               "match old mode (%o)"),
3954                                        patch->new_mode, new_name,
3955                                        patch->old_mode);
3956                        else
3957                                return error(_("new mode (%o) of %s does not "
3958                                               "match old mode (%o) of %s"),
3959                                        patch->new_mode, new_name,
3960                                        patch->old_mode, old_name);
3961                }
3962        }
3963
3964        if (!state->unsafe_paths && check_unsafe_path(patch))
3965                return -128;
3966
3967        /*
3968         * An attempt to read from or delete a path that is beyond a
3969         * symbolic link will be prevented by load_patch_target() that
3970         * is called at the beginning of apply_data() so we do not
3971         * have to worry about a patch marked with "is_delete" bit
3972         * here.  We however need to make sure that the patch result
3973         * is not deposited to a path that is beyond a symbolic link
3974         * here.
3975         */
3976        if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3977                return error(_("affected file '%s' is beyond a symbolic link"),
3978                             patch->new_name);
3979
3980        if (apply_data(state, patch, &st, ce) < 0)
3981                return error(_("%s: patch does not apply"), name);
3982        patch->rejected = 0;
3983        return 0;
3984}
3985
3986static int check_patch_list(struct apply_state *state, struct patch *patch)
3987{
3988        int err = 0;
3989
3990        prepare_symlink_changes(state, patch);
3991        prepare_fn_table(state, patch);
3992        while (patch) {
3993                int res;
3994                if (state->apply_verbosity > verbosity_normal)
3995                        say_patch_name(stderr,
3996                                       _("Checking patch %s..."), patch);
3997                res = check_patch(state, patch);
3998                if (res == -128)
3999                        return -128;
4000                err |= res;
4001                patch = patch->next;
4002        }
4003        return err;
4004}
4005
4006static int read_apply_cache(struct apply_state *state)
4007{
4008        if (state->index_file)
4009                return read_cache_from(state->index_file);
4010        else
4011                return read_cache();
4012}
4013
4014/* This function tries to read the object name from the current index */
4015static int get_current_oid(struct apply_state *state, const char *path,
4016                           struct object_id *oid)
4017{
4018        int pos;
4019
4020        if (read_apply_cache(state) < 0)
4021                return -1;
4022        pos = cache_name_pos(path, strlen(path));
4023        if (pos < 0)
4024                return -1;
4025        oidcpy(oid, &active_cache[pos]->oid);
4026        return 0;
4027}
4028
4029static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4030{
4031        /*
4032         * A usable gitlink patch has only one fragment (hunk) that looks like:
4033         * @@ -1 +1 @@
4034         * -Subproject commit <old sha1>
4035         * +Subproject commit <new sha1>
4036         * or
4037         * @@ -1 +0,0 @@
4038         * -Subproject commit <old sha1>
4039         * for a removal patch.
4040         */
4041        struct fragment *hunk = p->fragments;
4042        static const char heading[] = "-Subproject commit ";
4043        char *preimage;
4044
4045        if (/* does the patch have only one hunk? */
4046            hunk && !hunk->next &&
4047            /* is its preimage one line? */
4048            hunk->oldpos == 1 && hunk->oldlines == 1 &&
4049            /* does preimage begin with the heading? */
4050            (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4051            starts_with(++preimage, heading) &&
4052            /* does it record full SHA-1? */
4053            !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4054            preimage[sizeof(heading) + GIT_SHA1_HEXSZ - 1] == '\n' &&
4055            /* does the abbreviated name on the index line agree with it? */
4056            starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
4057                return 0; /* it all looks fine */
4058
4059        /* we may have full object name on the index line */
4060        return get_oid_hex(p->old_sha1_prefix, oid);
4061}
4062
4063/* Build an index that contains the just the files needed for a 3way merge */
4064static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4065{
4066        struct patch *patch;
4067        struct index_state result = { NULL };
4068        static struct lock_file lock;
4069        int res;
4070
4071        /* Once we start supporting the reverse patch, it may be
4072         * worth showing the new sha1 prefix, but until then...
