apply.con commit Merge branch 'jk/hashcmp-memcmp' into next (27c4aa5)
   1/*
   2 * apply.c
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This applies patches on top of some (arbitrary) version of the SCM.
   7 *
   8 */
   9
  10#include "cache.h"
  11#include "config.h"
  12#include "blob.h"
  13#include "delta.h"
  14#include "diff.h"
  15#include "dir.h"
  16#include "xdiff-interface.h"
  17#include "ll-merge.h"
  18#include "lockfile.h"
  19#include "parse-options.h"
  20#include "quote.h"
  21#include "rerere.h"
  22#include "apply.h"
  23
  24static void git_apply_config(void)
  25{
  26        git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
  27        git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
  28        git_config(git_default_config, NULL);
  29}
  30
  31static int parse_whitespace_option(struct apply_state *state, const char *option)
  32{
  33        if (!option) {
  34                state->ws_error_action = warn_on_ws_error;
  35                return 0;
  36        }
  37        if (!strcmp(option, "warn")) {
  38                state->ws_error_action = warn_on_ws_error;
  39                return 0;
  40        }
  41        if (!strcmp(option, "nowarn")) {
  42                state->ws_error_action = nowarn_ws_error;
  43                return 0;
  44        }
  45        if (!strcmp(option, "error")) {
  46                state->ws_error_action = die_on_ws_error;
  47                return 0;
  48        }
  49        if (!strcmp(option, "error-all")) {
  50                state->ws_error_action = die_on_ws_error;
  51                state->squelch_whitespace_errors = 0;
  52                return 0;
  53        }
  54        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
  55                state->ws_error_action = correct_ws_error;
  56                return 0;
  57        }
  58        return error(_("unrecognized whitespace option '%s'"), option);
  59}
  60
  61static int parse_ignorewhitespace_option(struct apply_state *state,
  62                                                 const char *option)
  63{
  64        if (!option || !strcmp(option, "no") ||
  65            !strcmp(option, "false") || !strcmp(option, "never") ||
  66            !strcmp(option, "none")) {
  67                state->ws_ignore_action = ignore_ws_none;
  68                return 0;
  69        }
  70        if (!strcmp(option, "change")) {
  71                state->ws_ignore_action = ignore_ws_change;
  72                return 0;
  73        }
  74        return error(_("unrecognized whitespace ignore option '%s'"), option);
  75}
  76
  77int init_apply_state(struct apply_state *state,
  78                     const char *prefix,
  79                     struct lock_file *lock_file)
  80{
  81        memset(state, 0, sizeof(*state));
  82        state->prefix = prefix;
  83        state->prefix_length = state->prefix ? strlen(state->prefix) : 0;
  84        state->lock_file = lock_file;
  85        state->newfd = -1;
  86        state->apply = 1;
  87        state->line_termination = '\n';
  88        state->p_value = 1;
  89        state->p_context = UINT_MAX;
  90        state->squelch_whitespace_errors = 5;
  91        state->ws_error_action = warn_on_ws_error;
  92        state->ws_ignore_action = ignore_ws_none;
  93        state->linenr = 1;
  94        string_list_init(&state->fn_table, 0);
  95        string_list_init(&state->limit_by_name, 0);
  96        string_list_init(&state->symlink_changes, 0);
  97        strbuf_init(&state->root, 0);
  98
  99        git_apply_config();
 100        if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
 101                return -1;
 102        if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
 103                return -1;
 104        return 0;
 105}
 106
 107void clear_apply_state(struct apply_state *state)
 108{
 109        string_list_clear(&state->limit_by_name, 0);
 110        string_list_clear(&state->symlink_changes, 0);
 111        strbuf_release(&state->root);
 112
 113        /* &state->fn_table is cleared at the end of apply_patch() */
 114}
 115
 116static void mute_routine(const char *msg, va_list params)
 117{
 118        /* do nothing */
 119}
 120
 121int check_apply_state(struct apply_state *state, int force_apply)
 122{
 123        int is_not_gitdir = !startup_info->have_repository;
 124
 125        if (state->apply_with_reject && state->threeway)
 126                return error(_("--reject and --3way cannot be used together."));
 127        if (state->cached && state->threeway)
 128                return error(_("--cached and --3way cannot be used together."));
 129        if (state->threeway) {
 130                if (is_not_gitdir)
 131                        return error(_("--3way outside a repository"));
 132                state->check_index = 1;
 133        }
 134        if (state->apply_with_reject) {
 135                state->apply = 1;
 136                if (state->apply_verbosity == verbosity_normal)
 137                        state->apply_verbosity = verbosity_verbose;
 138        }
 139        if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
 140                state->apply = 0;
 141        if (state->check_index && is_not_gitdir)
 142                return error(_("--index outside a repository"));
 143        if (state->cached) {
 144                if (is_not_gitdir)
 145                        return error(_("--cached outside a repository"));
 146                state->check_index = 1;
 147        }
 148        if (state->check_index)
 149                state->unsafe_paths = 0;
 150        if (!state->lock_file)
 151                return error("BUG: state->lock_file should not be NULL");
 152
 153        if (state->apply_verbosity <= verbosity_silent) {
 154                state->saved_error_routine = get_error_routine();
 155                state->saved_warn_routine = get_warn_routine();
 156                set_error_routine(mute_routine);
 157                set_warn_routine(mute_routine);
 158        }
 159
 160        return 0;
 161}
 162
 163static void set_default_whitespace_mode(struct apply_state *state)
 164{
 165        if (!state->whitespace_option && !apply_default_whitespace)
 166                state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
 167}
 168
 169/*
 170 * This represents one "hunk" from a patch, starting with
 171 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 172 * patch text is pointed at by patch, and its byte length
 173 * is stored in size.  leading and trailing are the number
 174 * of context lines.
 175 */
 176struct fragment {
 177        unsigned long leading, trailing;
 178        unsigned long oldpos, oldlines;
 179        unsigned long newpos, newlines;
 180        /*
 181         * 'patch' is usually borrowed from buf in apply_patch(),
 182         * but some codepaths store an allocated buffer.
 183         */
 184        const char *patch;
 185        unsigned free_patch:1,
 186                rejected:1;
 187        int size;
 188        int linenr;
 189        struct fragment *next;
 190};
 191
 192/*
 193 * When dealing with a binary patch, we reuse "leading" field
 194 * to store the type of the binary hunk, either deflated "delta"
 195 * or deflated "literal".
 196 */
 197#define binary_patch_method leading
 198#define BINARY_DELTA_DEFLATED   1
 199#define BINARY_LITERAL_DEFLATED 2
 200
 201/*
 202 * This represents a "patch" to a file, both metainfo changes
 203 * such as creation/deletion, filemode and content changes represented
 204 * as a series of fragments.
 205 */
 206struct patch {
 207        char *new_name, *old_name, *def_name;
 208        unsigned int old_mode, new_mode;
 209        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 210        int rejected;
 211        unsigned ws_rule;
 212        int lines_added, lines_deleted;
 213        int score;
 214        int extension_linenr; /* first line specifying delete/new/rename/copy */
 215        unsigned int is_toplevel_relative:1;
 216        unsigned int inaccurate_eof:1;
 217        unsigned int is_binary:1;
 218        unsigned int is_copy:1;
 219        unsigned int is_rename:1;
 220        unsigned int recount:1;
 221        unsigned int conflicted_threeway:1;
 222        unsigned int direct_to_threeway:1;
 223        struct fragment *fragments;
 224        char *result;
 225        size_t resultsize;
 226        char old_sha1_prefix[41];
 227        char new_sha1_prefix[41];
 228        struct patch *next;
 229
 230        /* three-way fallback result */
 231        struct object_id threeway_stage[3];
 232};
 233
 234static void free_fragment_list(struct fragment *list)
 235{
 236        while (list) {
 237                struct fragment *next = list->next;
 238                if (list->free_patch)
 239                        free((char *)list->patch);
 240                free(list);
 241                list = next;
 242        }
 243}
 244
 245static void free_patch(struct patch *patch)
 246{
 247        free_fragment_list(patch->fragments);
 248        free(patch->def_name);
 249        free(patch->old_name);
 250        free(patch->new_name);
 251        free(patch->result);
 252        free(patch);
 253}
 254
 255static void free_patch_list(struct patch *list)
 256{
 257        while (list) {
 258                struct patch *next = list->next;
 259                free_patch(list);
 260                list = next;
 261        }
 262}
 263
 264/*
 265 * A line in a file, len-bytes long (includes the terminating LF,
 266 * except for an incomplete line at the end if the file ends with
 267 * one), and its contents hashes to 'hash'.
 268 */
 269struct line {
 270        size_t len;
 271        unsigned hash : 24;
 272        unsigned flag : 8;
 273#define LINE_COMMON     1
 274#define LINE_PATCHED    2
 275};
 276
 277/*
 278 * This represents a "file", which is an array of "lines".
 279 */
 280struct image {
 281        char *buf;
 282        size_t len;
 283        size_t nr;
 284        size_t alloc;
 285        struct line *line_allocated;
 286        struct line *line;
 287};
 288
 289static uint32_t hash_line(const char *cp, size_t len)
 290{
 291        size_t i;
 292        uint32_t h;
 293        for (i = 0, h = 0; i < len; i++) {
 294                if (!isspace(cp[i])) {
 295                        h = h * 3 + (cp[i] & 0xff);
 296                }
 297        }
 298        return h;
 299}
 300
 301/*
 302 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 303 * whitespace difference. Returns 1 if they match, 0 otherwise
 304 */
 305static int fuzzy_matchlines(const char *s1, size_t n1,
 306                            const char *s2, size_t n2)
 307{
 308        const char *last1 = s1 + n1 - 1;
 309        const char *last2 = s2 + n2 - 1;
 310        int result = 0;
 311
 312        /* ignore line endings */
 313        while ((*last1 == '\r') || (*last1 == '\n'))
 314                last1--;
 315        while ((*last2 == '\r') || (*last2 == '\n'))
 316                last2--;
 317
 318        /* skip leading whitespaces, if both begin with whitespace */
 319        if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
 320                while (isspace(*s1) && (s1 <= last1))
 321                        s1++;
 322                while (isspace(*s2) && (s2 <= last2))
 323                        s2++;
 324        }
 325        /* early return if both lines are empty */
 326        if ((s1 > last1) && (s2 > last2))
 327                return 1;
 328        while (!result) {
 329                result = *s1++ - *s2++;
 330                /*
 331                 * Skip whitespace inside. We check for whitespace on
 332                 * both buffers because we don't want "a b" to match
 333                 * "ab"
 334                 */
 335                if (isspace(*s1) && isspace(*s2)) {
 336                        while (isspace(*s1) && s1 <= last1)
 337                                s1++;
 338                        while (isspace(*s2) && s2 <= last2)
 339                                s2++;
 340                }
 341                /*
 342                 * If we reached the end on one side only,
 343                 * lines don't match
 344                 */
 345                if (
 346                    ((s2 > last2) && (s1 <= last1)) ||
 347                    ((s1 > last1) && (s2 <= last2)))
 348                        return 0;
 349                if ((s1 > last1) && (s2 > last2))
 350                        break;
 351        }
 352
 353        return !result;
 354}
 355
 356static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 357{
 358        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 359        img->line_allocated[img->nr].len = len;
 360        img->line_allocated[img->nr].hash = hash_line(bol, len);
 361        img->line_allocated[img->nr].flag = flag;
 362        img->nr++;
 363}
 364
 365/*
 366 * "buf" has the file contents to be patched (read from various sources).
 367 * attach it to "image" and add line-based index to it.
 368 * "image" now owns the "buf".
 369 */
 370static void prepare_image(struct image *image, char *buf, size_t len,
 371                          int prepare_linetable)
 372{
 373        const char *cp, *ep;
 374
 375        memset(image, 0, sizeof(*image));
 376        image->buf = buf;
 377        image->len = len;
 378
 379        if (!prepare_linetable)
 380                return;
 381
 382        ep = image->buf + image->len;
 383        cp = image->buf;
 384        while (cp < ep) {
 385                const char *next;
 386                for (next = cp; next < ep && *next != '\n'; next++)
 387                        ;
 388                if (next < ep)
 389                        next++;
 390                add_line_info(image, cp, next - cp, 0);
 391                cp = next;
 392        }
 393        image->line = image->line_allocated;
 394}
 395
 396static void clear_image(struct image *image)
 397{
 398        free(image->buf);
 399        free(image->line_allocated);
 400        memset(image, 0, sizeof(*image));
 401}
 402
 403/* fmt must contain _one_ %s and no other substitution */
 404static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 405{
 406        struct strbuf sb = STRBUF_INIT;
 407
 408        if (patch->old_name && patch->new_name &&
 409            strcmp(patch->old_name, patch->new_name)) {
 410                quote_c_style(patch->old_name, &sb, NULL, 0);
 411                strbuf_addstr(&sb, " => ");
 412                quote_c_style(patch->new_name, &sb, NULL, 0);
 413        } else {
 414                const char *n = patch->new_name;
 415                if (!n)
 416                        n = patch->old_name;
 417                quote_c_style(n, &sb, NULL, 0);
 418        }
 419        fprintf(output, fmt, sb.buf);
 420        fputc('\n', output);
 421        strbuf_release(&sb);
 422}
 423
 424#define SLOP (16)
 425
 426static int read_patch_file(struct strbuf *sb, int fd)
 427{
 428        if (strbuf_read(sb, fd, 0) < 0)
 429                return error_errno("git apply: failed to read");
 430
 431        /*
 432         * Make sure that we have some slop in the buffer
 433         * so that we can do speculative "memcmp" etc, and
 434         * see to it that it is NUL-filled.
 435         */
 436        strbuf_grow(sb, SLOP);
 437        memset(sb->buf + sb->len, 0, SLOP);
 438        return 0;
 439}
 440
 441static unsigned long linelen(const char *buffer, unsigned long size)
 442{
 443        unsigned long len = 0;
 444        while (size--) {
 445                len++;
 446                if (*buffer++ == '\n')
 447                        break;
 448        }
 449        return len;
 450}
 451
 452static int is_dev_null(const char *str)
 453{
 454        return skip_prefix(str, "/dev/null", &str) && isspace(*str);
 455}
 456
 457#define TERM_SPACE      1
 458#define TERM_TAB        2
 459
 460static int name_terminate(int c, int terminate)
 461{
 462        if (c == ' ' && !(terminate & TERM_SPACE))
 463                return 0;
 464        if (c == '\t' && !(terminate & TERM_TAB))
 465                return 0;
 466
 467        return 1;
 468}
 469
 470/* remove double slashes to make --index work with such filenames */
 471static char *squash_slash(char *name)
 472{
 473        int i = 0, j = 0;
 474
 475        if (!name)
 476                return NULL;
 477
 478        while (name[i]) {
 479                if ((name[j++] = name[i++]) == '/')
 480                        while (name[i] == '/')
 481                                i++;
 482        }
 483        name[j] = '\0';
 484        return name;
 485}
 486
 487static char *find_name_gnu(struct apply_state *state,
 488                           const char *line,
 489                           const char *def,
 490                           int p_value)
 491{
 492        struct strbuf name = STRBUF_INIT;
 493        char *cp;
 494
 495        /*
 496         * Proposed "new-style" GNU patch/diff format; see
 497         * http://marc.info/?l=git&m=112927316408690&w=2
 498         */
 499        if (unquote_c_style(&name, line, NULL)) {
 500                strbuf_release(&name);
 501                return NULL;
 502        }
 503
 504        for (cp = name.buf; p_value; p_value--) {
 505                cp = strchr(cp, '/');
 506                if (!cp) {
 507                        strbuf_release(&name);
 508                        return NULL;
 509                }
 510                cp++;
 511        }
 512
 513        strbuf_remove(&name, 0, cp - name.buf);
 514        if (state->root.len)
 515                strbuf_insert(&name, 0, state->root.buf, state->root.len);
 516        return squash_slash(strbuf_detach(&name, NULL));
 517}
 518
 519static size_t sane_tz_len(const char *line, size_t len)
 520{
 521        const char *tz, *p;
 522
 523        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 524                return 0;
 525        tz = line + len - strlen(" +0500");
 526
 527        if (tz[1] != '+' && tz[1] != '-')
 528                return 0;
 529
 530        for (p = tz + 2; p != line + len; p++)
 531                if (!isdigit(*p))
 532                        return 0;
 533
 534        return line + len - tz;
 535}
 536
 537static size_t tz_with_colon_len(const char *line, size_t len)
 538{
 539        const char *tz, *p;
 540
 541        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 542                return 0;
 543        tz = line + len - strlen(" +08:00");
 544
 545        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 546                return 0;
 547        p = tz + 2;
 548        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 549            !isdigit(*p++) || !isdigit(*p++))
 550                return 0;
 551
 552        return line + len - tz;
 553}
 554
 555static size_t date_len(const char *line, size_t len)
 556{
 557        const char *date, *p;
 558
 559        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 560                return 0;
 561        p = date = line + len - strlen("72-02-05");
 562
 563        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 564            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 565            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 566                return 0;
 567
 568        if (date - line >= strlen("19") &&
 569            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 570                date -= strlen("19");
 571
 572        return line + len - date;
 573}
 574
 575static size_t short_time_len(const char *line, size_t len)
 576{
 577        const char *time, *p;
 578
 579        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 580                return 0;
 581        p = time = line + len - strlen(" 07:01:32");
 582
 583        /* Permit 1-digit hours? */
 584        if (*p++ != ' ' ||
 585            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 586            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 587            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 588                return 0;
 589
 590        return line + len - time;
 591}
 592
 593static size_t fractional_time_len(const char *line, size_t len)
 594{
 595        const char *p;
 596        size_t n;
 597
 598        /* Expected format: 19:41:17.620000023 */
 599        if (!len || !isdigit(line[len - 1]))
 600                return 0;
 601        p = line + len - 1;
 602
 603        /* Fractional seconds. */
 604        while (p > line && isdigit(*p))
 605                p--;
 606        if (*p != '.')