4073         */
4074        for (patch = list; patch; patch = patch->next) {
4075                struct object_id oid;
4076                struct cache_entry *ce;
4077                const char *name;
4078
4079                name = patch->old_name ? patch->old_name : patch->new_name;
4080                if (0 < patch->is_new)
4081                        continue;
4082
4083                if (S_ISGITLINK(patch->old_mode)) {
4084                        if (!preimage_oid_in_gitlink_patch(patch, &oid))
4085                                ; /* ok, the textual part looks sane */
4086                        else
4087                                return error(_("sha1 information is lacking or "
4088                                               "useless for submodule %s"), name);
4089                } else if (!get_sha1_blob(patch->old_sha1_prefix, oid.hash)) {
4090                        ; /* ok */
4091                } else if (!patch->lines_added && !patch->lines_deleted) {
4092                        /* mode-only change: update the current */
4093                        if (get_current_oid(state, patch->old_name, &oid))
4094                                return error(_("mode change for %s, which is not "
4095                                               "in current HEAD"), name);
4096                } else
4097                        return error(_("sha1 information is lacking or useless "
4098                                       "(%s)."), name);
4099
4100                ce = make_cache_entry(patch->old_mode, oid.hash, name, 0, 0);
4101                if (!ce)
4102                        return error(_("make_cache_entry failed for path '%s'"),
4103                                     name);
4104                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4105                        free(ce);
4106                        return error(_("could not add %s to temporary index"),
4107                                     name);
4108                }
4109        }
4110
4111        hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4112        res = write_locked_index(&result, &lock, COMMIT_LOCK);
4113        discard_index(&result);
4114
4115        if (res)
4116                return error(_("could not write temporary index to %s"),
4117                             state->fake_ancestor);
4118
4119        return 0;
4120}
4121
4122static void stat_patch_list(struct apply_state *state, struct patch *patch)
4123{
4124        int files, adds, dels;
4125
4126        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4127                files++;
4128                adds += patch->lines_added;
4129                dels += patch->lines_deleted;
4130                show_stats(state, patch);
4131        }
4132
4133        print_stat_summary(stdout, files, adds, dels);
4134}
4135
4136static void numstat_patch_list(struct apply_state *state,
4137                               struct patch *patch)
4138{
4139        for ( ; patch; patch = patch->next) {
4140                const char *name;
4141                name = patch->new_name ? patch->new_name : patch->old_name;
4142                if (patch->is_binary)
4143                        printf("-\t-\t");
4144                else
4145                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4146                write_name_quoted(name, stdout, state->line_termination);
4147        }
4148}
4149
4150static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4151{
4152        if (mode)
4153                printf(" %s mode %06o %s\n", newdelete, mode, name);
4154        else
4155                printf(" %s %s\n", newdelete, name);
4156}
4157
4158static void show_mode_change(struct patch *p, int show_name)
4159{
4160        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4161                if (show_name)
4162                        printf(" mode change %06o => %06o %s\n",
4163                               p->old_mode, p->new_mode, p->new_name);
4164                else
4165                        printf(" mode change %06o => %06o\n",
4166                               p->old_mode, p->new_mode);
4167        }
4168}
4169
4170static void show_rename_copy(struct patch *p)
4171{
4172        const char *renamecopy = p->is_rename ? "rename" : "copy";
4173        const char *old, *new;
4174
4175        /* Find common prefix */
4176        old = p->old_name;
4177        new = p->new_name;
4178        while (1) {
4179                const char *slash_old, *slash_new;
4180                slash_old = strchr(old, '/');
4181                slash_new = strchr(new, '/');
4182                if (!slash_old ||
4183                    !slash_new ||
4184                    slash_old - old != slash_new - new ||
4185                    memcmp(old, new, slash_new - new))
4186                        break;
4187                old = slash_old + 1;
4188                new = slash_new + 1;
4189        }
4190        /* p->old_name thru old is the common prefix, and old and new
4191         * through the end of names are renames
4192         */
4193        if (old != p->old_name)
4194                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4195                       (int)(old - p->old_name), p->old_name,
4196                       old, new, p->score);
4197        else
4198                printf(" %s %s => %s (%d%%)\n", renamecopy,
4199                       p->old_name, p->new_name, p->score);
4200        show_mode_change(p, 0);
4201}
4202
4203static void summary_patch_list(struct patch *patch)
4204{
4205        struct patch *p;
4206
4207        for (p = patch; p; p = p->next) {
4208                if (p->is_new)
4209                        show_file_mode_name("create", p->new_mode, p->new_name);
4210                else if (p->is_delete)
4211                        show_file_mode_name("delete", p->old_mode, p->old_name);
4212                else {
4213                        if (p->is_rename || p->is_copy)
4214                                show_rename_copy(p);
4215                        else {
4216                                if (p->score) {
4217                                        printf(" rewrite %s (%d%%)\n",
4218                                               p->new_name, p->score);
4219                                        show_mode_change(p, 0);
4220                                }
4221                                else
4222                                        show_mode_change(p, 1);
4223                        }
4224                }
4225        }
4226}
4227
4228static void patch_stats(struct apply_state *state, struct patch *patch)
4229{
4230        int lines = patch->lines_added + patch->lines_deleted;
4231
4232        if (lines > state->max_change)
4233                state->max_change = lines;
4234        if (patch->old_name) {
4235                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4236                if (!len)
4237                        len = strlen(patch->old_name);
4238                if (len > state->max_len)
4239                        state->max_len = len;
4240        }
4241        if (patch->new_name) {
4242                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4243                if (!len)
4244                        len = strlen(patch->new_name);
4245                if (len > state->max_len)
4246                        state->max_len = len;
4247        }
4248}
4249
4250static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4251{
4252        if (state->update_index) {
4253                if (remove_file_from_cache(patch->old_name) < 0)
4254                        return error(_("unable to remove %s from index"), patch->old_name);
4255        }
4256        if (!state->cached) {
4257                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4258                        remove_path(patch->old_name);
4259                }
4260        }
4261        return 0;
4262}
4263
4264static int add_index_file(struct apply_state *state,
4265                          const char *path,
4266                          unsigned mode,
4267                          void *buf,
4268                          unsigned long size)
4269{
4270        struct stat st;
4271        struct cache_entry *ce;
4272        int namelen = strlen(path);
4273        unsigned ce_size = cache_entry_size(namelen);
4274
4275        if (!state->update_index)
4276                return 0;
4277
4278        ce = xcalloc(1, ce_size);
4279        memcpy(ce->name, path, namelen);
4280        ce->ce_mode = create_ce_mode(mode);
4281        ce->ce_flags = create_ce_flags(0);
4282        ce->ce_namelen = namelen;
4283        if (S_ISGITLINK(mode)) {
4284                const char *s;
4285
4286                if (!skip_prefix(buf, "Subproject commit ", &s) ||
4287                    get_oid_hex(s, &ce->oid)) {
4288                        free(ce);
4289                       return error(_("corrupt patch for submodule %s"), path);
4290                }
4291        } else {
4292                if (!state->cached) {
4293                        if (lstat(path, &st) < 0) {
4294                                free(ce);
4295                                return error_errno(_("unable to stat newly "
4296                                                     "created file '%s'"),
4297                                                   path);
4298                        }
4299                        fill_stat_cache_info(ce, &st);
4300                }
4301                if (write_sha1_file(buf, size, blob_type, ce->oid.hash) < 0) {
4302                        free(ce);
4303                        return error(_("unable to create backing store "
4304                                       "for newly created file %s"), path);
4305                }
4306        }
4307        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4308                free(ce);
4309                return error(_("unable to add cache entry for %s"), path);
4310        }
4311
4312        return 0;
4313}
4314
4315/*
4316 * Returns:
4317 *  -1 if an unrecoverable error happened
4318 *   0 if everything went well
4319 *   1 if a recoverable error happened
4320 */
4321static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4322{
4323        int fd, res;
4324        struct strbuf nbuf = STRBUF_INIT;
4325
4326        if (S_ISGITLINK(mode)) {
4327                struct stat st;
4328                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4329                        return 0;
4330                return !!mkdir(path, 0777);
4331        }
4332
4333        if (has_symlinks && S_ISLNK(mode))
4334                /* Although buf:size is counted string, it also is NUL
4335                 * terminated.
4336                 */
4337                return !!symlink(buf, path);
4338
4339        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4340        if (fd < 0)
4341                return 1;
4342
4343        if (convert_to_working_tree(path, buf, size, &nbuf)) {
4344                size = nbuf.len;
4345                buf  = nbuf.buf;
4346        }
4347
4348        res = write_in_full(fd, buf, size) < 0;
4349        if (res)
4350                error_errno(_("failed to write to '%s'"), path);
4351        strbuf_release(&nbuf);
4352
4353        if (close(fd) < 0 && !res)
4354                return error_errno(_("closing file '%s'"), path);
4355
4356        return res ? -1 : 0;
4357}
4358
4359/*
4360 * We optimistically assume that the directories exist,
4361 * which is true 99% of the time anyway. If they don't,
4362 * we create them and try again.