 607                return 0;
 608
 609        /* Hours, minutes, and whole seconds. */
 610        n = short_time_len(line, p - line);
 611        if (!n)
 612                return 0;
 613
 614        return line + len - p + n;
 615}
 616
 617static size_t trailing_spaces_len(const char *line, size_t len)
 618{
 619        const char *p;
 620
 621        /* Expected format: ' ' x (1 or more)  */
 622        if (!len || line[len - 1] != ' ')
 623                return 0;
 624
 625        p = line + len;
 626        while (p != line) {
 627                p--;
 628                if (*p != ' ')
 629                        return line + len - (p + 1);
 630        }
 631
 632        /* All spaces! */
 633        return len;
 634}
 635
 636static size_t diff_timestamp_len(const char *line, size_t len)
 637{
 638        const char *end = line + len;
 639        size_t n;
 640
 641        /*
 642         * Posix: 2010-07-05 19:41:17
 643         * GNU: 2010-07-05 19:41:17.620000023 -0500
 644         */
 645
 646        if (!isdigit(end[-1]))
 647                return 0;
 648
 649        n = sane_tz_len(line, end - line);
 650        if (!n)
 651                n = tz_with_colon_len(line, end - line);
 652        end -= n;
 653
 654        n = short_time_len(line, end - line);
 655        if (!n)
 656                n = fractional_time_len(line, end - line);
 657        end -= n;
 658
 659        n = date_len(line, end - line);
 660        if (!n) /* No date.  Too bad. */
 661                return 0;
 662        end -= n;
 663
 664        if (end == line)        /* No space before date. */
 665                return 0;
 666        if (end[-1] == '\t') {  /* Success! */
 667                end--;
 668                return line + len - end;
 669        }
 670        if (end[-1] != ' ')     /* No space before date. */
 671                return 0;
 672
 673        /* Whitespace damage. */
 674        end -= trailing_spaces_len(line, end - line);
 675        return line + len - end;
 676}
 677
 678static char *find_name_common(struct apply_state *state,
 679                              const char *line,
 680                              const char *def,
 681                              int p_value,
 682                              const char *end,
 683                              int terminate)
 684{
 685        int len;
 686        const char *start = NULL;
 687
 688        if (p_value == 0)
 689                start = line;
 690        while (line != end) {
 691                char c = *line;
 692
 693                if (!end && isspace(c)) {
 694                        if (c == '\n')
 695                                break;
 696                        if (name_terminate(c, terminate))
 697                                break;
 698                }
 699                line++;
 700                if (c == '/' && !--p_value)
 701                        start = line;
 702        }
 703        if (!start)
 704                return squash_slash(xstrdup_or_null(def));
 705        len = line - start;
 706        if (!len)
 707                return squash_slash(xstrdup_or_null(def));
 708
 709        /*
 710         * Generally we prefer the shorter name, especially
 711         * if the other one is just a variation of that with
 712         * something else tacked on to the end (ie "file.orig"
 713         * or "file~").
 714         */
 715        if (def) {
 716                int deflen = strlen(def);
 717                if (deflen < len && !strncmp(start, def, deflen))
 718                        return squash_slash(xstrdup(def));
 719        }
 720
 721        if (state->root.len) {
 722                char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
 723                return squash_slash(ret);
 724        }
 725
 726        return squash_slash(xmemdupz(start, len));
 727}
 728
 729static char *find_name(struct apply_state *state,
 730                       const char *line,
 731                       char *def,
 732                       int p_value,
 733                       int terminate)
 734{
 735        if (*line == '"') {
 736                char *name = find_name_gnu(state, line, def, p_value);
 737                if (name)
 738                        return name;
 739        }
 740
 741        return find_name_common(state, line, def, p_value, NULL, terminate);
 742}
 743
 744static char *find_name_traditional(struct apply_state *state,
 745                                   const char *line,
 746                                   char *def,
 747                                   int p_value)
 748{
 749        size_t len;
 750        size_t date_len;
 751
 752        if (*line == '"') {
 753                char *name = find_name_gnu(state, line, def, p_value);
 754                if (name)
 755                        return name;
 756        }
 757
 758        len = strchrnul(line, '\n') - line;
 759        date_len = diff_timestamp_len(line, len);
 760        if (!date_len)
 761                return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
 762        len -= date_len;
 763
 764        return find_name_common(state, line, def, p_value, line + len, 0);
 765}
 766
 767/*
 768 * Given the string after "--- " or "+++ ", guess the appropriate
 769 * p_value for the given patch.
 770 */
 771static int guess_p_value(struct apply_state *state, const char *nameline)
 772{
 773        char *name, *cp;
 774        int val = -1;
 775
 776        if (is_dev_null(nameline))
 777                return -1;
 778        name = find_name_traditional(state, nameline, NULL, 0);
 779        if (!name)
 780                return -1;
 781        cp = strchr(name, '/');
 782        if (!cp)
 783                val = 0;
 784        else if (state->prefix) {
 785                /*
 786                 * Does it begin with "a/$our-prefix" and such?  Then this is
 787                 * very likely to apply to our directory.
 788                 */
 789                if (!strncmp(name, state->prefix, state->prefix_length))
 790                        val = count_slashes(state->prefix);
 791                else {
 792                        cp++;
 793                        if (!strncmp(cp, state->prefix, state->prefix_length))
 794                                val = count_slashes(state->prefix) + 1;
 795                }
 796        }
 797        free(name);
 798        return val;
 799}
 800
 801/*
 802 * Does the ---/+++ line have the POSIX timestamp after the last HT?
 803 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 804 * this such a timestamp?
 805 */
 806static int has_epoch_timestamp(const char *nameline)
 807{
 808        /*
 809         * We are only interested in epoch timestamp; any non-zero
 810         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 811         * For the same reason, the date must be either 1969-12-31 or
 812         * 1970-01-01, and the seconds part must be "00".
 813         */
 814        const char stamp_regexp[] =
 815                "^(1969-12-31|1970-01-01)"
 816                " "
 817                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 818                " "
 819                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 820        const char *timestamp = NULL, *cp, *colon;
 821        static regex_t *stamp;
 822        regmatch_t m[10];
 823        int zoneoffset;
 824        int hourminute;
 825        int status;
 826
 827        for (cp = nameline; *cp != '\n'; cp++) {
 828                if (*cp == '\t')
 829                        timestamp = cp + 1;
 830        }
 831        if (!timestamp)
 832                return 0;
 833        if (!stamp) {
 834                stamp = xmalloc(sizeof(*stamp));
 835                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 836                        warning(_("Cannot prepare timestamp regexp %s"),
 837                                stamp_regexp);
 838                        return 0;
 839                }
 840        }
 841
 842        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 843        if (status) {
 844                if (status != REG_NOMATCH)
 845                        warning(_("regexec returned %d for input: %s"),
 846                                status, timestamp);
 847                return 0;
 848        }
 849
 850        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 851        if (*colon == ':')
 852                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 853        else
 854                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 855        if (timestamp[m[3].rm_so] == '-')
 856                zoneoffset = -zoneoffset;
 857
 858        /*
 859         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 860         * (west of GMT) or 1970-01-01 (east of GMT)
 861         */
 862        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 863            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 864                return 0;
 865
 866        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 867                      strtol(timestamp + 14, NULL, 10) -
 868                      zoneoffset);
 869
 870        return ((zoneoffset < 0 && hourminute == 1440) ||
 871                (0 <= zoneoffset && !hourminute));
 872}
 873
 874/*
 875 * Get the name etc info from the ---/+++ lines of a traditional patch header
 876 *
 877 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 878 * files, we can happily check the index for a match, but for creating a
 879 * new file we should try to match whatever "patch" does. I have no idea.
 880 */
 881static int parse_traditional_patch(struct apply_state *state,
 882                                   const char *first,
 883                                   const char *second,
 884                                   struct patch *patch)
 885{
 886        char *name;
 887
 888        first += 4;     /* skip "--- " */
 889        second += 4;    /* skip "+++ " */
 890        if (!state->p_value_known) {
 891                int p, q;
 892                p = guess_p_value(state, first);
 893                q = guess_p_value(state, second);
 894                if (p < 0) p = q;
 895                if (0 <= p && p == q) {
 896                        state->p_value = p;
 897                        state->p_value_known = 1;
 898                }
 899        }
 900        if (is_dev_null(first)) {
 901                patch->is_new = 1;
 902                patch->is_delete = 0;
 903                name = find_name_traditional(state, second, NULL, state->p_value);
 904                patch->new_name = name;
 905        } else if (is_dev_null(second)) {
 906                patch->is_new = 0;
 907                patch->is_delete = 1;
 908                name = find_name_traditional(state, first, NULL, state->p_value);
 909                patch->old_name = name;
 910        } else {
 911                char *first_name;
 912                first_name = find_name_traditional(state, first, NULL, state->p_value);
 913                name = find_name_traditional(state, second, first_name, state->p_value);
 914                free(first_name);
 915                if (has_epoch_timestamp(first)) {
 916                        patch->is_new = 1;
 917                        patch->is_delete = 0;
 918                        patch->new_name = name;
 919                } else if (has_epoch_timestamp(second)) {
 920                        patch->is_new = 0;
 921                        patch->is_delete = 1;
 922                        patch->old_name = name;
 923                } else {
 924                        patch->old_name = name;
 925                        patch->new_name = xstrdup_or_null(name);
 926                }
 927        }
 928        if (!name)
 929                return error(_("unable to find filename in patch at line %d"), state->linenr);
 930
 931        return 0;
 932}
 933
 934static int gitdiff_hdrend(struct apply_state *state,
 935                          const char *line,
 936                          struct patch *patch)
 937{
 938        return 1;
 939}
 940
 941/*
 942 * We're anal about diff header consistency, to make
 943 * sure that we don't end up having strange ambiguous
 944 * patches floating around.
 945 *
 946 * As a result, gitdiff_{old|new}name() will check
 947 * their names against any previous information, just
 948 * to make sure..
 949 */
 950#define DIFF_OLD_NAME 0
 951#define DIFF_NEW_NAME 1
 952
 953static int gitdiff_verify_name(struct apply_state *state,
 954                               const char *line,
 955                               int isnull,
 956                               char **name,
 957                               int side)
 958{
 959        if (!*name && !isnull) {
 960                *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
 961                return 0;
 962        }
 963
 964        if (*name) {
 965                char *another;
 966                if (isnull)
 967                        return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
 968                                     *name, state->linenr);
 969                another = find_name(state, line, NULL, state->p_value, TERM_TAB);
 970                if (!another || strcmp(another, *name)) {
 971                        free(another);
 972                        return error((side == DIFF_NEW_NAME) ?
 973                            _("git apply: bad git-diff - inconsistent new filename on line %d") :
 974                            _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
 975                }
 976                free(another);
 977        } else {
 978                if (!starts_with(line, "/dev/null\n"))
 979                        return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
 980        }
 981
 982        return 0;
 983}
 984
 985static int gitdiff_oldname(struct apply_state *state,
 986                           const char *line,
 987                           struct patch *patch)
 988{
 989        return gitdiff_verify_name(state, line,
 990                                   patch->is_new, &patch->old_name,
 991                                   DIFF_OLD_NAME);
 992}
 993
 994static int gitdiff_newname(struct apply_state *state,
 995                           const char *line,
 996                           struct patch *patch)
 997{
 998        return gitdiff_verify_name(state, line,
 999                                   patch->is_delete, &patch->new_name,
1000                                   DIFF_NEW_NAME);
1001}
1002
1003static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
1004{
1005        char *end;
1006        *mode = strtoul(line, &end, 8);
1007        if (end == line || !isspace(*end))
1008                return error(_("invalid mode on line %d: %s"), linenr, line);
1009        return 0;
1010}
1011
1012static int gitdiff_oldmode(struct apply_state *state,
1013                           const char *line,
1014                           struct patch *patch)
1015{
1016        return parse_mode_line(line, state->linenr, &patch->old_mode);
1017}
1018
1019static int gitdiff_newmode(struct apply_state *state,
1020                           const char *line,
1021                           struct patch *patch)
1022{
1023        return parse_mode_line(line, state->linenr, &patch->new_mode);
1024}
1025
1026static int gitdiff_delete(struct apply_state *state,
1027                          const char *line,
1028                          struct patch *patch)
1029{
1030        patch->is_delete = 1;
1031        free(patch->old_name);
1032        patch->old_name = xstrdup_or_null(patch->def_name);
1033        return gitdiff_oldmode(state, line, patch);
1034}
1035
1036static int gitdiff_newfile(struct apply_state *state,
1037                           const char *line,
1038                           struct patch *patch)
1039{
1040        patch->is_new = 1;
1041        free(patch->new_name);
1042        patch->new_name = xstrdup_or_null(patch->def_name);
1043        return gitdiff_newmode(state, line, patch);
1044}
1045
1046static int gitdiff_copysrc(struct apply_state *state,
1047                           const char *line,
1048                           struct patch *patch)
1049{
1050        patch->is_copy = 1;
1051        free(patch->old_name);
1052        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1053        return 0;
1054}
1055
1056static int gitdiff_copydst(struct apply_state *state,
1057                           const char *line,
1058                           struct patch *patch)
1059{
1060        patch->is_copy = 1;
1061        free(patch->new_name);
1062        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1063        return 0;
1064}
1065
1066static int gitdiff_renamesrc(struct apply_state *state,
1067                             const char *line,
1068                             struct patch *patch)
1069{
1070        patch->is_rename = 1;
1071        free(patch->old_name);
1072        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1073        return 0;
1074}
1075
1076static int gitdiff_renamedst(struct apply_state *state,
1077                             const char *line,
1078                             struct patch *patch)
1079{
1080        patch->is_rename = 1;
1081        free(patch->new_name);
1082        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1083        return 0;
1084}
1085
1086static int gitdiff_similarity(struct apply_state *state,
1087                              const char *line,
1088                              struct patch *patch)
1089{
1090        unsigned long val = strtoul(line, NULL, 10);
1091        if (val <= 100)
1092                patch->score = val;
1093        return 0;
1094}
1095
1096static int gitdiff_dissimilarity(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_index(struct apply_state *state,
1107                         const char *line,
1108                         struct patch *patch)
1109{
1110        /*
1111         * index line is N hexadecimal, "..", N hexadecimal,
1112         * and optional space with octal mode.
1113         */
1114        const char *ptr, *eol;
1115        int len;
1116
1117        ptr = strchr(line, '.');
1118        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1119                return 0;
1120        len = ptr - line;
1121        memcpy(patch->old_sha1_prefix, line, len);
1122        patch->old_sha1_prefix[len] = 0;
1123
1124        line = ptr + 2;
1125        ptr = strchr(line, ' ');
1126        eol = strchrnul(line, '\n');
1127
1128        if (!ptr || eol < ptr)
1129                ptr = eol;
1130        len = ptr - line;
1131
1132        if (40 < len)
1133                return 0;
1134        memcpy(patch->new_sha1_prefix, line, len);
1135        patch->new_sha1_prefix[len] = 0;
1136        if (*ptr == ' ')
1137                return gitdiff_oldmode(state, ptr + 1, patch);
1138        return 0;
1139}
1140
1141/*
1142 * This is normal for a diff that doesn't change anything: we'll fall through
1143 * into the next diff. Tell the parser to break out.
1144 */
1145static int gitdiff_unrecognized(struct apply_state *state,
1146                                const char *line,
1147                                struct patch *patch)
1148{
1149        return 1;
1150}
1151
1152/*
1153 * Skip p_value leading components from "line"; as we do not accept
1154 * absolute paths, return NULL in that case.
1155 */
1156static const char *skip_tree_prefix(struct apply_state *state,
1157                                    const char *line,
1158                                    int llen)
1159{
1160        int nslash;
1161        int i;
1162
1163        if (!state->p_value)
1164                return (llen && line[0] == '/') ? NULL : line;
1165
1166        nslash = state->p_value;
1167        for (i = 0; i < llen; i++) {
1168                int ch = line[i];
1169                if (ch == '/' && --nslash <= 0)
1170                        return (i == 0) ? NULL : &line[i + 1];
1171        }
1172        return NULL;
1173}
1174
1175/*
1176 * This is to extract the same name that appears on "diff --git"
1177 * line.  We do not find and return anything if it is a rename
1178 * patch, and it is OK because we will find the name elsewhere.
1179 * We need to reliably find name only when it is mode-change only,
1180 * creation or deletion of an empty file.  In any of these cases,
1181 * both sides are the same name under a/ and b/ respectively.
1182 */
1183static char *git_header_name(struct apply_state *state,
1184                             const char *line,
1185                             int llen)
1186{
1187        const char *name;
1188        const char *second = NULL;
1189        size_t len, line_len;
1190
1191        line += strlen("diff --git ");
1192        llen -= strlen("diff --git ");
1193
1194        if (*line == '"') {
1195                const char *cp;
1196                struct strbuf first = STRBUF_INIT;
1197                struct strbuf sp = STRBUF_INIT;
1198
1199                if (unquote_c_style(&first, line, &second))
1200                        goto free_and_fail1;
1201
1202                /* strip the a/b prefix including trailing slash */
1203                cp = skip_tree_prefix(state, first.buf, first.len);
1204                if (!cp)
1205                        goto free_and_fail1;
1206                strbuf_remove(&first, 0, cp - first.buf);
1207
1208                /*
1209                 * second points at one past closing dq of name.
1210                 * find the second name.
1211                 */
1212                while ((second < line + llen) && isspace(*second))
1213                        second++;
1214
1215                if (line + llen <= second)
1216                        goto free_and_fail1;
1217                if (*second == '"') {
1218                        if (unquote_c_style(&sp, second, NULL))
1219                                goto free_and_fail1;
1220                        cp = skip_tree_prefix(state, sp.buf, sp.len);
1221                        if (!cp)
1222                                goto free_and_fail1;
1223                        /* They must match, otherwise ignore */
1224                        if (strcmp(cp, first.buf))
1225                                goto free_and_fail1;
1226                        strbuf_release(&sp);
1227                        return strbuf_detach(&first, NULL);
1228                }
1229
1230                /* unquoted second */
1231                cp = skip_tree_prefix(state, second, line + llen - second);
1232                if (!cp)
1233                        goto free_and_fail1;
1234                if (line + llen - cp != first.len ||
1235                    memcmp(first.buf, cp, first.len))
1236                        goto free_and_fail1;
1237                return strbuf_detach(&first, NULL);
1238
1239        free_and_fail1:
1240                strbuf_release(&first);
1241                strbuf_release(&sp);
1242                return NULL;
1243        }
1244
1245        /* unquoted first name */
1246        name = skip_tree_prefix(state, line, llen);
1247        if (!name)
1248                return NULL;
1249
1250        /*
1251         * since the first name is unquoted, a dq if exists must be
1252         * the beginning of the second name.