4363 *
4364 * Returns:
4365 *   -1 on error
4366 *   0 otherwise
4367 */
4368static int create_one_file(struct apply_state *state,
4369                           char *path,
4370                           unsigned mode,
4371                           const char *buf,
4372                           unsigned long size)
4373{
4374        int res;
4375
4376        if (state->cached)
4377                return 0;
4378
4379        res = try_create_file(path, mode, buf, size);
4380        if (res < 0)
4381                return -1;
4382        if (!res)
4383                return 0;
4384
4385        if (errno == ENOENT) {
4386                if (safe_create_leading_directories(path))
4387                        return 0;
4388                res = try_create_file(path, mode, buf, size);
4389                if (res < 0)
4390                        return -1;
4391                if (!res)
4392                        return 0;
4393        }
4394
4395        if (errno == EEXIST || errno == EACCES) {
4396                /* We may be trying to create a file where a directory
4397                 * used to be.
4398                 */
4399                struct stat st;
4400                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4401                        errno = EEXIST;
4402        }
4403
4404        if (errno == EEXIST) {
4405                unsigned int nr = getpid();
4406
4407                for (;;) {
4408                        char newpath[PATH_MAX];
4409                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4410                        res = try_create_file(newpath, mode, buf, size);
4411                        if (res < 0)
4412                                return -1;
4413                        if (!res) {
4414                                if (!rename(newpath, path))
4415                                        return 0;
4416                                unlink_or_warn(newpath);
4417                                break;
4418                        }
4419                        if (errno != EEXIST)
4420                                break;
4421                        ++nr;
4422                }
4423        }
4424        return error_errno(_("unable to write file '%s' mode %o"),
4425                           path, mode);
4426}
4427
4428static int add_conflicted_stages_file(struct apply_state *state,
4429                                       struct patch *patch)
4430{
4431        int stage, namelen;
4432        unsigned ce_size, mode;
4433        struct cache_entry *ce;
4434
4435        if (!state->update_index)
4436                return 0;
4437        namelen = strlen(patch->new_name);
4438        ce_size = cache_entry_size(namelen);
4439        mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4440
4441        remove_file_from_cache(patch->new_name);
4442        for (stage = 1; stage < 4; stage++) {
4443                if (is_null_oid(&patch->threeway_stage[stage - 1]))
4444                        continue;
4445                ce = xcalloc(1, ce_size);
4446                memcpy(ce->name, patch->new_name, namelen);
4447                ce->ce_mode = create_ce_mode(mode);
4448                ce->ce_flags = create_ce_flags(stage);
4449                ce->ce_namelen = namelen;
4450                oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4451                if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4452                        free(ce);
4453                        return error(_("unable to add cache entry for %s"),
4454                                     patch->new_name);
4455                }
4456        }
4457
4458        return 0;
4459}
4460
4461static int create_file(struct apply_state *state, struct patch *patch)
4462{
4463        char *path = patch->new_name;
4464        unsigned mode = patch->new_mode;
4465        unsigned long size = patch->resultsize;
4466        char *buf = patch->result;
4467
4468        if (!mode)
4469                mode = S_IFREG | 0644;
4470        if (create_one_file(state, path, mode, buf, size))
4471                return -1;
4472
4473        if (patch->conflicted_threeway)
4474                return add_conflicted_stages_file(state, patch);
4475        else
4476                return add_index_file(state, path, mode, buf, size);
4477}
4478
4479/* phase zero is to remove, phase one is to create */
4480static int write_out_one_result(struct apply_state *state,
4481                                struct patch *patch,
4482                                int phase)
4483{
4484        if (patch->is_delete > 0) {
4485                if (phase == 0)
4486                        return remove_file(state, patch, 1);
4487                return 0;
4488        }
4489        if (patch->is_new > 0 || patch->is_copy) {
4490                if (phase == 1)
4491                        return create_file(state, patch);
4492                return 0;
4493        }
4494        /*
4495         * Rename or modification boils down to the same
4496         * thing: remove the old, write the new
4497         */
4498        if (phase == 0)
4499                return remove_file(state, patch, patch->is_rename);
4500        if (phase == 1)
4501                return create_file(state, patch);
4502        return 0;
4503}
4504
4505static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4506{
4507        FILE *rej;
4508        char namebuf[PATH_MAX];
4509        struct fragment *frag;
4510        int cnt = 0;
4511        struct strbuf sb = STRBUF_INIT;
4512
4513        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4514                if (!frag->rejected)
4515                        continue;
4516                cnt++;
4517        }
4518
4519        if (!cnt) {
4520                if (state->apply_verbosity > verbosity_normal)
4521                        say_patch_name(stderr,
4522                                       _("Applied patch %s cleanly."), patch);
4523                return 0;
4524        }
4525
4526        /* This should not happen, because a removal patch that leaves
4527         * contents are marked "rejected" at the patch level.
4528         */
4529        if (!patch->new_name)
4530                die(_("internal error"));
4531
4532        /* Say this even without --verbose */
4533        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4534                            "Applying patch %%s with %d rejects...",
4535                            cnt),
4536                    cnt);
4537        if (state->apply_verbosity > verbosity_silent)
4538                say_patch_name(stderr, sb.buf, patch);
4539        strbuf_release(&sb);
4540
4541        cnt = strlen(patch->new_name);
4542        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4543                cnt = ARRAY_SIZE(namebuf) - 5;
4544                warning(_("truncating .rej filename to %.*s.rej"),
4545                        cnt - 1, patch->new_name);
4546        }
4547        memcpy(namebuf, patch->new_name, cnt);
4548        memcpy(namebuf + cnt, ".rej", 5);
4549
4550        rej = fopen(namebuf, "w");
4551        if (!rej)
4552                return error_errno(_("cannot open %s"), namebuf);
4553
4554        /* Normal git tools never deal with .rej, so do not pretend
4555         * this is a git patch by saying --git or giving extended
4556         * headers.  While at it, maybe please "kompare" that wants
4557         * the trailing TAB and some garbage at the end of line ;-).
4558         */
4559        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4560                patch->new_name, patch->new_name);
4561        for (cnt = 1, frag = patch->fragments;
4562             frag;
4563             cnt++, frag = frag->next) {
4564                if (!frag->rejected) {
4565                        if (state->apply_verbosity > verbosity_silent)
4566                                fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4567                        continue;
4568                }
4569                if (state->apply_verbosity > verbosity_silent)
4570                        fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4571                fprintf(rej, "%.*s", frag->size, frag->patch);
4572                if (frag->patch[frag->size-1] != '\n')
4573                        fputc('\n', rej);
4574        }
4575        fclose(rej);
4576        return -1;
4577}
4578
4579/*
4580 * Returns:
4581 *  -1 if an error happened
4582 *   0 if the patch applied cleanly
4583 *   1 if the patch did not apply cleanly
4584 */
4585static int write_out_results(struct apply_state *state, struct patch *list)
4586{
4587        int phase;
4588        int errs = 0;
4589        struct patch *l;
4590        struct string_list cpath = STRING_LIST_INIT_DUP;
4591
4592        for (phase = 0; phase < 2; phase++) {
4593                l = list;
4594                while (l) {
4595                        if (l->rejected)
4596                                errs = 1;
4597                        else {
4598                                if (write_out_one_result(state, l, phase)) {
4599                                        string_list_clear(&cpath, 0);
4600                                        return -1;
4601                                }
4602                                if (phase == 1) {
4603                                        if (write_out_one_reject(state, l))
4604                                                errs = 1;
4605                                        if (l->conflicted_threeway) {
4606                                                string_list_append(&cpath, l->new_name);
4607                                                errs = 1;
4608                                        }
4609                                }
4610                        }
4611                        l = l->next;
4612                }
4613        }
4614
4615        if (cpath.nr) {
4616                struct string_list_item *item;
4617
4618                string_list_sort(&cpath);
4619                if (state->apply_verbosity > verbosity_silent) {
4620                        for_each_string_list_item(item, &cpath)
4621                                fprintf(stderr, "U %s\n", item->string);
4622                }
4623                string_list_clear(&cpath, 0);
4624
4625                rerere(0);
4626        }
4627
4628        return errs;
4629}
4630
4631/*
4632 * Try to apply a patch.