1253         */
1254        for (second = name; second < line + llen; second++) {
1255                if (*second == '"') {
1256                        struct strbuf sp = STRBUF_INIT;
1257                        const char *np;
1258
1259                        if (unquote_c_style(&sp, second, NULL))
1260                                goto free_and_fail2;
1261
1262                        np = skip_tree_prefix(state, sp.buf, sp.len);
1263                        if (!np)
1264                                goto free_and_fail2;
1265
1266                        len = sp.buf + sp.len - np;
1267                        if (len < second - name &&
1268                            !strncmp(np, name, len) &&
1269                            isspace(name[len])) {
1270                                /* Good */
1271                                strbuf_remove(&sp, 0, np - sp.buf);
1272                                return strbuf_detach(&sp, NULL);
1273                        }
1274
1275                free_and_fail2:
1276                        strbuf_release(&sp);
1277                        return NULL;
1278                }
1279        }
1280
1281        /*
1282         * Accept a name only if it shows up twice, exactly the same
1283         * form.
1284         */
1285        second = strchr(name, '\n');
1286        if (!second)
1287                return NULL;
1288        line_len = second - name;
1289        for (len = 0 ; ; len++) {
1290                switch (name[len]) {
1291                default:
1292                        continue;
1293                case '\n':
1294                        return NULL;
1295                case '\t': case ' ':
1296                        /*
1297                         * Is this the separator between the preimage
1298                         * and the postimage pathname?  Again, we are
1299                         * only interested in the case where there is
1300                         * no rename, as this is only to set def_name
1301                         * and a rename patch has the names elsewhere
1302                         * in an unambiguous form.
1303                         */
1304                        if (!name[len + 1])
1305                                return NULL; /* no postimage name */
1306                        second = skip_tree_prefix(state, name + len + 1,
1307                                                  line_len - (len + 1));
1308                        if (!second)
1309                                return NULL;
1310                        /*
1311                         * Does len bytes starting at "name" and "second"
1312                         * (that are separated by one HT or SP we just
1313                         * found) exactly match?
1314                         */
1315                        if (second[len] == '\n' && !strncmp(name, second, len))
1316                                return xmemdupz(name, len);
1317                }
1318        }
1319}
1320
1321static int check_header_line(struct apply_state *state, struct patch *patch)
1322{
1323        int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1324                         (patch->is_rename == 1) + (patch->is_copy == 1);
1325        if (extensions > 1)
1326                return error(_("inconsistent header lines %d and %d"),
1327                             patch->extension_linenr, state->linenr);
1328        if (extensions && !patch->extension_linenr)
1329                patch->extension_linenr = state->linenr;
1330        return 0;
1331}
1332
1333/* Verify that we recognize the lines following a git header */
1334static int parse_git_header(struct apply_state *state,
1335                            const char *line,
1336                            int len,
1337                            unsigned int size,
1338                            struct patch *patch)
1339{
1340        unsigned long offset;
1341
1342        /* A git diff has explicit new/delete information, so we don't guess */
1343        patch->is_new = 0;
1344        patch->is_delete = 0;
1345
1346        /*
1347         * Some things may not have the old name in the
1348         * rest of the headers anywhere (pure mode changes,
1349         * or removing or adding empty files), so we get
1350         * the default name from the header.
1351         */
1352        patch->def_name = git_header_name(state, line, len);
1353        if (patch->def_name && state->root.len) {
1354                char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1355                free(patch->def_name);
1356                patch->def_name = s;
1357        }
1358
1359        line += len;
1360        size -= len;
1361        state->linenr++;
1362        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1363                static const struct opentry {
1364                        const char *str;
1365                        int (*fn)(struct apply_state *, const char *, struct patch *);
1366                } optable[] = {
1367                        { "@@ -", gitdiff_hdrend },
1368                        { "--- ", gitdiff_oldname },
1369                        { "+++ ", gitdiff_newname },
1370                        { "old mode ", gitdiff_oldmode },
1371                        { "new mode ", gitdiff_newmode },
1372                        { "deleted file mode ", gitdiff_delete },
1373                        { "new file mode ", gitdiff_newfile },
1374                        { "copy from ", gitdiff_copysrc },
1375                        { "copy to ", gitdiff_copydst },
1376                        { "rename old ", gitdiff_renamesrc },
1377                        { "rename new ", gitdiff_renamedst },
1378                        { "rename from ", gitdiff_renamesrc },
1379                        { "rename to ", gitdiff_renamedst },
1380                        { "similarity index ", gitdiff_similarity },
1381                        { "dissimilarity index ", gitdiff_dissimilarity },
1382                        { "index ", gitdiff_index },
1383                        { "", gitdiff_unrecognized },
1384                };
1385                int i;
1386
1387                len = linelen(line, size);
1388                if (!len || line[len-1] != '\n')
1389                        break;
1390                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1391                        const struct opentry *p = optable + i;
1392                        int oplen = strlen(p->str);
1393                        int res;
1394                        if (len < oplen || memcmp(p->str, line, oplen))
1395                                continue;
1396                        res = p->fn(state, line + oplen, patch);
1397                        if (res < 0)
1398                                return -1;
1399                        if (check_header_line(state, patch))
1400                                return -1;
1401                        if (res > 0)
1402                                return offset;
1403                        break;
1404                }
1405        }
1406
1407        return offset;
1408}
1409
1410static int parse_num(const char *line, unsigned long *p)
1411{
1412        char *ptr;
1413
1414        if (!isdigit(*line))
1415                return 0;
1416        *p = strtoul(line, &ptr, 10);
1417        return ptr - line;
1418}
1419
1420static int parse_range(const char *line, int len, int offset, const char *expect,
1421                       unsigned long *p1, unsigned long *p2)
1422{
1423        int digits, ex;
1424
1425        if (offset < 0 || offset >= len)
1426                return -1;
1427        line += offset;
1428        len -= offset;
1429
1430        digits = parse_num(line, p1);
1431        if (!digits)
1432                return -1;
1433
1434        offset += digits;
1435        line += digits;
1436        len -= digits;
1437
1438        *p2 = 1;
1439        if (*line == ',') {
1440                digits = parse_num(line+1, p2);
1441                if (!digits)
1442                        return -1;
1443
1444                offset += digits+1;
1445                line += digits+1;
1446                len -= digits+1;
1447        }
1448
1449        ex = strlen(expect);
1450        if (ex > len)
1451                return -1;
1452        if (memcmp(line, expect, ex))
1453                return -1;
1454
1455        return offset + ex;
1456}
1457
1458static void recount_diff(const char *line, int size, struct fragment *fragment)
1459{
1460        int oldlines = 0, newlines = 0, ret = 0;
1461
1462        if (size < 1) {
1463                warning("recount: ignore empty hunk");
1464                return;
1465        }
1466
1467        for (;;) {
1468                int len = linelen(line, size);
1469                size -= len;
1470                line += len;
1471
1472                if (size < 1)
1473                        break;
1474
1475                switch (*line) {
1476                case ' ': case '\n':
1477                        newlines++;
1478                        /* fall through */
1479                case '-':
1480                        oldlines++;
1481                        continue;
1482                case '+':
1483                        newlines++;
1484                        continue;
1485                case '\\':
1486                        continue;
1487                case '@':
1488                        ret = size < 3 || !starts_with(line, "@@ ");
1489                        break;
1490                case 'd':
1491                        ret = size < 5 || !starts_with(line, "diff ");
1492                        break;
1493                default:
1494                        ret = -1;
1495                        break;
1496                }
1497                if (ret) {
1498                        warning(_("recount: unexpected line: %.*s"),
1499                                (int)linelen(line, size), line);
1500                        return;
1501                }
1502                break;
1503        }
1504        fragment->oldlines = oldlines;
1505        fragment->newlines = newlines;
1506}
1507
1508/*
1509 * Parse a unified diff fragment header of the
1510 * form "@@ -a,b +c,d @@"
1511 */
1512static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1513{
1514        int offset;
1515
1516        if (!len || line[len-1] != '\n')
1517                return -1;
1518
1519        /* Figure out the number of lines in a fragment */
1520        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1521        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1522
1523        return offset;
1524}
1525
1526/*
1527 * Find file diff header
1528 *
1529 * Returns:
1530 *  -1 if no header was found
1531 *  -128 in case of error
1532 *   the size of the header in bytes (called "offset") otherwise
1533 */
1534static int find_header(struct apply_state *state,
1535                       const char *line,
1536                       unsigned long size,
1537                       int *hdrsize,
1538                       struct patch *patch)
1539{
1540        unsigned long offset, len;
1541
1542        patch->is_toplevel_relative = 0;
1543        patch->is_rename = patch->is_copy = 0;
1544        patch->is_new = patch->is_delete = -1;
1545        patch->old_mode = patch->new_mode = 0;
1546        patch->old_name = patch->new_name = NULL;
1547        for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1548                unsigned long nextlen;
1549
1550                len = linelen(line, size);
1551                if (!len)
1552                        break;
1553
1554                /* Testing this early allows us to take a few shortcuts.. */
1555                if (len < 6)
1556                        continue;
1557
1558                /*
1559                 * Make sure we don't find any unconnected patch fragments.
1560                 * That's a sign that we didn't find a header, and that a
1561                 * patch has become corrupted/broken up.
1562                 */
1563                if (!memcmp("@@ -", line, 4)) {
1564                        struct fragment dummy;
1565                        if (parse_fragment_header(line, len, &dummy) < 0)
1566                                continue;
1567                        error(_("patch fragment without header at line %d: %.*s"),
1568                                     state->linenr, (int)len-1, line);
1569                        return -128;
1570                }
1571
1572                if (size < len + 6)
1573                        break;
1574
1575                /*
1576                 * Git patch? It might not have a real patch, just a rename
1577                 * or mode change, so we handle that specially
1578                 */
1579                if (!memcmp("diff --git ", line, 11)) {
1580                        int git_hdr_len = parse_git_header(state, line, len, size, patch);
1581                        if (git_hdr_len < 0)
1582                                return -128;
1583                        if (git_hdr_len <= len)
1584                                continue;
1585                        if (!patch->old_name && !patch->new_name) {
1586                                if (!patch->def_name) {
1587                                        error(Q_("git diff header lacks filename information when removing "
1588                                                        "%d leading pathname component (line %d)",
1589                                                        "git diff header lacks filename information when removing "
1590                                                        "%d leading pathname components (line %d)",
1591                                                        state->p_value),
1592                                                     state->p_value, state->linenr);
1593                                        return -128;
1594                                }
1595                                patch->old_name = xstrdup(patch->def_name);
1596                                patch->new_name = xstrdup(patch->def_name);
1597                        }
1598                        if ((!patch->new_name && !patch->is_delete) ||
1599                            (!patch->old_name && !patch->is_new)) {
1600                                error(_("git diff header lacks filename information "
1601                                             "(line %d)"), state->linenr);
1602                                return -128;
1603                        }
1604                        patch->is_toplevel_relative = 1;
1605                        *hdrsize = git_hdr_len;
1606                        return offset;
1607                }
1608
1609                /* --- followed by +++ ? */
1610                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1611                        continue;
1612
1613                /*
1614                 * We only accept unified patches, so we want it to
1615                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1616                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1617                 */
1618                nextlen = linelen(line + len, size - len);
1619                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1620                        continue;
1621
1622                /* Ok, we'll consider it a patch */
1623                if (parse_traditional_patch(state, line, line+len, patch))
1624                        return -128;
1625                *hdrsize = len + nextlen;
1626                state->linenr += 2;
1627                return offset;
1628        }
1629        return -1;
1630}
1631
1632static void record_ws_error(struct apply_state *state,
1633                            unsigned result,
1634                            const char *line,
1635                            int len,
1636                            int linenr)
1637{
1638        char *err;
1639
1640        if (!result)
1641                return;
1642
1643        state->whitespace_error++;
1644        if (state->squelch_whitespace_errors &&
1645            state->squelch_whitespace_errors < state->whitespace_error)
1646                return;
1647
1648        err = whitespace_error_string(result);
1649        if (state->apply_verbosity > verbosity_silent)
1650                fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1651                        state->patch_input_file, linenr, err, len, line);
1652        free(err);
1653}
1654
1655static void check_whitespace(struct apply_state *state,
1656                             const char *line,
1657                             int len,
1658                             unsigned ws_rule)
1659{
1660        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1661
1662        record_ws_error(state, result, line + 1, len - 2, state->linenr);
1663}
1664
1665/*
1666 * Parse a unified diff. Note that this really needs to parse each
1667 * fragment separately, since the only way to know the difference
1668 * between a "---" that is part of a patch, and a "---" that starts
1669 * the next patch is to look at the line counts..
1670 */
1671static int parse_fragment(struct apply_state *state,
1672                          const char *line,
1673                          unsigned long size,
1674                          struct patch *patch,
1675                          struct fragment *fragment)
1676{
1677        int added, deleted;
1678        int len = linelen(line, size), offset;
1679        unsigned long oldlines, newlines;
1680        unsigned long leading, trailing;
1681
1682        offset = parse_fragment_header(line, len, fragment);
1683        if (offset < 0)
1684                return -1;
1685        if (offset > 0 && patch->recount)
1686                recount_diff(line + offset, size - offset, fragment);
1687        oldlines = fragment->oldlines;
1688        newlines = fragment->newlines;
1689        leading = 0;
1690        trailing = 0;
1691
1692        /* Parse the thing.. */
1693        line += len;
1694        size -= len;
1695        state->linenr++;
1696        added = deleted = 0;
1697        for (offset = len;
1698             0 < size;
1699             offset += len, size -= len, line += len, state->linenr++) {
1700                if (!oldlines && !newlines)
1701                        break;
1702                len = linelen(line, size);
1703                if (!len || line[len-1] != '\n')
1704                        return -1;
1705                switch (*line) {
1706                default:
1707                        return -1;
1708                case '\n': /* newer GNU diff, an empty context line */
1709                case ' ':
1710                        oldlines--;
1711                        newlines--;
1712                        if (!deleted && !added)
1713                                leading++;
1714                        trailing++;
1715                        if (!state->apply_in_reverse &&
1716                            state->ws_error_action == correct_ws_error)
1717                                check_whitespace(state, line, len, patch->ws_rule);
1718                        break;
1719                case '-':
1720                        if (state->apply_in_reverse &&
1721                            state->ws_error_action != nowarn_ws_error)
1722                                check_whitespace(state, line, len, patch->ws_rule);
1723                        deleted++;
1724                        oldlines--;
1725                        trailing = 0;
1726                        break;
1727                case '+':
1728                        if (!state->apply_in_reverse &&
1729                            state->ws_error_action != nowarn_ws_error)
1730                                check_whitespace(state, line, len, patch->ws_rule);
1731                        added++;
1732                        newlines--;
1733                        trailing = 0;
1734                        break;
1735
1736                /*
1737                 * We allow "\ No newline at end of file". Depending
1738                 * on locale settings when the patch was produced we
1739                 * don't know what this line looks like. The only
1740                 * thing we do know is that it begins with "\ ".
1741                 * Checking for 12 is just for sanity check -- any
1742                 * l10n of "\ No newline..." is at least that long.
1743                 */
1744                case '\\':
1745                        if (len < 12 || memcmp(line, "\\ ", 2))
1746                                return -1;
1747                        break;
1748                }
1749        }
1750        if (oldlines || newlines)
1751                return -1;
1752        if (!deleted && !added)
1753                return -1;
1754
1755        fragment->leading = leading;
1756        fragment->trailing = trailing;
1757
1758        /*
1759         * If a fragment ends with an incomplete line, we failed to include
1760         * it in the above loop because we hit oldlines == newlines == 0
1761         * before seeing it.
1762         */
1763        if (12 < size && !memcmp(line, "\\ ", 2))
1764                offset += linelen(line, size);
1765
1766        patch->lines_added += added;
1767        patch->lines_deleted += deleted;
1768
1769        if (0 < patch->is_new && oldlines)
1770                return error(_("new file depends on old contents"));
1771        if (0 < patch->is_delete && newlines)
1772                return error(_("deleted file still has contents"));
1773        return offset;
1774}
1775
1776/*
1777 * We have seen "diff --git a/... b/..." header (or a traditional patch
1778 * header).  Read hunks that belong to this patch into fragments and hang
1779 * them to the given patch structure.
1780 *
1781 * The (fragment->patch, fragment->size) pair points into the memory given
1782 * by the caller, not a copy, when we return.
1783 *
1784 * Returns:
1785 *   -1 in case of error,
1786 *   the number of bytes in the patch otherwise.
1787 */
1788static int parse_single_patch(struct apply_state *state,
1789                              const char *line,
1790                              unsigned long size,
1791                              struct patch *patch)
1792{
1793        unsigned long offset = 0;
1794        unsigned long oldlines = 0, newlines = 0, context = 0;
1795        struct fragment **fragp = &patch->fragments;
1796
1797        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1798                struct fragment *fragment;
1799                int len;
1800
1801                fragment = xcalloc(1, sizeof(*fragment));
1802                fragment->linenr = state->linenr;
1803                len = parse_fragment(state, line, size, patch, fragment);
1804                if (len <= 0) {
1805                        free(fragment);
1806                        return error(_("corrupt patch at line %d"), state->linenr);
1807                }
1808                fragment->patch = line;
1809                fragment->size = len;
1810                oldlines += fragment->oldlines;
1811                newlines += fragment->newlines;
1812                context += fragment->leading + fragment->trailing;
1813
1814                *fragp = fragment;
1815                fragp = &fragment->next;
1816
1817                offset += len;
1818                line += len;
1819                size -= len;
1820        }
1821
1822        /*
1823         * If something was removed (i.e. we have old-lines) it cannot
1824         * be creation, and if something was added it cannot be
1825         * deletion.  However, the reverse is not true; --unified=0
1826         * patches that only add are not necessarily creation even
1827         * though they do not have any old lines, and ones that only
1828         * delete are not necessarily deletion.