4633 *
4634 * Returns:
4635 *  -128 if a bad error happened (like patch unreadable)
4636 *  -1 if patch did not apply and user cannot deal with it
4637 *   0 if the patch applied
4638 *   1 if the patch did not apply but user might fix it
4639 */
4640static int apply_patch(struct apply_state *state,
4641                       int fd,
4642                       const char *filename,
4643                       int options)
4644{
4645        size_t offset;
4646        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4647        struct patch *list = NULL, **listp = &list;
4648        int skipped_patch = 0;
4649        int res = 0;
4650
4651        state->patch_input_file = filename;
4652        if (read_patch_file(&buf, fd) < 0)
4653                return -128;
4654        offset = 0;
4655        while (offset < buf.len) {
4656                struct patch *patch;
4657                int nr;
4658
4659                patch = xcalloc(1, sizeof(*patch));
4660                patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4661                patch->recount =  !!(options & APPLY_OPT_RECOUNT);
4662                nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4663                if (nr < 0) {
4664                        free_patch(patch);
4665                        if (nr == -128) {
4666                                res = -128;
4667                                goto end;
4668                        }
4669                        break;
4670                }
4671                if (state->apply_in_reverse)
4672                        reverse_patches(patch);
4673                if (use_patch(state, patch)) {
4674                        patch_stats(state, patch);
4675                        *listp = patch;
4676                        listp = &patch->next;
4677                }
4678                else {
4679                        if (state->apply_verbosity > verbosity_normal)
4680                                say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4681                        free_patch(patch);
4682                        skipped_patch++;
4683                }
4684                offset += nr;
4685        }
4686
4687        if (!list && !skipped_patch) {
4688                error(_("unrecognized input"));
4689                res = -128;
4690                goto end;
4691        }
4692
4693        if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4694                state->apply = 0;
4695
4696        state->update_index = state->check_index && state->apply;
4697        if (state->update_index && state->newfd < 0) {
4698                if (state->index_file)
4699                        state->newfd = hold_lock_file_for_update(state->lock_file,
4700                                                                 state->index_file,
4701                                                                 LOCK_DIE_ON_ERROR);
4702                else
4703                        state->newfd = hold_locked_index(state->lock_file, LOCK_DIE_ON_ERROR);
4704        }
4705
4706        if (state->check_index && read_apply_cache(state) < 0) {
4707                error(_("unable to read index file"));
4708                res = -128;
4709                goto end;
4710        }
4711
4712        if (state->check || state->apply) {
4713                int r = check_patch_list(state, list);
4714                if (r == -128) {
4715                        res = -128;
4716                        goto end;
4717                }
4718                if (r < 0 && !state->apply_with_reject) {
4719                        res = -1;
4720                        goto end;
4721                }
4722        }
4723
4724        if (state->apply) {
4725                int write_res = write_out_results(state, list);
4726                if (write_res < 0) {
4727                        res = -128;
4728                        goto end;
4729                }
4730                if (write_res > 0) {
4731                        /* with --3way, we still need to write the index out */
4732                        res = state->apply_with_reject ? -1 : 1;
4733                        goto end;
4734                }
4735        }
4736
4737        if (state->fake_ancestor &&
4738            build_fake_ancestor(state, list)) {
4739                res = -128;
4740                goto end;
4741        }
4742
4743        if (state->diffstat && state->apply_verbosity > verbosity_silent)
4744                stat_patch_list(state, list);
4745
4746        if (state->numstat && state->apply_verbosity > verbosity_silent)
4747                numstat_patch_list(state, list);
4748
4749        if (state->summary && state->apply_verbosity > verbosity_silent)
4750                summary_patch_list(list);
4751
4752end:
4753        free_patch_list(list);
4754        strbuf_release(&buf);
4755        string_list_clear(&state->fn_table, 0);
4756        return res;
4757}
4758
4759static int apply_option_parse_exclude(const struct option *opt,
4760                                      const char *arg, int unset)
4761{
4762        struct apply_state *state = opt->value;
4763        add_name_limit(state, arg, 1);
4764        return 0;
4765}
4766
4767static int apply_option_parse_include(const struct option *opt,
4768                                      const char *arg, int unset)
4769{
4770        struct apply_state *state = opt->value;
4771        add_name_limit(state, arg, 0);
4772        state->has_include = 1;
4773        return 0;
4774}
4775
4776static int apply_option_parse_p(const struct option *opt,
4777                                const char *arg,
4778                                int unset)
4779{
4780        struct apply_state *state = opt->value;
4781        state->p_value = atoi(arg);
4782        state->p_value_known = 1;
4783        return 0;
4784}
4785
4786static int apply_option_parse_space_change(const struct option *opt,
4787                                           const char *arg, int unset)
4788{
4789        struct apply_state *state = opt->value;
4790        if (unset)
4791                state->ws_ignore_action = ignore_ws_none;
4792        else
4793                state->ws_ignore_action = ignore_ws_change;
4794        return 0;
4795}
4796
4797static int apply_option_parse_whitespace(const struct option *opt,
4798                                         const char *arg, int unset)
4799{
4800        struct apply_state *state = opt->value;
4801        state->whitespace_option = arg;
4802        if (parse_whitespace_option(state, arg))
4803                exit(1);
4804        return 0;
4805}
4806
4807static int apply_option_parse_directory(const struct option *opt,
4808                                        const char *arg, int unset)
4809{
4810        struct apply_state *state = opt->value;
4811        strbuf_reset(&state->root);
4812        strbuf_addstr(&state->root, arg);
4813        strbuf_complete(&state->root, '/');
4814        return 0;
4815}
4816
4817int apply_all_patches(struct apply_state *state,
4818                      int argc,
4819                      const char **argv,
4820                      int options)
4821{
4822        int i;
4823        int res;
4824        int errs = 0;
4825        int read_stdin = 1;
4826
4827        for (i = 0; i < argc; i++) {
4828                const char *arg = argv[i];
4829                char *to_free = NULL;
4830                int fd;
4831
4832                if (!strcmp(arg, "-")) {
4833                        res = apply_patch(state, 0, "<stdin>", options);
4834                        if (res < 0)
4835                                goto end;
4836                        errs |= res;
4837                        read_stdin = 0;
4838                        continue;
4839                } else
4840                        arg = to_free = prefix_filename(state->prefix, arg);
4841
4842                fd = open(arg, O_RDONLY);
4843                if (fd < 0) {
4844                        error(_("can't open patch '%s': %s"), arg, strerror(errno));
4845                        res = -128;
4846                        free(to_free);
4847                        goto end;
4848                }
4849                read_stdin = 0;
4850                set_default_whitespace_mode(state);
4851                res = apply_patch(state, fd, arg, options);
4852                close(fd);
4853                free(to_free);
4854                if (res < 0)
4855                        goto end;
4856                errs |= res;
4857        }
4858        set_default_whitespace_mode(state);
4859        if (read_stdin) {
4860                res = apply_patch(state, 0, "<stdin>", options);
4861                if (res < 0)
4862                        goto end;
4863                errs |= res;
4864        }
4865
4866        if (state->whitespace_error) {
4867                if (state->squelch_whitespace_errors &&
4868                    state->squelch_whitespace_errors < state->whitespace_error) {
4869                        int squelched =
4870                                state->whitespace_error - state->squelch_whitespace_errors;
4871                        warning(Q_("squelched %d whitespace error",
4872                                   "squelched %d whitespace errors",
4873                                   squelched),
4874                                squelched);
4875                }
4876                if (state->ws_error_action == die_on_ws_error) {
4877                        error(Q_("%d line adds whitespace errors.",
4878                                 "%d lines add whitespace errors.",
4879                                 state->whitespace_error),
4880                              state->whitespace_error);
4881                        res = -128;
4882                        goto end;
4883                }
4884                if (state->applied_after_fixing_ws && state->apply)
4885                        warning(Q_("%d line applied after"
4886                                   " fixing whitespace errors.",
4887                                   "%d lines applied after"
4888                                   " fixing whitespace errors.",
4889                                   state->applied_after_fixing_ws),
4890                                state->applied_after_fixing_ws);
4891                else if (state->whitespace_error)
4892                        warning(Q_("%d line adds whitespace errors.",
4893                                   "%d lines add whitespace errors.",
4894                                   state->whitespace_error),
4895                                state->whitespace_error);
4896        }
4897
4898        if (state->update_index) {
4899                res = write_locked_index(&the_index, state->lock_file, COMMIT_LOCK);
4900                if (res) {
4901                        error(_("Unable to write new index file"));
4902                        res = -128;
4903                        goto end;
4904                }
4905                state->newfd = -1;
4906        }
4907
4908        res = !!errs;
4909
4910end:
4911        if (state->newfd >= 0) {
4912                rollback_lock_file(state->lock_file);