1829         *
1830         * Unfortunately, a real creation/deletion patch do _not_ have
1831         * any context line by definition, so we cannot safely tell it
1832         * apart with --unified=0 insanity.  At least if the patch has
1833         * more than one hunk it is not creation or deletion.
1834         */
1835        if (patch->is_new < 0 &&
1836            (oldlines || (patch->fragments && patch->fragments->next)))
1837                patch->is_new = 0;
1838        if (patch->is_delete < 0 &&
1839            (newlines || (patch->fragments && patch->fragments->next)))
1840                patch->is_delete = 0;
1841
1842        if (0 < patch->is_new && oldlines)
1843                return error(_("new file %s depends on old contents"), patch->new_name);
1844        if (0 < patch->is_delete && newlines)
1845                return error(_("deleted file %s still has contents"), patch->old_name);
1846        if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1847                fprintf_ln(stderr,
1848                           _("** warning: "
1849                             "file %s becomes empty but is not deleted"),
1850                           patch->new_name);
1851
1852        return offset;
1853}
1854
1855static inline int metadata_changes(struct patch *patch)
1856{
1857        return  patch->is_rename > 0 ||
1858                patch->is_copy > 0 ||
1859                patch->is_new > 0 ||
1860                patch->is_delete ||
1861                (patch->old_mode && patch->new_mode &&
1862                 patch->old_mode != patch->new_mode);
1863}
1864
1865static char *inflate_it(const void *data, unsigned long size,
1866                        unsigned long inflated_size)
1867{
1868        git_zstream stream;
1869        void *out;
1870        int st;
1871
1872        memset(&stream, 0, sizeof(stream));
1873
1874        stream.next_in = (unsigned char *)data;
1875        stream.avail_in = size;
1876        stream.next_out = out = xmalloc(inflated_size);
1877        stream.avail_out = inflated_size;
1878        git_inflate_init(&stream);
1879        st = git_inflate(&stream, Z_FINISH);
1880        git_inflate_end(&stream);
1881        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1882                free(out);
1883                return NULL;
1884        }
1885        return out;
1886}
1887
1888/*
1889 * Read a binary hunk and return a new fragment; fragment->patch
1890 * points at an allocated memory that the caller must free, so
1891 * it is marked as "->free_patch = 1".
1892 */
1893static struct fragment *parse_binary_hunk(struct apply_state *state,
1894                                          char **buf_p,
1895                                          unsigned long *sz_p,
1896                                          int *status_p,
1897                                          int *used_p)
1898{
1899        /*
1900         * Expect a line that begins with binary patch method ("literal"
1901         * or "delta"), followed by the length of data before deflating.
1902         * a sequence of 'length-byte' followed by base-85 encoded data
1903         * should follow, terminated by a newline.
1904         *
1905         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1906         * and we would limit the patch line to 66 characters,
1907         * so one line can fit up to 13 groups that would decode
1908         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1909         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1910         */
1911        int llen, used;
1912        unsigned long size = *sz_p;
1913        char *buffer = *buf_p;
1914        int patch_method;
1915        unsigned long origlen;
1916        char *data = NULL;
1917        int hunk_size = 0;
1918        struct fragment *frag;
1919
1920        llen = linelen(buffer, size);
1921        used = llen;
1922
1923        *status_p = 0;
1924
1925        if (starts_with(buffer, "delta ")) {
1926                patch_method = BINARY_DELTA_DEFLATED;
1927                origlen = strtoul(buffer + 6, NULL, 10);
1928        }
1929        else if (starts_with(buffer, "literal ")) {
1930                patch_method = BINARY_LITERAL_DEFLATED;
1931                origlen = strtoul(buffer + 8, NULL, 10);
1932        }
1933        else
1934                return NULL;
1935
1936        state->linenr++;
1937        buffer += llen;
1938        while (1) {
1939                int byte_length, max_byte_length, newsize;
1940                llen = linelen(buffer, size);
1941                used += llen;
1942                state->linenr++;
1943                if (llen == 1) {
1944                        /* consume the blank line */
1945                        buffer++;
1946                        size--;
1947                        break;
1948                }
1949                /*
1950                 * Minimum line is "A00000\n" which is 7-byte long,
1951                 * and the line length must be multiple of 5 plus 2.
1952                 */
1953                if ((llen < 7) || (llen-2) % 5)
1954                        goto corrupt;
1955                max_byte_length = (llen - 2) / 5 * 4;
1956                byte_length = *buffer;
1957                if ('A' <= byte_length && byte_length <= 'Z')
1958                        byte_length = byte_length - 'A' + 1;
1959                else if ('a' <= byte_length && byte_length <= 'z')
1960                        byte_length = byte_length - 'a' + 27;
1961                else
1962                        goto corrupt;
1963                /* if the input length was not multiple of 4, we would
1964                 * have filler at the end but the filler should never
1965                 * exceed 3 bytes
1966                 */
1967                if (max_byte_length < byte_length ||
1968                    byte_length <= max_byte_length - 4)
1969                        goto corrupt;
1970                newsize = hunk_size + byte_length;
1971                data = xrealloc(data, newsize);
1972                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1973                        goto corrupt;
1974                hunk_size = newsize;
1975                buffer += llen;
1976                size -= llen;
1977        }
1978
1979        frag = xcalloc(1, sizeof(*frag));
1980        frag->patch = inflate_it(data, hunk_size, origlen);
1981        frag->free_patch = 1;
1982        if (!frag->patch)
1983                goto corrupt;
1984        free(data);
1985        frag->size = origlen;
1986        *buf_p = buffer;
1987        *sz_p = size;
1988        *used_p = used;
1989        frag->binary_patch_method = patch_method;
1990        return frag;
1991
1992 corrupt:
1993        free(data);
1994        *status_p = -1;
1995        error(_("corrupt binary patch at line %d: %.*s"),
1996              state->linenr-1, llen-1, buffer);
1997        return NULL;
1998}
1999
2000/*
2001 * Returns:
2002 *   -1 in case of error,
2003 *   the length of the parsed binary patch otherwise
2004 */
2005static int parse_binary(struct apply_state *state,
2006                        char *buffer,
2007                        unsigned long size,
2008                        struct patch *patch)
2009{
2010        /*
2011         * We have read "GIT binary patch\n"; what follows is a line
2012         * that says the patch method (currently, either "literal" or
2013         * "delta") and the length of data before deflating; a
2014         * sequence of 'length-byte' followed by base-85 encoded data
2015         * follows.
2016         *
2017         * When a binary patch is reversible, there is another binary
2018         * hunk in the same format, starting with patch method (either
2019         * "literal" or "delta") with the length of data, and a sequence
2020         * of length-byte + base-85 encoded data, terminated with another
2021         * empty line.  This data, when applied to the postimage, produces
2022         * the preimage.
2023         */
2024        struct fragment *forward;
2025        struct fragment *reverse;
2026        int status;
2027        int used, used_1;
2028
2029        forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2030        if (!forward && !status)
2031                /* there has to be one hunk (forward hunk) */
2032                return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2033        if (status)
2034                /* otherwise we already gave an error message */
2035                return status;
2036
2037        reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2038        if (reverse)
2039                used += used_1;
2040        else if (status) {
2041                /*
2042                 * Not having reverse hunk is not an error, but having
2043                 * a corrupt reverse hunk is.
2044                 */
2045                free((void*) forward->patch);
2046                free(forward);
2047                return status;
2048        }
2049        forward->next = reverse;
2050        patch->fragments = forward;
2051        patch->is_binary = 1;
2052        return used;
2053}
2054
2055static void prefix_one(struct apply_state *state, char **name)
2056{
2057        char *old_name = *name;
2058        if (!old_name)
2059                return;
2060        *name = prefix_filename(state->prefix, *name);
2061        free(old_name);
2062}
2063
2064static void prefix_patch(struct apply_state *state, struct patch *p)
2065{
2066        if (!state->prefix || p->is_toplevel_relative)
2067                return;
2068        prefix_one(state, &p->new_name);
2069        prefix_one(state, &p->old_name);
2070}
2071
2072/*
2073 * include/exclude
2074 */
2075
2076static void add_name_limit(struct apply_state *state,
2077                           const char *name,
2078                           int exclude)
2079{
2080        struct string_list_item *it;
2081
2082        it = string_list_append(&state->limit_by_name, name);
2083        it->util = exclude ? NULL : (void *) 1;
2084}
2085
2086static int use_patch(struct apply_state *state, struct patch *p)
2087{
2088        const char *pathname = p->new_name ? p->new_name : p->old_name;
2089        int i;
2090
2091        /* Paths outside are not touched regardless of "--include" */
2092        if (0 < state->prefix_length) {
2093                int pathlen = strlen(pathname);
2094                if (pathlen <= state->prefix_length ||
2095                    memcmp(state->prefix, pathname, state->prefix_length))
2096                        return 0;
2097        }
2098
2099        /* See if it matches any of exclude/include rule */
2100        for (i = 0; i < state->limit_by_name.nr; i++) {
2101                struct string_list_item *it = &state->limit_by_name.items[i];
2102                if (!wildmatch(it->string, pathname, 0))
2103                        return (it->util != NULL);
2104        }
2105
2106        /*
2107         * If we had any include, a path that does not match any rule is
2108         * not used.  Otherwise, we saw bunch of exclude rules (or none)
2109         * and such a path is used.
2110         */
2111        return !state->has_include;
2112}
2113
2114/*
2115 * Read the patch text in "buffer" that extends for "size" bytes; stop
2116 * reading after seeing a single patch (i.e. changes to a single file).
2117 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2118 *
2119 * Returns:
2120 *   -1 if no header was found or parse_binary() failed,
2121 *   -128 on another error,
2122 *   the number of bytes consumed otherwise,
2123 *     so that the caller can call us again for the next patch.
2124 */
2125static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2126{
2127        int hdrsize, patchsize;
2128        int offset = find_header(state, buffer, size, &hdrsize, patch);
2129
2130        if (offset < 0)
2131                return offset;
2132
2133        prefix_patch(state, patch);
2134
2135        if (!use_patch(state, patch))
2136                patch->ws_rule = 0;
2137        else
2138                patch->ws_rule = whitespace_rule(patch->new_name
2139                                                 ? patch->new_name
2140                                                 : patch->old_name);
2141
2142        patchsize = parse_single_patch(state,
2143                                       buffer + offset + hdrsize,
2144                                       size - offset - hdrsize,
2145                                       patch);
2146
2147        if (patchsize < 0)
2148                return -128;
2149
2150        if (!patchsize) {
2151                static const char git_binary[] = "GIT binary patch\n";
2152                int hd = hdrsize + offset;
2153                unsigned long llen = linelen(buffer + hd, size - hd);
2154
2155                if (llen == sizeof(git_binary) - 1 &&
2156                    !memcmp(git_binary, buffer + hd, llen)) {
2157                        int used;
2158                        state->linenr++;
2159                        used = parse_binary(state, buffer + hd + llen,
2160                                            size - hd - llen, patch);
2161                        if (used < 0)
2162                                return -1;
2163                        if (used)
2164                                patchsize = used + llen;
2165                        else
2166                                patchsize = 0;
2167                }
2168                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2169                        static const char *binhdr[] = {
2170                                "Binary files ",
2171                                "Files ",
2172                                NULL,
2173                        };
2174                        int i;
2175                        for (i = 0; binhdr[i]; i++) {
2176                                int len = strlen(binhdr[i]);
2177                                if (len < size - hd &&
2178                                    !memcmp(binhdr[i], buffer + hd, len)) {
2179                                        state->linenr++;
2180                                        patch->is_binary = 1;
2181                                        patchsize = llen;
2182                                        break;
2183                                }
2184                        }
2185                }
2186
2187                /* Empty patch cannot be applied if it is a text patch
2188                 * without metadata change.  A binary patch appears
2189                 * empty to us here.
2190                 */
2191                if ((state->apply || state->check) &&
2192                    (!patch->is_binary && !metadata_changes(patch))) {
2193                        error(_("patch with only garbage at line %d"), state->linenr);
2194                        return -128;
2195                }
2196        }
2197
2198        return offset + hdrsize + patchsize;
2199}
2200
2201static void reverse_patches(struct patch *p)
2202{
2203        for (; p; p = p->next) {
2204                struct fragment *frag = p->fragments;
2205
2206                SWAP(p->new_name, p->old_name);
2207                SWAP(p->new_mode, p->old_mode);
2208                SWAP(p->is_new, p->is_delete);
2209                SWAP(p->lines_added, p->lines_deleted);
2210                SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2211
2212                for (; frag; frag = frag->next) {
2213                        SWAP(frag->newpos, frag->oldpos);
2214                        SWAP(frag->newlines, frag->oldlines);
2215                }
2216        }
2217}
2218
2219static const char pluses[] =
2220"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2221static const char minuses[]=
2222"----------------------------------------------------------------------";
2223
2224static void show_stats(struct apply_state *state, struct patch *patch)
2225{
2226        struct strbuf qname = STRBUF_INIT;
2227        char *cp = patch->new_name ? patch->new_name : patch->old_name;
2228        int max, add, del;
2229
2230        quote_c_style(cp, &qname, NULL, 0);
2231
2232        /*
2233         * "scale" the filename
2234         */
2235        max = state->max_len;
2236        if (max > 50)
2237                max = 50;
2238
2239        if (qname.len > max) {
2240                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2241                if (!cp)
2242                        cp = qname.buf + qname.len + 3 - max;
2243                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2244        }
2245
2246        if (patch->is_binary) {
2247                printf(" %-*s |  Bin\n", max, qname.buf);
2248                strbuf_release(&qname);
2249                return;
2250        }
2251
2252        printf(" %-*s |", max, qname.buf);
2253        strbuf_release(&qname);
2254
2255        /*
2256         * scale the add/delete
2257         */
2258        max = max + state->max_change > 70 ? 70 - max : state->max_change;
2259        add = patch->lines_added;
2260        del = patch->lines_deleted;
2261
2262        if (state->max_change > 0) {
2263                int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2264                add = (add * max + state->max_change / 2) / state->max_change;
2265                del = total - add;
2266        }
2267        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2268                add, pluses, del, minuses);
2269}
2270
2271static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2272{
2273        switch (st->st_mode & S_IFMT) {
2274        case S_IFLNK:
2275                if (strbuf_readlink(buf, path, st->st_size) < 0)
2276                        return error(_("unable to read symlink %s"), path);
2277                return 0;
2278        case S_IFREG:
2279                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2280                        return error(_("unable to open or read %s"), path);
2281                convert_to_git(&the_index, path, buf->buf, buf->len, buf, 0);
2282                return 0;
2283        default:
2284                return -1;
2285        }
2286}
2287
2288/*
2289 * Update the preimage, and the common lines in postimage,
2290 * from buffer buf of length len. If postlen is 0 the postimage
2291 * is updated in place, otherwise it's updated on a new buffer
2292 * of length postlen
2293 */
2294
2295static void update_pre_post_images(struct image *preimage,
2296                                   struct image *postimage,
2297                                   char *buf,
2298                                   size_t len, size_t postlen)
2299{
2300        int i, ctx, reduced;
2301        char *new, *old, *fixed;
2302        struct image fixed_preimage;
2303
2304        /*
2305         * Update the preimage with whitespace fixes.  Note that we
2306         * are not losing preimage->buf -- apply_one_fragment() will
2307         * free "oldlines".
2308         */
2309        prepare_image(&fixed_preimage, buf, len, 1);
2310        assert(postlen
2311               ? fixed_preimage.nr == preimage->nr
2312               : fixed_preimage.nr <= preimage->nr);
2313        for (i = 0; i < fixed_preimage.nr; i++)
2314                fixed_preimage.line[i].flag = preimage->line[i].flag;
2315        free(preimage->line_allocated);
2316        *preimage = fixed_preimage;
2317
2318        /*
2319         * Adjust the common context lines in postimage. This can be
2320         * done in-place when we are shrinking it with whitespace
2321         * fixing, but needs a new buffer when ignoring whitespace or
2322         * expanding leading tabs to spaces.
2323         *
2324         * We trust the caller to tell us if the update can be done
2325         * in place (postlen==0) or not.
2326         */
2327        old = postimage->buf;
2328        if (postlen)
2329                new = postimage->buf = xmalloc(postlen);
2330        else
2331                new = old;
2332        fixed = preimage->buf;
2333
2334        for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2335                size_t l_len = postimage->line[i].len;
2336                if (!(postimage->line[i].flag & LINE_COMMON)) {
2337                        /* an added line -- no counterparts in preimage */
2338                        memmove(new, old, l_len);
2339                        old += l_len;
2340                        new += l_len;
2341                        continue;
2342                }
2343
2344                /* a common context -- skip it in the original postimage */
2345                old += l_len;
2346
2347                /* and find the corresponding one in the fixed preimage */
2348                while (ctx < preimage->nr &&
2349                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2350                        fixed += preimage->line[ctx].len;
2351                        ctx++;
2352                }
2353
2354                /*
2355                 * preimage is expected to run out, if the caller
2356                 * fixed addition of trailing blank lines.
2357                 */
2358                if (preimage->nr <= ctx) {
2359                        reduced++;
2360                        continue;
2361                }
2362
2363                /* and copy it in, while fixing the line length */
2364                l_len = preimage->line[ctx].len;
2365                memcpy(new, fixed, l_len);
2366                new += l_len;
2367                fixed += l_len;
2368                postimage->line[i].len = l_len;
2369                ctx++;
2370        }
2371
2372        if (postlen
2373            ? postlen < new - postimage->buf
2374            : postimage->len < new - postimage->buf)
2375                die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2376                    (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2377
2378        /* Fix the length of the whole thing */
2379        postimage->len = new - postimage->buf;
2380        postimage->nr -= reduced;
2381}
2382
2383static int line_by_line_fuzzy_match(struct image *img,
2384                                    struct image *preimage,
2385                                    struct image *postimage,
2386                                    unsigned long try,
2387                                    int try_lno,
2388                                    int preimage_limit)
2389{
2390        int i;
2391        size_t imgoff = 0;
2392        size_t preoff = 0;
2393        size_t postlen = postimage->len;
2394        size_t extra_chars;
2395        char *buf;
2396        char *preimage_eof;
2397        char *preimage_end;
2398        struct strbuf fixed;
2399        char *fixed_buf;
2400        size_t fixed_len;
2401
2402        for (i = 0; i < preimage_limit; i++) {
2403                size_t prelen = preimage->line[i].len;
2404                size_t imglen = img->line[try_lno+i].len;
2405
2406                if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2407                                      preimage->buf + preoff, prelen))
2408                        return 0;
2409                if (preimage->line[i].flag & LINE_COMMON)
2410                        postlen += imglen - prelen;
2411                imgoff += imglen;
2412                preoff += prelen;
2413        }
2414
2415        /*
2416         * Ok, the preimage matches with whitespace fuzz.