4913                state->newfd = -1;
4914        }
4915
4916        if (state->apply_verbosity <= verbosity_silent) {
4917                set_error_routine(state->saved_error_routine);
4918                set_warn_routine(state->saved_warn_routine);
4919        }
4920
4921        if (res > -1)
4922                return res;
4923        return (res == -1 ? 1 : 128);
4924}
4925
4926int apply_parse_options(int argc, const char **argv,
4927                        struct apply_state *state,
4928                        int *force_apply, int *options,
4929                        const char * const *apply_usage)
4930{
4931        struct option builtin_apply_options[] = {
4932                { OPTION_CALLBACK, 0, "exclude", state, N_("path"),
4933                        N_("don't apply changes matching the given path"),
4934                        0, apply_option_parse_exclude },
4935                { OPTION_CALLBACK, 0, "include", state, N_("path"),
4936                        N_("apply changes matching the given path"),
4937                        0, apply_option_parse_include },
4938                { OPTION_CALLBACK, 'p', NULL, state, N_("num"),
4939                        N_("remove <num> leading slashes from traditional diff paths"),
4940                        0, apply_option_parse_p },
4941                OPT_BOOL(0, "no-add", &state->no_add,
4942                        N_("ignore additions made by the patch")),
4943                OPT_BOOL(0, "stat", &state->diffstat,
4944                        N_("instead of applying the patch, output diffstat for the input")),
4945                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4946                OPT_NOOP_NOARG(0, "binary"),
4947                OPT_BOOL(0, "numstat", &state->numstat,
4948                        N_("show number of added and deleted lines in decimal notation")),
4949                OPT_BOOL(0, "summary", &state->summary,
4950                        N_("instead of applying the patch, output a summary for the input")),
4951                OPT_BOOL(0, "check", &state->check,
4952                        N_("instead of applying the patch, see if the patch is applicable")),
4953                OPT_BOOL(0, "index", &state->check_index,
4954                        N_("make sure the patch is applicable to the current index")),
4955                OPT_BOOL(0, "cached", &state->cached,
4956                        N_("apply a patch without touching the working tree")),
4957                OPT_BOOL(0, "unsafe-paths", &state->unsafe_paths,
4958                        N_("accept a patch that touches outside the working area")),
4959                OPT_BOOL(0, "apply", force_apply,
4960                        N_("also apply the patch (use with --stat/--summary/--check)")),
4961                OPT_BOOL('3', "3way", &state->threeway,
4962                         N_( "attempt three-way merge if a patch does not apply")),
4963                OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
4964                        N_("build a temporary index based on embedded index information")),
4965                /* Think twice before adding "--nul" synonym to this */
4966                OPT_SET_INT('z', NULL, &state->line_termination,
4967                        N_("paths are separated with NUL character"), '\0'),
4968                OPT_INTEGER('C', NULL, &state->p_context,
4969                                N_("ensure at least <n> lines of context match")),
4970                { OPTION_CALLBACK, 0, "whitespace", state, N_("action"),
4971                        N_("detect new or modified lines that have whitespace errors"),
4972                        0, apply_option_parse_whitespace },
4973                { OPTION_CALLBACK, 0, "ignore-space-change", state, NULL,
4974                        N_("ignore changes in whitespace when finding context"),
4975                        PARSE_OPT_NOARG, apply_option_parse_space_change },
4976                { OPTION_CALLBACK, 0, "ignore-whitespace", state, NULL,
4977                        N_("ignore changes in whitespace when finding context"),
4978                        PARSE_OPT_NOARG, apply_option_parse_space_change },
4979                OPT_BOOL('R', "reverse", &state->apply_in_reverse,
4980                        N_("apply the patch in reverse")),
4981                OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
4982                        N_("don't expect at least one line of context")),
4983                OPT_BOOL(0, "reject", &state->apply_with_reject,
4984                        N_("leave the rejected hunks in corresponding *.rej files")),
4985                OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
4986                        N_("allow overlapping hunks")),
4987                OPT__VERBOSE(&state->apply_verbosity, N_("be verbose")),
4988                OPT_BIT(0, "inaccurate-eof", options,
4989                        N_("tolerate incorrectly detected missing new-line at the end of file"),
4990                        APPLY_OPT_INACCURATE_EOF),
4991                OPT_BIT(0, "recount", options,
4992                        N_("do not trust the line counts in the hunk headers"),
4993                        APPLY_OPT_RECOUNT),
4994                { OPTION_CALLBACK, 0, "directory", state, N_("root"),
4995                        N_("prepend <root> to all filenames"),
4996                        0, apply_option_parse_directory },
4997                OPT_END()
4998        };
4999
5000        return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);
5001}