2417         *
2418         * imgoff now holds the true length of the target that
2419         * matches the preimage before the end of the file.
2420         *
2421         * Count the number of characters in the preimage that fall
2422         * beyond the end of the file and make sure that all of them
2423         * are whitespace characters. (This can only happen if
2424         * we are removing blank lines at the end of the file.)
2425         */
2426        buf = preimage_eof = preimage->buf + preoff;
2427        for ( ; i < preimage->nr; i++)
2428                preoff += preimage->line[i].len;
2429        preimage_end = preimage->buf + preoff;
2430        for ( ; buf < preimage_end; buf++)
2431                if (!isspace(*buf))
2432                        return 0;
2433
2434        /*
2435         * Update the preimage and the common postimage context
2436         * lines to use the same whitespace as the target.
2437         * If whitespace is missing in the target (i.e.
2438         * if the preimage extends beyond the end of the file),
2439         * use the whitespace from the preimage.
2440         */
2441        extra_chars = preimage_end - preimage_eof;
2442        strbuf_init(&fixed, imgoff + extra_chars);
2443        strbuf_add(&fixed, img->buf + try, imgoff);
2444        strbuf_add(&fixed, preimage_eof, extra_chars);
2445        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2446        update_pre_post_images(preimage, postimage,
2447                               fixed_buf, fixed_len, postlen);
2448        return 1;
2449}
2450
2451static int match_fragment(struct apply_state *state,
2452                          struct image *img,
2453                          struct image *preimage,
2454                          struct image *postimage,
2455                          unsigned long try,
2456                          int try_lno,
2457                          unsigned ws_rule,
2458                          int match_beginning, int match_end)
2459{
2460        int i;
2461        char *fixed_buf, *buf, *orig, *target;
2462        struct strbuf fixed;
2463        size_t fixed_len, postlen;
2464        int preimage_limit;
2465
2466        if (preimage->nr + try_lno <= img->nr) {
2467                /*
2468                 * The hunk falls within the boundaries of img.
2469                 */
2470                preimage_limit = preimage->nr;
2471                if (match_end && (preimage->nr + try_lno != img->nr))
2472                        return 0;
2473        } else if (state->ws_error_action == correct_ws_error &&
2474                   (ws_rule & WS_BLANK_AT_EOF)) {
2475                /*
2476                 * This hunk extends beyond the end of img, and we are
2477                 * removing blank lines at the end of the file.  This
2478                 * many lines from the beginning of the preimage must
2479                 * match with img, and the remainder of the preimage
2480                 * must be blank.
2481                 */
2482                preimage_limit = img->nr - try_lno;
2483        } else {
2484                /*
2485                 * The hunk extends beyond the end of the img and
2486                 * we are not removing blanks at the end, so we
2487                 * should reject the hunk at this position.
2488                 */
2489                return 0;
2490        }
2491
2492        if (match_beginning && try_lno)
2493                return 0;
2494
2495        /* Quick hash check */
2496        for (i = 0; i < preimage_limit; i++)
2497                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2498                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2499                        return 0;
2500
2501        if (preimage_limit == preimage->nr) {
2502                /*
2503                 * Do we have an exact match?  If we were told to match
2504                 * at the end, size must be exactly at try+fragsize,
2505                 * otherwise try+fragsize must be still within the preimage,
2506                 * and either case, the old piece should match the preimage
2507                 * exactly.
2508                 */
2509                if ((match_end
2510                     ? (try + preimage->len == img->len)
2511                     : (try + preimage->len <= img->len)) &&
2512                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2513                        return 1;
2514        } else {
2515                /*
2516                 * The preimage extends beyond the end of img, so
2517                 * there cannot be an exact match.
2518                 *
2519                 * There must be one non-blank context line that match
2520                 * a line before the end of img.
2521                 */
2522                char *buf_end;
2523
2524                buf = preimage->buf;
2525                buf_end = buf;
2526                for (i = 0; i < preimage_limit; i++)
2527                        buf_end += preimage->line[i].len;
2528
2529                for ( ; buf < buf_end; buf++)
2530                        if (!isspace(*buf))
2531                                break;
2532                if (buf == buf_end)
2533                        return 0;
2534        }
2535
2536        /*
2537         * No exact match. If we are ignoring whitespace, run a line-by-line
2538         * fuzzy matching. We collect all the line length information because
2539         * we need it to adjust whitespace if we match.
2540         */
2541        if (state->ws_ignore_action == ignore_ws_change)
2542                return line_by_line_fuzzy_match(img, preimage, postimage,
2543                                                try, try_lno, preimage_limit);
2544
2545        if (state->ws_error_action != correct_ws_error)
2546                return 0;
2547
2548        /*
2549         * The hunk does not apply byte-by-byte, but the hash says
2550         * it might with whitespace fuzz. We weren't asked to
2551         * ignore whitespace, we were asked to correct whitespace
2552         * errors, so let's try matching after whitespace correction.
2553         *
2554         * While checking the preimage against the target, whitespace
2555         * errors in both fixed, we count how large the corresponding
2556         * postimage needs to be.  The postimage prepared by
2557         * apply_one_fragment() has whitespace errors fixed on added
2558         * lines already, but the common lines were propagated as-is,
2559         * which may become longer when their whitespace errors are
2560         * fixed.
2561         */
2562
2563        /* First count added lines in postimage */
2564        postlen = 0;
2565        for (i = 0; i < postimage->nr; i++) {
2566                if (!(postimage->line[i].flag & LINE_COMMON))
2567                        postlen += postimage->line[i].len;
2568        }
2569
2570        /*
2571         * The preimage may extend beyond the end of the file,
2572         * but in this loop we will only handle the part of the
2573         * preimage that falls within the file.
2574         */
2575        strbuf_init(&fixed, preimage->len + 1);
2576        orig = preimage->buf;
2577        target = img->buf + try;
2578        for (i = 0; i < preimage_limit; i++) {
2579                size_t oldlen = preimage->line[i].len;
2580                size_t tgtlen = img->line[try_lno + i].len;
2581                size_t fixstart = fixed.len;
2582                struct strbuf tgtfix;
2583                int match;
2584
2585                /* Try fixing the line in the preimage */
2586                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2587
2588                /* Try fixing the line in the target */
2589                strbuf_init(&tgtfix, tgtlen);
2590                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2591
2592                /*
2593                 * If they match, either the preimage was based on
2594                 * a version before our tree fixed whitespace breakage,
2595                 * or we are lacking a whitespace-fix patch the tree
2596                 * the preimage was based on already had (i.e. target
2597                 * has whitespace breakage, the preimage doesn't).
2598                 * In either case, we are fixing the whitespace breakages
2599                 * so we might as well take the fix together with their
2600                 * real change.
2601                 */
2602                match = (tgtfix.len == fixed.len - fixstart &&
2603                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2604                                             fixed.len - fixstart));
2605
2606                /* Add the length if this is common with the postimage */
2607                if (preimage->line[i].flag & LINE_COMMON)
2608                        postlen += tgtfix.len;
2609
2610                strbuf_release(&tgtfix);
2611                if (!match)
2612                        goto unmatch_exit;
2613
2614                orig += oldlen;
2615                target += tgtlen;
2616        }
2617
2618
2619        /*
2620         * Now handle the lines in the preimage that falls beyond the
2621         * end of the file (if any). They will only match if they are
2622         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2623         * false).
2624         */
2625        for ( ; i < preimage->nr; i++) {
2626                size_t fixstart = fixed.len; /* start of the fixed preimage */
2627                size_t oldlen = preimage->line[i].len;
2628                int j;
2629
2630                /* Try fixing the line in the preimage */
2631                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2632
2633                for (j = fixstart; j < fixed.len; j++)
2634                        if (!isspace(fixed.buf[j]))
2635                                goto unmatch_exit;
2636
2637                orig += oldlen;
2638        }
2639
2640        /*
2641         * Yes, the preimage is based on an older version that still
2642         * has whitespace breakages unfixed, and fixing them makes the
2643         * hunk match.  Update the context lines in the postimage.
2644         */
2645        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2646        if (postlen < postimage->len)
2647                postlen = 0;
2648        update_pre_post_images(preimage, postimage,
2649                               fixed_buf, fixed_len, postlen);
2650        return 1;
2651
2652 unmatch_exit:
2653        strbuf_release(&fixed);
2654        return 0;
2655}
2656
2657static int find_pos(struct apply_state *state,
2658                    struct image *img,
2659                    struct image *preimage,
2660                    struct image *postimage,
2661                    int line,
2662                    unsigned ws_rule,
2663                    int match_beginning, int match_end)
2664{
2665        int i;
2666        unsigned long backwards, forwards, try;
2667        int backwards_lno, forwards_lno, try_lno;
2668
2669        /*
2670         * If match_beginning or match_end is specified, there is no
2671         * point starting from a wrong line that will never match and
2672         * wander around and wait for a match at the specified end.
2673         */
2674        if (match_beginning)
2675                line = 0;
2676        else if (match_end)
2677                line = img->nr - preimage->nr;
2678
2679        /*
2680         * Because the comparison is unsigned, the following test
2681         * will also take care of a negative line number that can
2682         * result when match_end and preimage is larger than the target.
2683         */
2684        if ((size_t) line > img->nr)
2685                line = img->nr;
2686
2687        try = 0;
2688        for (i = 0; i < line; i++)
2689                try += img->line[i].len;
2690
2691        /*
2692         * There's probably some smart way to do this, but I'll leave
2693         * that to the smart and beautiful people. I'm simple and stupid.
2694         */
2695        backwards = try;
2696        backwards_lno = line;
2697        forwards = try;
2698        forwards_lno = line;
2699        try_lno = line;
2700
2701        for (i = 0; ; i++) {
2702                if (match_fragment(state, img, preimage, postimage,
2703                                   try, try_lno, ws_rule,
2704                                   match_beginning, match_end))
2705                        return try_lno;
2706
2707        again:
2708                if (backwards_lno == 0 && forwards_lno == img->nr)
2709                        break;
2710
2711                if (i & 1) {
2712                        if (backwards_lno == 0) {
2713                                i++;
2714                                goto again;
2715                        }
2716                        backwards_lno--;
2717                        backwards -= img->line[backwards_lno].len;
2718                        try = backwards;
2719                        try_lno = backwards_lno;
2720                } else {
2721                        if (forwards_lno == img->nr) {
2722                                i++;
2723                                goto again;
2724                        }
2725                        forwards += img->line[forwards_lno].len;
2726                        forwards_lno++;
2727                        try = forwards;
2728                        try_lno = forwards_lno;
2729                }
2730
2731        }
2732        return -1;
2733}
2734
2735static void remove_first_line(struct image *img)
2736{
2737        img->buf += img->line[0].len;
2738        img->len -= img->line[0].len;
2739        img->line++;
2740        img->nr--;
2741}
2742
2743static void remove_last_line(struct image *img)
2744{
2745        img->len -= img->line[--img->nr].len;
2746}
2747
2748/*
2749 * The change from "preimage" and "postimage" has been found to
2750 * apply at applied_pos (counts in line numbers) in "img".
2751 * Update "img" to remove "preimage" and replace it with "postimage".
2752 */
2753static void update_image(struct apply_state *state,
2754                         struct image *img,
2755                         int applied_pos,
2756                         struct image *preimage,
2757                         struct image *postimage)
2758{
2759        /*
2760         * remove the copy of preimage at offset in img
2761         * and replace it with postimage
2762         */
2763        int i, nr;
2764        size_t remove_count, insert_count, applied_at = 0;
2765        char *result;
2766        int preimage_limit;
2767
2768        /*
2769         * If we are removing blank lines at the end of img,
2770         * the preimage may extend beyond the end.
2771         * If that is the case, we must be careful only to
2772         * remove the part of the preimage that falls within
2773         * the boundaries of img. Initialize preimage_limit
2774         * to the number of lines in the preimage that falls
2775         * within the boundaries.
2776         */
2777        preimage_limit = preimage->nr;
2778        if (preimage_limit > img->nr - applied_pos)
2779                preimage_limit = img->nr - applied_pos;
2780
2781        for (i = 0; i < applied_pos; i++)
2782                applied_at += img->line[i].len;
2783
2784        remove_count = 0;
2785        for (i = 0; i < preimage_limit; i++)
2786                remove_count += img->line[applied_pos + i].len;
2787        insert_count = postimage->len;
2788
2789        /* Adjust the contents */
2790        result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2791        memcpy(result, img->buf, applied_at);
2792        memcpy(result + applied_at, postimage->buf, postimage->len);
2793        memcpy(result + applied_at + postimage->len,
2794               img->buf + (applied_at + remove_count),
2795               img->len - (applied_at + remove_count));
2796        free(img->buf);
2797        img->buf = result;
2798        img->len += insert_count - remove_count;
2799        result[img->len] = '\0';
2800
2801        /* Adjust the line table */
2802        nr = img->nr + postimage->nr - preimage_limit;
2803        if (preimage_limit < postimage->nr) {
2804                /*
2805                 * NOTE: this knows that we never call remove_first_line()
2806                 * on anything other than pre/post image.
2807                 */
2808                REALLOC_ARRAY(img->line, nr);
2809                img->line_allocated = img->line;
2810        }
2811        if (preimage_limit != postimage->nr)
2812                MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2813                           img->line + applied_pos + preimage_limit,
2814                           img->nr - (applied_pos + preimage_limit));
2815        COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2816        if (!state->allow_overlap)
2817                for (i = 0; i < postimage->nr; i++)
2818                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2819        img->nr = nr;
2820}
2821
2822/*
2823 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2824 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2825 * replace the part of "img" with "postimage" text.
2826 */
2827static int apply_one_fragment(struct apply_state *state,
2828                              struct image *img, struct fragment *frag,
2829                              int inaccurate_eof, unsigned ws_rule,
2830                              int nth_fragment)
2831{
2832        int match_beginning, match_end;
2833        const char *patch = frag->patch;
2834        int size = frag->size;
2835        char *old, *oldlines;
2836        struct strbuf newlines;
2837        int new_blank_lines_at_end = 0;
2838        int found_new_blank_lines_at_end = 0;
2839        int hunk_linenr = frag->linenr;
2840        unsigned long leading, trailing;
2841        int pos, applied_pos;
2842        struct image preimage;
2843        struct image postimage;
2844
2845        memset(&preimage, 0, sizeof(preimage));
2846        memset(&postimage, 0, sizeof(postimage));
2847        oldlines = xmalloc(size);
2848        strbuf_init(&newlines, size);
2849
2850        old = oldlines;
2851        while (size > 0) {
2852                char first;
2853                int len = linelen(patch, size);
2854                int plen;
2855                int added_blank_line = 0;
2856                int is_blank_context = 0;
2857                size_t start;
2858
2859                if (!len)
2860                        break;
2861
2862                /*
2863                 * "plen" is how much of the line we should use for
2864                 * the actual patch data. Normally we just remove the
2865                 * first character on the line, but if the line is
2866                 * followed by "\ No newline", then we also remove the
2867                 * last one (which is the newline, of course).
2868                 */
2869                plen = len - 1;
2870                if (len < size && patch[len] == '\\')
2871                        plen--;
2872                first = *patch;
2873                if (state->apply_in_reverse) {
2874                        if (first == '-')
2875                                first = '+';
2876                        else if (first == '+')
2877                                first = '-';
2878                }
2879
2880                switch (first) {
2881                case '\n':
2882                        /* Newer GNU diff, empty context line */
2883                        if (plen < 0)
2884                                /* ... followed by '\No newline'; nothing */
2885                                break;
2886                        *old++ = '\n';
2887                        strbuf_addch(&newlines, '\n');
2888                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2889                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2890                        is_blank_context = 1;
2891                        break;
2892                case ' ':
2893                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2894                            ws_blank_line(patch + 1, plen, ws_rule))
2895                                is_blank_context = 1;
2896                case '-':
2897                        memcpy(old, patch + 1, plen);
2898                        add_line_info(&preimage, old, plen,
2899                                      (first == ' ' ? LINE_COMMON : 0));
2900                        old += plen;
2901                        if (first == '-')
2902                                break;
2903                /* Fall-through for ' ' */
2904                case '+':
2905                        /* --no-add does not add new lines */
2906                        if (first == '+' && state->no_add)
2907                                break;
2908
2909                        start = newlines.len;
2910                        if (first != '+' ||
2911                            !state->whitespace_error ||
2912                            state->ws_error_action != correct_ws_error) {
2913                                strbuf_add(&newlines, patch + 1, plen);
2914                        }
2915                        else {
2916                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2917                        }
2918                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2919                                      (first == '+' ? 0 : LINE_COMMON));
2920                        if (first == '+' &&
2921                            (ws_rule & WS_BLANK_AT_EOF) &&
2922                            ws_blank_line(patch + 1, plen, ws_rule))
2923                                added_blank_line = 1;
2924                        break;
2925                case '@': case '\\':
2926                        /* Ignore it, we already handled it */
2927                        break;
2928                default:
2929                        if (state->apply_verbosity > verbosity_normal)
2930                                error(_("invalid start of line: '%c'"), first);
2931                        applied_pos = -1;
2932                        goto out;
2933                }
2934                if (added_blank_line) {
2935                        if (!new_blank_lines_at_end)
2936                                found_new_blank_lines_at_end = hunk_linenr;
2937                        new_blank_lines_at_end++;
2938                }
2939                else if (is_blank_context)
2940                        ;
2941                else
2942                        new_blank_lines_at_end = 0;
2943                patch += len;
2944                size -= len;
2945                hunk_linenr++;
2946        }
2947        if (inaccurate_eof &&
2948            old > oldlines && old[-1] == '\n' &&
2949            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2950                old--;
2951                strbuf_setlen(&newlines, newlines.len - 1);
2952        }
2953
2954        leading = frag->leading;
2955        trailing = frag->trailing;
2956
2957        /*
2958         * A hunk to change lines at the beginning would begin with
2959         * @@ -1,L +N,M @@
2960         * but we need to be careful.  -U0 that inserts before the second
2961         * line also has this pattern.
2962         *
2963         * And a hunk to add to an empty file would begin with
2964         * @@ -0,0 +N,M @@
2965         *
2966         * In other words, a hunk that is (frag->oldpos <= 1) with or
2967         * without leading context must match at the beginning.
2968         */
2969        match_beginning = (!frag->oldpos ||
2970                           (frag->oldpos == 1 && !state->unidiff_zero));
2971
2972        /*
2973         * A hunk without trailing lines must match at the end.
2974         * However, we simply cannot tell if a hunk must match end
2975         * from the lack of trailing lines if the patch was generated
2976         * with unidiff without any context.
2977         */
2978        match_end = !state->unidiff_zero && !trailing;
2979
2980        pos = frag->newpos ? (frag->newpos - 1) : 0;
2981        preimage.buf = oldlines;
2982        preimage.len = old - oldlines;
2983        postimage.buf = newlines.buf;
2984        postimage.len = newlines.len;
2985        preimage.line = preimage.line_allocated;
2986        postimage.line = postimage.line_allocated;
2987
2988        for (;;) {
2989
2990                applied_pos = find_pos(state, img, &preimage, &postimage, pos,
2991                                       ws_rule, match_beginning, match_end);
2992
2993                if (applied_pos >= 0)
2994                        break;
2995
2996                /* Am I at my context limits? */
2997                if ((leading <= state->p_context) && (trailing <= state->p_context))
2998                        break;
2999                if (match_beginning || match_end) {
3000                        match_beginning = match_end = 0;
3001                        continue;
3002                }
3003
3004                /*
3005                 * Reduce the number of context lines; reduce both
3006                 * leading and trailing if they are equal otherwise
3007                 * just reduce the larger context.
3008                 */
3009                if (leading >= trailing) {
3010                        remove_first_line(&preimage);
3011                        remove_first_line(&postimage);
3012                        pos--;
3013                        leading--;
3014                }
3015                if (trailing > leading) {
3016                        remove_last_line(&preimage);
3017                        remove_last_line(&postimage);
3018                        trailing--;
3019                }
3020        }
3021
3022        if (applied_pos >= 0) {
3023                if (new_blank_lines_at_end &&
3024                    preimage.nr + applied_pos >= img->nr &&
3025                    (ws_rule & WS_BLANK_AT_EOF) &&
3026                    state->ws_error_action != nowarn_ws_error) {
3027                        record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3028                                        found_new_blank_lines_at_end);
3029                        if (state->ws_error_action == correct_ws_error) {
3030                                while (new_blank_lines_at_end--)
3031                                        remove_last_line(&postimage);
3032                        }
3033                        /*
3034                         * We would want to prevent write_out_results()
3035                         * from taking place in apply_patch() that follows
3036                         * the callchain led us here, which is:
3037                         * apply_patch->check_patch_list->check_patch->
3038                         * apply_data->apply_fragments->apply_one_fragment
3039                         */
3040                        if (state->ws_error_action == die_on_ws_error)
3041                                state->apply = 0;
3042                }
3043
3044                if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3045                        int offset = applied_pos - pos;
3046                        if (state->apply_in_reverse)
3047                                offset = 0 - offset;
3048                        fprintf_ln(stderr,
3049                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
3050                                      "Hunk #%d succeeded at %d (offset %d lines).",
3051                                      offset),
3052                                   nth_fragment, applied_pos + 1, offset);
3053                }
3054
3055                /*
3056                 * Warn if it was necessary to reduce the number
3057                 * of context lines.
3058                 */
3059                if ((leading != frag->leading ||
3060                     trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3061                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3062                                             " to apply fragment at %d"),
3063                                   leading, trailing, applied_pos+1);
3064                update_image(state, img, applied_pos, &preimage, &postimage);
3065        } else {
3066                if (state->apply_verbosity > verbosity_normal)
3067                        error(_("while searching for:\n%.*s"),
3068                              (int)(old - oldlines), oldlines);
3069        }
3070
3071out:
3072        free(oldlines);
3073        strbuf_release(&newlines);
3074        free(preimage.line_allocated);
3075        free(postimage.line_allocated);
3076
3077        return (applied_pos < 0);
3078}
3079
3080static int apply_binary_fragment(struct apply_state *state,
3081                                 struct image *img,
3082                                 struct patch *patch)
3083{
3084        struct fragment *fragment = patch->fragments;
3085        unsigned long len;
3086        void *dst;
3087
3088        if (!fragment)
3089                return error(_("missing binary patch data for '%s'"),
3090                             patch->new_name ?
3091                             patch->new_name :
3092                             patch->old_name);
3093
3094        /* Binary patch is irreversible without the optional second hunk */
3095        if (state->apply_in_reverse) {
3096                if (!fragment->next)
3097                        return error(_("cannot reverse-apply a binary patch "
3098                                       "without the reverse hunk to '%s'"),
3099                                     patch->new_name
3100                                     ? patch->new_name : patch->old_name);
3101                fragment = fragment->next;
3102        }
3103        switch (fragment->binary_patch_method) {
3104        case BINARY_DELTA_DEFLATED:
3105                dst = patch_delta(img->buf, img->len, fragment->patch,
3106                                  fragment->size, &len);
3107                if (!dst)
3108                        return -1;
3109                clear_image(img);
3110                img->buf = dst;
3111                img->len = len;
3112                return 0;
3113        case BINARY_LITERAL_DEFLATED:
3114                clear_image(img);
3115                img->len = fragment->size;
3116                img->buf = xmemdupz(fragment->patch, img->len);
3117                return 0;
3118        }
3119        return -1;
3120}
3121
3122/*
3123 * Replace "img" with the result of applying the binary patch.
3124 * The binary patch data itself in patch->fragment is still kept
3125 * but the preimage prepared by the caller in "img" is freed here
3126 * or in the helper function apply_binary_fragment() this calls.
3127 */
3128static int apply_binary(struct apply_state *state,
3129                        struct image *img,
3130                        struct patch *patch)
3131{
3132        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3133        struct object_id oid;
3134
3135        /*
3136         * For safety, we require patch index line to contain
3137         * full 40-byte textual SHA1 for old and new, at least for now.
3138         */
3139        if (strlen(patch->old_sha1_prefix) != 40 ||
3140            strlen(patch->new_sha1_prefix) != 40 ||
3141            get_oid_hex(patch->old_sha1_prefix, &oid) ||
3142            get_oid_hex(patch->new_sha1_prefix, &oid))
3143                return error(_("cannot apply binary patch to '%s' "
3144                               "without full index line"), name);
3145
3146        if (patch->old_name) {
3147                /*
3148                 * See if the old one matches what the patch
3149                 * applies to.
3150                 */
3151                hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3152                if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3153                        return error(_("the patch applies to '%s' (%s), "
3154                                       "which does not match the "
3155                                       "current contents."),
3156                                     name, oid_to_hex(&oid));
3157        }
3158        else {
3159                /* Otherwise, the old one must be empty. */
3160                if (img->len)
3161                        return error(_("the patch applies to an empty "
3162                                       "'%s' but it is not empty"), name);
3163        }
3164
3165        get_oid_hex(patch->new_sha1_prefix, &oid);
3166        if (is_null_oid(&oid)) {
3167                clear_image(img);
3168                return 0; /* deletion patch */
3169        }
3170
3171        if (has_sha1_file(oid.hash)) {
3172                /* We already have the postimage */
3173                enum object_type type;
3174                unsigned long size;
3175                char *result;
3176
3177                result = read_sha1_file(oid.hash, &type, &size);
3178                if (!result)
3179                        return error(_("the necessary postimage %s for "
3180                                       "'%s' cannot be read"),
3181                                     patch->new_sha1_prefix, name);
3182                clear_image(img);
3183                img->buf = result;
3184                img->len = size;
3185        } else {
3186                /*
3187                 * We have verified buf matches the preimage;
3188                 * apply the patch data to it, which is stored
3189                 * in the patch->fragments->{patch,size}.
3190                 */
3191                if (apply_binary_fragment(state, img, patch))
3192                        return error(_("binary patch does not apply to '%s'"),
3193                                     name);
3194
3195                /* verify that the result matches */
3196                hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3197                if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3198                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3199                                name, patch->new_sha1_prefix, oid_to_hex(&oid));
3200        }
3201
3202        return 0;
3203}
3204
3205static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3206{
3207        struct fragment *frag = patch->fragments;
3208        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3209        unsigned ws_rule = patch->ws_rule;
3210        unsigned inaccurate_eof = patch->inaccurate_eof;
3211        int nth = 0;
3212
3213        if (patch->is_binary)
3214                return apply_binary(state, img, patch);
3215
3216        while (frag) {
3217                nth++;
3218                if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3219                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
3220                        if (!state->apply_with_reject)
3221                                return -1;
3222                        frag->rejected = 1;
3223                }
3224                frag = frag->next;
3225        }
3226        return 0;
3227}
3228
3229static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3230{
3231        if (S_ISGITLINK(mode)) {
3232                strbuf_grow(buf, 100);
3233                strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3234        } else {
3235                enum object_type type;
3236                unsigned long sz;
3237                char *result;
3238
3239                result = read_sha1_file(oid->hash, &type, &sz);
3240                if (!result)
3241                        return -1;
3242                /* XXX read_sha1_file NUL-terminates */
3243                strbuf_attach(buf, result, sz, sz + 1);
3244        }
3245        return 0;
3246}
3247
3248static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3249{
3250        if (!ce)
3251                return 0;
3252        return read_blob_object(buf, &ce->oid, ce->ce_mode);
3253}
3254
3255static struct patch *in_fn_table(struct apply_state *state, const char *name)
3256{
3257        struct string_list_item *item;
3258
3259        if (name == NULL)
3260                return NULL;
3261
3262        item = string_list_lookup(&state->fn_table, name);
3263        if (item != NULL)
3264                return (struct patch *)item->util;
3265
3266        return NULL;
3267}
3268
3269/*
3270 * item->util in the filename table records the status of the path.
3271 * Usually it points at a patch (whose result records the contents
3272 * of it after applying it), but it could be PATH_WAS_DELETED for a
3273 * path that a previously applied patch has already removed, or
3274 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3275 *
3276 * The latter is needed to deal with a case where two paths A and B
3277 * are swapped by first renaming A to B and then renaming B to A;
3278 * moving A to B should not be prevented due to presence of B as we
3279 * will remove it in a later patch.
3280 */
3281#define PATH_TO_BE_DELETED ((struct patch *) -2)
3282#define PATH_WAS_DELETED ((struct patch *) -1)
3283
3284static int to_be_deleted(struct patch *patch)
3285{
3286        return patch == PATH_TO_BE_DELETED;
3287}
3288
3289static int was_deleted(struct patch *patch)
3290{
3291        return patch == PATH_WAS_DELETED;
3292}
3293
3294static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3295{
3296        struct string_list_item *item;
3297
3298        /*
3299         * Always add new_name unless patch is a deletion
3300         * This should cover the cases for normal diffs,
3301         * file creations and copies
3302         */
3303        if (patch->new_name != NULL) {
3304                item = string_list_insert(&state->fn_table, patch->new_name);
3305                item->util = patch;
3306        }
3307
3308        /*
3309         * store a failure on rename/deletion cases because
3310         * later chunks shouldn't patch old names
3311         */
3312        if ((patch->new_name == NULL) || (patch->is_rename)) {
3313                item = string_list_insert(&state->fn_table, patch->old_name);
3314                item->util = PATH_WAS_DELETED;
3315        }
3316}
3317
3318static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3319{
3320        /*
3321         * store information about incoming file deletion
3322         */
3323        while (patch) {
3324                if ((patch->new_name == NULL) || (patch->is_rename)) {
3325                        struct string_list_item *item;
3326                        item = string_list_insert(&state->fn_table, patch->old_name);
3327                        item->util = PATH_TO_BE_DELETED;
3328                }
3329                patch = patch->next;
3330        }
3331}
3332
3333static int checkout_target(struct index_state *istate,
3334                           struct cache_entry *ce, struct stat *st)
3335{
3336        struct checkout costate = CHECKOUT_INIT;
3337
3338        costate.refresh_cache = 1;
3339        costate.istate = istate;
3340        if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3341                return error(_("cannot checkout %s"), ce->name);
3342        return 0;
3343}
3344
3345static struct patch *previous_patch(struct apply_state *state,
3346                                    struct patch *patch,
3347                                    int *gone)
3348{
3349        struct patch *previous;
3350
3351        *gone = 0;
3352        if (patch->is_copy || patch->is_rename)
3353                return NULL; /* "git" patches do not depend on the order */
3354
3355        previous = in_fn_table(state, patch->old_name);
3356        if (!previous)
3357                return NULL;
3358
3359        if (to_be_deleted(previous))
3360                return NULL; /* the deletion hasn't happened yet */
3361
3362        if (was_deleted(previous))
3363                *gone = 1;
3364
3365        return previous;
3366}
3367
3368static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3369{
3370        if (S_ISGITLINK(ce->ce_mode)) {
3371                if (!S_ISDIR(st->st_mode))
3372                        return -1;
3373                return 0;
3374        }
3375        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3376}
3377
3378#define SUBMODULE_PATCH_WITHOUT_INDEX 1
3379
3380static int load_patch_target(struct apply_state *state,
3381                             struct strbuf *buf,
3382                             const struct cache_entry *ce,
3383                             struct stat *st,
3384                             const char *name,
3385                             unsigned expected_mode)
3386{
3387        if (state->cached || state->check_index) {
3388                if (read_file_or_gitlink(ce, buf))
3389                        return error(_("failed to read %s"), name);
3390        } else if (name) {
3391                if (S_ISGITLINK(expected_mode)) {
3392                        if (ce)
3393                                return read_file_or_gitlink(ce, buf);
3394                        else
3395                                return SUBMODULE_PATCH_WITHOUT_INDEX;
3396                } else if (has_symlink_leading_path(name, strlen(name))) {
3397                        return error(_("reading from '%s' beyond a symbolic link"), name);
3398                } else {
3399                        if (read_old_data(st, name, buf))
3400                                return error(_("failed to read %s"), name);
3401                }
3402        }
3403        return 0;
3404}
3405
3406/*
3407 * We are about to apply "patch"; populate the "image" with the
3408 * current version we have, from the working tree or from the index,
3409 * depending on the situation e.g. --cached/--index.  If we are
3410 * applying a non-git patch that incrementally updates the tree,
3411 * we read from the result of a previous diff.
3412 */
3413static int load_preimage(struct apply_state *state,
3414                         struct image *image,
3415                         struct patch *patch, struct stat *st,
3416                         const struct cache_entry *ce)
3417{
3418        struct strbuf buf = STRBUF_INIT;
3419        size_t len;
3420        char *img;
3421        struct patch *previous;
3422        int status;
3423
3424        previous = previous_patch(state, patch, &status);
3425        if (status)
3426                return error(_("path %s has been renamed/deleted"),
3427                             patch->old_name);
3428        if (previous) {
3429                /* We have a patched copy in memory; use that. */
3430                strbuf_add(&buf, previous->result, previous->resultsize);
3431        } else {
3432                status = load_patch_target(state, &buf, ce, st,
3433                                           patch->old_name, patch->old_mode);
3434                if (status < 0)
3435                        return status;
3436                else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3437                        /*
3438                         * There is no way to apply subproject
3439                         * patch without looking at the index.
3440                         * NEEDSWORK: shouldn't this be flagged
3441                         * as an error???
3442                         */
3443                        free_fragment_list(patch->fragments);
3444                        patch->fragments = NULL;
3445                } else if (status) {
3446                        return error(_("failed to read %s"), patch->old_name);
3447                }
3448        }
3449
3450        img = strbuf_detach(&buf, &len);
3451        prepare_image(image, img, len, !patch->is_binary);
3452        return 0;
3453}
3454
3455static int three_way_merge(struct image *image,
3456                           char *path,
3457                           const struct object_id *base,
3458                           const struct object_id *ours,
3459                           const struct object_id *theirs)
3460{
3461        mmfile_t base_file, our_file, their_file;
3462        mmbuffer_t result = { NULL };
3463        int status;
3464
3465        read_mmblob(&base_file, base);
3466        read_mmblob(&our_file, ours);
3467        read_mmblob(&their_file, theirs);
3468        status = ll_merge(&result, path,
3469                          &base_file, "base",
3470                          &our_file, "ours",
3471                          &their_file, "theirs", NULL);
3472        free(base_file.ptr);
3473        free(our_file.ptr);
3474        free(their_file.ptr);
3475        if (status < 0 || !result.ptr) {
3476                free(result.ptr);
3477                return -1;
3478        }
3479        clear_image(image);
3480        image->buf = result.ptr;
3481        image->len = result.size;
3482
3483        return status;
3484}
3485
3486/*
3487 * When directly falling back to add/add three-way merge, we read from
3488 * the current contents of the new_name.  In no cases other than that
3489 * this function will be called.
3490 */
3491static int load_current(struct apply_state *state,
3492                        struct image *image,
3493                        struct patch *patch)
3494{
3495        struct strbuf buf = STRBUF_INIT;
3496        int status, pos;
3497        size_t len;
3498        char *img;
3499        struct stat st;
3500        struct cache_entry *ce;
3501        char *name = patch->new_name;
3502        unsigned mode = patch->new_mode;
3503
3504        if (!patch->is_new)
3505                die("BUG: patch to %s is not a creation", patch->old_name);
3506
3507        pos = cache_name_pos(name, strlen(name));
3508        if (pos < 0)
3509                return error(_("%s: does not exist in index"), name);
3510        ce = active_cache[pos];
3511        if (lstat(name, &st)) {
3512                if (errno != ENOENT)
3513                        return error_errno("%s", name);
3514                if (checkout_target(&the_index, ce, &st))
3515                        return -1;
3516        }
3517        if (verify_index_match(ce, &st))
3518                return error(_("%s: does not match index"), name);
3519
3520        status = load_patch_target(state, &buf, ce, &st, name, mode);
3521        if (status < 0)
3522                return status;
3523        else if (status)
3524                return -1;
3525        img = strbuf_detach(&buf, &len);
3526        prepare_image(image, img, len, !patch->is_binary);
3527        return 0;
3528}
3529
3530static int try_threeway(struct apply_state *state,
3531                        struct image *image,
3532                        struct patch *patch,
3533                        struct stat *st,
3534                        const struct cache_entry *ce)
3535{
3536        struct object_id pre_oid, post_oid, our_oid;
3537        struct strbuf buf = STRBUF_INIT;
3538        size_t len;
3539        int status;
3540        char *img;
3541        struct image tmp_image;
3542
3543        /* No point falling back to 3-way merge in these cases */
3544        if (patch->is_delete ||
3545            S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3546                return -1;
3547
3548        /* Preimage the patch was prepared for */
3549        if (patch->is_new)
3550                write_sha1_file("", 0, blob_type, pre_oid.hash);
3551        else if (get_oid(patch->old_sha1_prefix, &pre_oid) ||
3552                 read_blob_object(&buf, &pre_oid, patch->old_mode))
3553                return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3554
3555        if (state->apply_verbosity > verbosity_silent)
3556                fprintf(stderr, _("Falling back to three-way merge...\n"));
3557
3558        img = strbuf_detach(&buf, &len);
3559        prepare_image(&tmp_image, img, len, 1);
3560        /* Apply the patch to get the post image */
3561        if (apply_fragments(state, &tmp_image, patch) < 0) {
3562                clear_image(&tmp_image);
3563                return -1;
3564        }
3565        /* post_oid is theirs */
3566        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3567        clear_image(&tmp_image);
3568
3569        /* our_oid is ours */
3570        if (patch->is_new) {
3571                if (load_current(state, &tmp_image, patch))
3572                        return error(_("cannot read the current contents of '%s'"),
3573                                     patch->new_name);
3574        } else {
3575                if (load_preimage(state, &tmp_image, patch, st, ce))
3576                        return error(_("cannot read the current contents of '%s'"),
3577                                     patch->old_name);
3578        }
3579        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3580        clear_image(&tmp_image);
3581
3582        /* in-core three-way merge between post and our using pre as base */
3583        status = three_way_merge(image, patch->new_name,
3584                                 &pre_oid, &our_oid, &post_oid);
3585        if (status < 0) {
3586                if (state->apply_verbosity > verbosity_silent)
3587                        fprintf(stderr,
3588                                _("Failed to fall back on three-way merge...\n"));
3589                return status;
3590        }
3591
3592        if (status) {
3593                patch->conflicted_threeway = 1;
3594                if (patch->is_new)
3595                        oidclr(&patch->threeway_stage[0]);
3596                else
3597                        oidcpy(&patch->threeway_stage[0], &pre_oid);
3598                oidcpy(&patch->threeway_stage[1], &our_oid);
3599                oidcpy(&patch->threeway_stage[2], &post_oid);
3600                if (state->apply_verbosity > verbosity_silent)
3601                        fprintf(stderr,
3602                                _("Applied patch to '%s' with conflicts.\n"),
3603                                patch->new_name);
3604        } else {
3605                if (state->apply_verbosity > verbosity_silent)
3606                        fprintf(stderr,
3607                                _("Applied patch to '%s' cleanly.\n"),
3608                                patch->new_name);
3609        }
3610        return 0;
3611}
3612
3613static int apply_data(struct apply_state *state, struct patch *patch,
3614                      struct stat *st, const struct cache_entry *ce)
3615{
3616        struct image image;
3617
3618        if (load_preimage(state, &image, patch, st, ce) < 0)
3619                return -1;
3620
3621        if (patch->direct_to_threeway ||
3622            apply_fragments(state, &image, patch) < 0) {
3623                /* Note: with --reject, apply_fragments() returns 0 */
3624                if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3625                        return -1;
3626        }
3627        patch->result = image.buf;
3628        patch->resultsize = image.len;
3629        add_to_fn_table(state, patch);
3630        free(image.line_allocated);
3631
3632        if (0 < patch->is_delete && patch->resultsize)
3633                return error(_("removal patch leaves file contents"));
3634
3635        return 0;
3636}
3637
3638/*
3639 * If "patch" that we are looking at modifies or deletes what we have,
3640 * we would want it not to lose any local modification we have, either
3641 * in the working tree or in the index.
3642 *
3643 * This also decides if a non-git patch is a creation patch or a
3644 * modification to an existing empty file.  We do not check the state
3645 * of the current tree for a creation patch in this function; the caller
3646 * check_patch() separately makes sure (and errors out otherwise) that
3647 * the path the patch creates does not exist in the current tree.
3648 */
3649static int check_preimage(struct apply_state *state,
3650                          struct patch *patch,
3651                          struct cache_entry **ce,
3652                          struct stat *st)
3653{
3654        const char *old_name = patch->old_name;
3655        struct patch *previous = NULL;
3656        int stat_ret = 0, status;
3657        unsigned st_mode = 0;
3658
3659        if (!old_name)
3660                return 0;
3661
3662        assert(patch->is_new <= 0);
3663        previous = previous_patch(state, patch, &status);
3664
3665        if (status)
3666                return error(_("path %s has been renamed/deleted"), old_name);
3667        if (previous) {
3668                st_mode = previous->new_mode;
3669        } else if (!state->cached) {
3670                stat_ret = lstat(old_name, st);
3671                if (stat_ret && errno != ENOENT)
3672                        return error_errno("%s", old_name);
3673        }
3674
3675        if (state->check_index && !previous) {
3676                int pos = cache_name_pos(old_name, strlen(old_name));
3677                if (pos < 0) {
3678                        if (patch->is_new < 0)
3679                                goto is_new;
3680                        return error(_("%s: does not exist in index"), old_name);
3681                }
3682                *ce = active_cache[pos];
3683                if (stat_ret < 0) {
3684                        if (checkout_target(&the_index, *ce, st))
3685                                return -1;
3686                }
3687                if (!state->cached && verify_index_match(*ce, st))
3688                        return error(_("%s: does not match index"), old_name);
3689                if (state->cached)
3690                        st_mode = (*ce)->ce_mode;
3691        } else if (stat_ret < 0) {
3692                if (patch->is_new < 0)
3693                        goto is_new;
3694                return error_errno("%s", old_name);
3695        }
3696
3697        if (!state->cached && !previous)
3698                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3699
3700        if (patch->is_new < 0)
3701                patch->is_new = 0;
3702        if (!patch->old_mode)
3703                patch->old_mode = st_mode;
3704        if ((st_mode ^ patch->old_mode) & S_IFMT)
3705                return error(_("%s: wrong type"), old_name);
3706        if (st_mode != patch->old_mode)
3707                warning(_("%s has type %o, expected %o"),
3708                        old_name, st_mode, patch->old_mode);
3709        if (!patch->new_mode && !patch->is_delete)
3710                patch->new_mode = st_mode;
3711        return 0;
3712
3713 is_new:
3714        patch->is_new = 1;
3715        patch->is_delete = 0;
3716        FREE_AND_NULL(patch->old_name);
3717        return 0;
3718}
3719
3720
3721#define EXISTS_IN_INDEX 1
3722#define EXISTS_IN_WORKTREE 2
3723
3724static int check_to_create(struct apply_state *state,
3725                           const char *new_name,
3726                           int ok_if_exists)
3727{
3728        struct stat nst;
3729
3730        if (state->check_index &&
3731            cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3732            !ok_if_exists)
3733                return EXISTS_IN_INDEX;
3734        if (state->cached)
3735                return 0;
3736
3737        if (!lstat(new_name, &nst)) {
3738                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3739                        return 0;
3740                /*
3741                 * A leading component of new_name might be a symlink
3742                 * that is going to be removed with this patch, but
3743                 * still pointing at somewhere that has the path.
3744                 * In such a case, path "new_name" does not exist as
3745                 * far as git is concerned.
3746                 */
3747                if (has_symlink_leading_path(new_name, strlen(new_name)))
3748                        return 0;
3749
3750                return EXISTS_IN_WORKTREE;
3751        } else if (!is_missing_file_error(errno)) {
3752                return error_errno("%s", new_name);
3753        }
3754        return 0;
3755}
3756
3757static uintptr_t register_symlink_changes(struct apply_state *state,
3758                                          const char *path,
3759                                          uintptr_t what)
3760{
3761        struct string_list_item *ent;
3762
3763        ent = string_list_lookup(&state->symlink_changes, path);
3764        if (!ent) {
3765                ent = string_list_insert(&state->symlink_changes, path);
3766                ent->util = (void *)0;
3767        }
3768        ent->util = (void *)(what | ((uintptr_t)ent->util));
3769        return (uintptr_t)ent->util;
3770}
3771
3772static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3773{
3774        struct string_list_item *ent;
3775
3776        ent = string_list_lookup(&state->symlink_changes, path);
3777        if (!ent)
3778                return 0;
3779        return (uintptr_t)ent->util;
3780}
3781
3782static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3783{
3784        for ( ; patch; patch = patch->next) {
3785                if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3786                    (patch->is_rename || patch->is_delete))
3787                        /* the symlink at patch->old_name is removed */
3788                        register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3789
3790                if (patch->new_name && S_ISLNK(patch->new_mode))
3791                        /* the symlink at patch->new_name is created or remains */
3792                        register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3793        }
3794}
3795
3796static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3797{
3798        do {
3799                unsigned int change;
3800
3801                while (--name->len && name->buf[name->len] != '/')
3802                        ; /* scan backwards */
3803                if (!name->len)
3804                        break;
3805                name->buf[name->len] = '\0';
3806                change = check_symlink_changes(state, name->buf);
3807                if (change & APPLY_SYMLINK_IN_RESULT)
3808                        return 1;
3809                if (change & APPLY_SYMLINK_GOES_AWAY)
3810                        /*
3811                         * This cannot be "return 0", because we may
3812                         * see a new one created at a higher level.
3813                         */
3814                        continue;
3815
3816                /* otherwise, check the preimage */
3817                if (state->check_index) {
3818                        struct cache_entry *ce;
3819
3820                        ce = cache_file_exists(name->buf, name->len, ignore_case);
3821                        if (ce && S_ISLNK(ce->ce_mode))
3822                                return 1;
3823                } else {
3824                        struct stat st;
3825                        if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3826                                return 1;
3827                }
3828        } while (1);
3829        return 0;
3830}
3831
3832static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3833{
3834        int ret;
3835        struct strbuf name = STRBUF_INIT;
3836
3837        assert(*name_ != '\0');
3838        strbuf_addstr(&name, name_);
3839        ret = path_is_beyond_symlink_1(state, &name);
3840        strbuf_release(&name);
3841
3842        return ret;
3843}
3844
3845static int check_unsafe_path(struct patch *patch)
3846{
3847        const char *old_name = NULL;
3848        const char *new_name = NULL;
3849        if (patch->is_delete)
3850                old_name = patch->old_name;
3851        else if (!patch->is_new && !patch->is_copy)
3852                old_name = patch->old_name;
3853        if (!patch->is_delete)
3854                new_name = patch->new_name;
3855
3856        if (old_name && !verify_path(old_name))
3857                return error(_("invalid path '%s'"), old_name);
3858        if (new_name && !verify_path(new_name))
3859                return error(_("invalid path '%s'"), new_name);
3860        return 0;
3861}
3862
3863/*
3864 * Check and apply the patch in-core; leave the result in patch->result
3865 * for the caller to write it out to the final destination.
3866 */
3867static int check_patch(struct apply_state *state, struct patch *patch)
3868{
3869        struct stat st;
3870        const char *old_name = patch->old_name;
3871        const char *new_name = patch->new_name;
3872        const char *name = old_name ? old_name : new_name;
3873        struct cache_entry *ce = NULL;
3874        struct patch *tpatch;
3875        int ok_if_exists;
3876        int status;
3877
3878        patch->rejected = 1; /* we will drop this after we succeed */
3879
3880        status = check_preimage(state, patch, &ce, &st);
3881        if (status)
3882                return status;
3883        old_name = patch->old_name;
3884
3885        /*
3886         * A type-change diff is always split into a patch to delete
3887         * old, immediately followed by a patch to create new (see
3888         * diff.c::run_diff()); in such a case it is Ok that the entry
3889         * to be deleted by the previous patch is still in the working
3890         * tree and in the index.
3891         *
3892         * A patch to swap-rename between A and B would first rename A
3893         * to B and then rename B to A.  While applying the first one,
3894         * the presence of B should not stop A from getting renamed to
3895         * B; ask to_be_deleted() about the later rename.  Removal of
3896         * B and rename from A to B is handled the same way by asking
3897         * was_deleted().
3898         */
3899        if ((tpatch = in_fn_table(state, new_name)) &&
3900            (was_deleted(tpatch) || to_be_deleted(tpatch)))
3901                ok_if_exists = 1;
3902        else
3903                ok_if_exists = 0;
3904
3905        if (new_name &&
3906            ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3907                int err = check_to_create(state, new_name, ok_if_exists);
3908
3909                if (err && state->threeway) {
3910                        patch->direct_to_threeway = 1;
3911                } else switch (err) {
3912                case 0:
3913                        break; /* happy */
3914                case EXISTS_IN_INDEX:
3915                        return error(_("%s: already exists in index"), new_name);
3916                        break;
3917                case EXISTS_IN_WORKTREE:
3918                        return error(_("%s: already exists in working directory"),
3919                                     new_name);
3920                default:
3921                        return err;
3922                }
3923
3924                if (!patch->new_mode) {
3925                        if (0 < patch->is_new)
3926                                patch->new_mode = S_IFREG | 0644;
3927                        else
3928                                patch->new_mode = patch->old_mode;
3929                }
3930        }
3931
3932        if (new_name && old_name) {
3933                int same = !strcmp(old_name, new_name);
3934                if (!patch->new_mode)
3935                        patch->new_mode = patch->old_mode;
3936                if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3937                        if (same)
3938                                return error(_("new mode (%o) of %s does not "
3939                                               "match old mode (%o)"),
3940                                        patch->new_mode, new_name,
3941                                        patch->old_mode);
3942                        else
3943                                return error(_("new mode (%o) of %s does not "
3944                                               "match old mode (%o) of %s"),
3945                                        patch->new_mode, new_name,
3946                                        patch->old_mode, old_name);
3947                }
3948        }
3949
3950        if (!state->unsafe_paths && check_unsafe_path(patch))
3951                return -128;
3952
3953        /*
3954         * An attempt to read from or delete a path that is beyond a
3955         * symbolic link will be prevented by load_patch_target() that
3956         * is called at the beginning of apply_data() so we do not
3957         * have to worry about a patch marked with "is_delete" bit
3958         * here.  We however need to make sure that the patch result
3959         * is not deposited to a path that is beyond a symbolic link
3960         * here.
3961         */
3962        if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3963                return error(_("affected file '%s' is beyond a symbolic link"),
3964                             patch->new_name);
3965
3966        if (apply_data(state, patch, &st, ce) < 0)
3967                return error(_("%s: patch does not apply"), name);
3968        patch->rejected = 0;
3969        return 0;
3970}
3971
3972static int check_patch_list(struct apply_state *state, struct patch *patch)
3973{
3974        int err = 0;
3975
3976        prepare_symlink_changes(state, patch);
3977        prepare_fn_table(state, patch);
3978        while (patch) {
3979                int res;
3980                if (state->apply_verbosity > verbosity_normal)
3981                        say_patch_name(stderr,
3982                                       _("Checking patch %s..."), patch);
3983                res = check_patch(state, patch);
3984                if (res == -128)
3985                        return -128;
3986                err |= res;
3987                patch = patch->next;
3988        }
3989        return err;
3990}
3991
3992static int read_apply_cache(struct apply_state *state)
3993{
3994        if (state->index_file)
3995                return read_cache_from(state->index_file);
3996        else
3997                return read_cache();
3998}
3999
4000/* This function tries to read the object name from the current index */
4001static int get_current_oid(struct apply_state *state, const char *path,
4002                           struct object_id *oid)
4003{
4004        int pos;
4005
4006        if (read_apply_cache(state) < 0)
4007                return -1;
4008        pos = cache_name_pos(path, strlen(path));
4009        if (pos < 0)
4010                return -1;
4011        oidcpy(oid, &active_cache[pos]->oid);
4012        return 0;
4013}
4014
4015static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4016{
4017        /*
4018         * A usable gitlink patch has only one fragment (hunk) that looks like:
4019         * @@ -1 +1 @@
4020         * -Subproject commit <old sha1>
4021         * +Subproject commit <new sha1>
4022         * or
4023         * @@ -1 +0,0 @@
4024         * -Subproject commit <old sha1>
4025         * for a removal patch.
4026         */
4027        struct fragment *hunk = p->fragments;
4028        static const char heading[] = "-Subproject commit ";
4029        char *preimage;
4030
4031        if (/* does the patch have only one hunk? */
4032            hunk && !hunk->next &&
4033            /* is its preimage one line? */
4034            hunk->oldpos == 1 && hunk->oldlines == 1 &&
4035            /* does preimage begin with the heading? */
4036            (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4037            starts_with(++preimage, heading) &&
4038            /* does it record full SHA-1? */
4039            !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4040            preimage[sizeof(heading) + GIT_SHA1_HEXSZ - 1] == '\n' &&
4041            /* does the abbreviated name on the index line agree with it? */
4042            starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
4043                return 0; /* it all looks fine */
4044
4045        /* we may have full object name on the index line */
4046        return get_oid_hex(p->old_sha1_prefix, oid);
4047}
4048
4049/* Build an index that contains the just the files needed for a 3way merge */
4050static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4051{
4052        struct patch *patch;
4053        struct index_state result = { NULL };
4054        static struct lock_file lock;
4055        int res;
4056
4057        /* Once we start supporting the reverse patch, it may be
4058         * worth showing the new sha1 prefix, but until then...
4059         */
4060        for (patch = list; patch; patch = patch->next) {
4061                struct object_id oid;
4062                struct cache_entry *ce;
4063                const char *name;
4064
4065                name = patch->old_name ? patch->old_name : patch->new_name;
4066                if (0 < patch->is_new)
4067                        continue;
4068
4069                if (S_ISGITLINK(patch->old_mode)) {
4070                        if (!preimage_oid_in_gitlink_patch(patch, &oid))
4071                                ; /* ok, the textual part looks sane */
4072                        else
4073                                return error(_("sha1 information is lacking or "
4074                                               "useless for submodule %s"), name);
4075                } else if (!get_oid_blob(patch->old_sha1_prefix, &oid)) {
4076                        ; /* ok */
4077                } else if (!patch->lines_added && !patch->lines_deleted) {
4078                        /* mode-only change: update the current */
4079                        if (get_current_oid(state, patch->old_name, &oid))
4080                                return error(_("mode change for %s, which is not "
4081                                               "in current HEAD"), name);
4082                } else
4083                        return error(_("sha1 information is lacking or useless "
4084                                       "(%s)."), name);
4085
4086                ce = make_cache_entry(patch->old_mode, oid.hash, name, 0, 0);
4087                if (!ce)
4088                        return error(_("make_cache_entry failed for path '%s'"),
4089                                     name);
4090                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4091                        free(ce);
4092                        return error(_("could not add %s to temporary index"),
4093                                     name);
4094                }
4095        }
4096
4097        hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4098        res = write_locked_index(&result, &lock, COMMIT_LOCK);
4099        discard_index(&result);
4100
4101        if (res)
4102                return error(_("could not write temporary index to %s"),
4103                             state->fake_ancestor);
4104
4105        return 0;
4106}
4107
4108static void stat_patch_list(struct apply_state *state, struct patch *patch)
4109{
4110        int files, adds, dels;
4111
4112        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4113                files++;
4114                adds += patch->lines_added;
4115                dels += patch->lines_deleted;
4116                show_stats(state, patch);
4117        }
4118
4119        print_stat_summary(stdout, files, adds, dels);
4120}
4121
4122static void numstat_patch_list(struct apply_state *state,
4123                               struct patch *patch)
4124{
4125        for ( ; patch; patch = patch->next) {
4126                const char *name;
4127                name = patch->new_name ? patch->new_name : patch->old_name;
4128                if (patch->is_binary)
4129                        printf("-\t-\t");
4130                else
4131                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4132                write_name_quoted(name, stdout, state->line_termination);
4133        }
4134}
4135
4136static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4137{
4138        if (mode)
4139                printf(" %s mode %06o %s\n", newdelete, mode, name);
4140        else
4141                printf(" %s %s\n", newdelete, name);
4142}
4143
4144static void show_mode_change(struct patch *p, int show_name)
4145{
4146        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4147                if (show_name)
4148                        printf(" mode change %06o => %06o %s\n",
4149                               p->old_mode, p->new_mode, p->new_name);
4150                else
4151                        printf(" mode change %06o => %06o\n",
4152                               p->old_mode, p->new_mode);
4153        }
4154}
4155
4156static void show_rename_copy(struct patch *p)
4157{
4158        const char *renamecopy = p->is_rename ? "rename" : "copy";
4159        const char *old, *new;
4160
4161        /* Find common prefix */
4162        old = p->old_name;
4163        new = p->new_name;
4164        while (1) {
4165                const char *slash_old, *slash_new;
4166                slash_old = strchr(old, '/');
4167                slash_new = strchr(new, '/');
4168                if (!slash_old ||
4169                    !slash_new ||
4170                    slash_old - old != slash_new - new ||
4171                    memcmp(old, new, slash_new - new))
4172                        break;
4173                old = slash_old + 1;
4174                new = slash_new + 1;
4175        }
4176        /* p->old_name thru old is the common prefix, and old and new
4177         * through the end of names are renames
4178         */
4179        if (old != p->old_name)
4180                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4181                       (int)(old - p->old_name), p->old_name,
4182                       old, new, p->score);
4183        else
4184                printf(" %s %s => %s (%d%%)\n", renamecopy,
4185                       p->old_name, p->new_name, p->score);
4186        show_mode_change(p, 0);
4187}
4188
4189static void summary_patch_list(struct patch *patch)
4190{
4191        struct patch *p;
4192
4193        for (p = patch; p; p = p->next) {
4194                if (p->is_new)
4195                        show_file_mode_name("create", p->new_mode, p->new_name);
4196                else if (p->is_delete)
4197                        show_file_mode_name("delete", p->old_mode, p->old_name);
4198                else {
4199                        if (p->is_rename || p->is_copy)
4200                                show_rename_copy(p);
4201                        else {
4202                                if (p->score) {
4203                                        printf(" rewrite %s (%d%%)\n",
4204                                               p->new_name, p->score);
4205                                        show_mode_change(p, 0);
4206                                }
4207                                else
4208                                        show_mode_change(p, 1);
4209                        }
4210                }
4211        }
4212}
4213
4214static void patch_stats(struct apply_state *state, struct patch *patch)
4215{
4216        int lines = patch->lines_added + patch->lines_deleted;
4217
4218        if (lines > state->max_change)
4219                state->max_change = lines;
4220        if (patch->old_name) {
4221                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4222                if (!len)
4223                        len = strlen(patch->old_name);
4224                if (len > state->max_len)
4225                        state->max_len = len;
4226        }
4227        if (patch->new_name) {
4228                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4229                if (!len)
4230                        len = strlen(patch->new_name);
4231                if (len > state->max_len)
4232                        state->max_len = len;
4233        }
4234}
4235
4236static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4237{
4238        if (state->update_index) {
4239                if (remove_file_from_cache(patch->old_name) < 0)
4240                        return error(_("unable to remove %s from index"), patch->old_name);
4241        }
4242        if (!state->cached) {
4243                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4244                        remove_path(patch->old_name);
4245                }
4246        }
4247        return 0;
4248}
4249
4250static int add_index_file(struct apply_state *state,
4251                          const char *path,
4252                          unsigned mode,
4253                          void *buf,
4254                          unsigned long size)
4255{
4256        struct stat st;
4257        struct cache_entry *ce;
4258        int namelen = strlen(path);
4259        unsigned ce_size = cache_entry_size(namelen);
4260
4261        if (!state->update_index)
4262                return 0;
4263
4264        ce = xcalloc(1, ce_size);
4265        memcpy(ce->name, path, namelen);
4266        ce->ce_mode = create_ce_mode(mode);
4267        ce->ce_flags = create_ce_flags(0);
4268        ce->ce_namelen = namelen;
4269        if (S_ISGITLINK(mode)) {
4270                const char *s;
4271
4272                if (!skip_prefix(buf, "Subproject commit ", &s) ||
4273                    get_oid_hex(s, &ce->oid)) {
4274                        free(ce);
4275                       return error(_("corrupt patch for submodule %s"), path);
4276                }
4277        } else {
4278                if (!state->cached) {
4279                        if (lstat(path, &st) < 0) {
4280                                free(ce);
4281                                return error_errno(_("unable to stat newly "
4282                                                     "created file '%s'"),
4283                                                   path);
4284                        }
4285                        fill_stat_cache_info(ce, &st);
4286                }
4287                if (write_sha1_file(buf, size, blob_type, ce->oid.hash) < 0) {
4288                        free(ce);
4289                        return error(_("unable to create backing store "
4290                                       "for newly created file %s"), path);
4291                }
4292        }
4293        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4294                free(ce);
4295                return error(_("unable to add cache entry for %s"), path);
4296        }
4297
4298        return 0;
4299}
4300
4301/*
4302 * Returns:
4303 *  -1 if an unrecoverable error happened
4304 *   0 if everything went well
4305 *   1 if a recoverable error happened
4306 */
4307static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4308{
4309        int fd, res;
4310        struct strbuf nbuf = STRBUF_INIT;
4311
4312        if (S_ISGITLINK(mode)) {
4313                struct stat st;
4314                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4315                        return 0;
4316                return !!mkdir(path, 0777);
4317        }
4318
4319        if (has_symlinks && S_ISLNK(mode))
4320                /* Although buf:size is counted string, it also is NUL
4321                 * terminated.
4322                 */
4323                return !!symlink(buf, path);
4324
4325        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4326        if (fd < 0)
4327                return 1;
4328
4329        if (convert_to_working_tree(path, buf, size, &nbuf)) {
4330                size = nbuf.len;
4331                buf  = nbuf.buf;
4332        }
4333
4334        res = write_in_full(fd, buf, size) < 0;
4335        if (res)
4336                error_errno(_("failed to write to '%s'"), path);
4337        strbuf_release(&nbuf);
4338
4339        if (close(fd) < 0 && !res)
4340                return error_errno(_("closing file '%s'"), path);
4341
4342        return res ? -1 : 0;
4343}
4344
4345/*
4346 * We optimistically assume that the directories exist,
4347 * which is true 99% of the time anyway. If they don't,
4348 * we create them and try again.
4349 *
4350 * Returns:
4351 *   -1 on error
4352 *   0 otherwise
4353 */
4354static int create_one_file(struct apply_state *state,
4355                           char *path,
4356                           unsigned mode,
4357                           const char *buf,
4358                           unsigned long size)
4359{
4360        int res;
4361
4362        if (state->cached)
4363                return 0;
4364
4365        res = try_create_file(path, mode, buf, size);
4366        if (res < 0)
4367                return -1;
4368        if (!res)
4369                return 0;
4370
4371        if (errno == ENOENT) {
4372                if (safe_create_leading_directories(path))
4373                        return 0;
4374                res = try_create_file(path, mode, buf, size);
4375                if (res < 0)
4376                        return -1;
4377                if (!res)
4378                        return 0;
4379        }
4380
4381        if (errno == EEXIST || errno == EACCES) {
4382                /* We may be trying to create a file where a directory
4383                 * used to be.
4384                 */
4385                struct stat st;
4386                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4387                        errno = EEXIST;
4388        }
4389
4390        if (errno == EEXIST) {
4391                unsigned int nr = getpid();
4392
4393                for (;;) {
4394                        char newpath[PATH_MAX];
4395                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4396                        res = try_create_file(newpath, mode, buf, size);
4397                        if (res < 0)
4398                                return -1;
4399                        if (!res) {
4400                                if (!rename(newpath, path))
4401                                        return 0;
4402                                unlink_or_warn(newpath);
4403                                break;
4404                        }
4405                        if (errno != EEXIST)
4406                                break;
4407                        ++nr;
4408                }
4409        }
4410        return error_errno(_("unable to write file '%s' mode %o"),
4411                           path, mode);
4412}
4413
4414static int add_conflicted_stages_file(struct apply_state *state,
4415                                       struct patch *patch)
4416{
4417        int stage, namelen;
4418        unsigned ce_size, mode;
4419        struct cache_entry *ce;
4420
4421        if (!state->update_index)
4422                return 0;
4423        namelen = strlen(patch->new_name);
4424        ce_size = cache_entry_size(namelen);
4425        mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4426
4427        remove_file_from_cache(patch->new_name);
4428        for (stage = 1; stage < 4; stage++) {
4429                if (is_null_oid(&patch->threeway_stage[stage - 1]))
4430                        continue;
4431                ce = xcalloc(1, ce_size);
4432                memcpy(ce->name, patch->new_name, namelen);
4433                ce->ce_mode = create_ce_mode(mode);
4434                ce->ce_flags = create_ce_flags(stage);
4435                ce->ce_namelen = namelen;
4436                oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4437                if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4438                        free(ce);
4439                        return error(_("unable to add cache entry for %s"),
4440                                     patch->new_name);
4441                }
4442        }
4443
4444        return 0;
4445}
4446
4447static int create_file(struct apply_state *state, struct patch *patch)
4448{
4449        char *path = patch->new_name;
4450        unsigned mode = patch->new_mode;
4451        unsigned long size = patch->resultsize;
4452        char *buf = patch->result;
4453
4454        if (!mode)
4455                mode = S_IFREG | 0644;
4456        if (create_one_file(state, path, mode, buf, size))
4457                return -1;
4458
4459        if (patch->conflicted_threeway)
4460                return add_conflicted_stages_file(state, patch);
4461        else
4462                return add_index_file(state, path, mode, buf, size);
4463}
4464
4465/* phase zero is to remove, phase one is to create */
4466static int write_out_one_result(struct apply_state *state,
4467                                struct patch *patch,
4468                                int phase)
4469{
4470        if (patch->is_delete > 0) {
4471                if (phase == 0)
4472                        return remove_file(state, patch, 1);
4473                return 0;
4474        }
4475        if (patch->is_new > 0 || patch->is_copy) {
4476                if (phase == 1)
4477                        return create_file(state, patch);
4478                return 0;
4479        }
4480        /*
4481         * Rename or modification boils down to the same
4482         * thing: remove the old, write the new
4483         */
4484        if (phase == 0)
4485                return remove_file(state, patch, patch->is_rename);
4486        if (phase == 1)
4487                return create_file(state, patch);
4488        return 0;
4489}
4490
4491static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4492{
4493        FILE *rej;
4494        char namebuf[PATH_MAX];
4495        struct fragment *frag;
4496        int cnt = 0;
4497        struct strbuf sb = STRBUF_INIT;
4498
4499        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4500                if (!frag->rejected)
4501                        continue;
4502                cnt++;
4503        }
4504
4505        if (!cnt) {
4506                if (state->apply_verbosity > verbosity_normal)
4507                        say_patch_name(stderr,
4508                                       _("Applied patch %s cleanly."), patch);
4509                return 0;
4510        }
4511
4512        /* This should not happen, because a removal patch that leaves
4513         * contents are marked "rejected" at the patch level.
4514         */
4515        if (!patch->new_name)
4516                die(_("internal error"));
4517
4518        /* Say this even without --verbose */
4519        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4520                            "Applying patch %%s with %d rejects...",
4521                            cnt),
4522                    cnt);
4523        if (state->apply_verbosity > verbosity_silent)
4524                say_patch_name(stderr, sb.buf, patch);
4525        strbuf_release(&sb);
4526
4527        cnt = strlen(patch->new_name);
4528        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4529                cnt = ARRAY_SIZE(namebuf) - 5;
4530                warning(_("truncating .rej filename to %.*s.rej"),
4531                        cnt - 1, patch->new_name);
4532        }
4533        memcpy(namebuf, patch->new_name, cnt);
4534        memcpy(namebuf + cnt, ".rej", 5);
4535
4536        rej = fopen(namebuf, "w");
4537        if (!rej)
4538                return error_errno(_("cannot open %s"), namebuf);
4539
4540        /* Normal git tools never deal with .rej, so do not pretend
4541         * this is a git patch by saying --git or giving extended
4542         * headers.  While at it, maybe please "kompare" that wants
4543         * the trailing TAB and some garbage at the end of line ;-).
4544         */
4545        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4546                patch->new_name, patch->new_name);
4547        for (cnt = 1, frag = patch->fragments;
4548             frag;
4549             cnt++, frag = frag->next) {
4550                if (!frag->rejected) {
4551                        if (state->apply_verbosity > verbosity_silent)
4552                                fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4553                        continue;
4554                }
4555                if (state->apply_verbosity > verbosity_silent)
4556                        fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4557                fprintf(rej, "%.*s", frag->size, frag->patch);
4558                if (frag->patch[frag->size-1] != '\n')
4559                        fputc('\n', rej);
4560        }
4561        fclose(rej);
4562        return -1;
4563}
4564
4565/*
4566 * Returns:
4567 *  -1 if an error happened
4568 *   0 if the patch applied cleanly
4569 *   1 if the patch did not apply cleanly
4570 */
4571static int write_out_results(struct apply_state *state, struct patch *list)
4572{
4573        int phase;
4574        int errs = 0;
4575        struct patch *l;
4576        struct string_list cpath = STRING_LIST_INIT_DUP;
4577
4578        for (phase = 0; phase < 2; phase++) {
4579                l = list;
4580                while (l) {
4581                        if (l->rejected)
4582                                errs = 1;
4583                        else {
4584                                if (write_out_one_result(state, l, phase)) {
4585                                        string_list_clear(&cpath, 0);
4586                                        return -1;
4587                                }
4588                                if (phase == 1) {
4589                                        if (write_out_one_reject(state, l))
4590                                                errs = 1;
4591                                        if (l->conflicted_threeway) {
4592                                                string_list_append(&cpath, l->new_name);
4593                                                errs = 1;
4594                                        }
4595                                }
4596                        }
4597                        l = l->next;
4598                }
4599        }
4600
4601        if (cpath.nr) {
4602                struct string_list_item *item;
4603
4604                string_list_sort(&cpath);
4605                if (state->apply_verbosity > verbosity_silent) {
4606                        for_each_string_list_item(item, &cpath)
4607                                fprintf(stderr, "U %s\n", item->string);
4608                }
4609                string_list_clear(&cpath, 0);
4610
4611                rerere(0);
4612        }
4613
4614        return errs;
4615}
4616
4617/*
4618 * Try to apply a patch.
4619 *
4620 * Returns:
4621 *  -128 if a bad error happened (like patch unreadable)
4622 *  -1 if patch did not apply and user cannot deal with it
4623 *   0 if the patch applied
4624 *   1 if the patch did not apply but user might fix it
4625 */
4626static int apply_patch(struct apply_state *state,
4627                       int fd,
4628                       const char *filename,
4629                       int options)
4630{
4631        size_t offset;
4632        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4633        struct patch *list = NULL, **listp = &list;
4634        int skipped_patch = 0;
4635        int res = 0;
4636
4637        state->patch_input_file = filename;
4638        if (read_patch_file(&buf, fd) < 0)
4639                return -128;
4640        offset = 0;
4641        while (offset < buf.len) {
4642                struct patch *patch;
4643                int nr;
4644
4645                patch = xcalloc(1, sizeof(*patch));
4646                patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4647                patch->recount =  !!(options & APPLY_OPT_RECOUNT);
4648                nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4649                if (nr < 0) {
4650                        free_patch(patch);
4651                        if (nr == -128) {
4652                                res = -128;
4653                                goto end;
4654                        }
4655                        break;
4656                }
4657                if (state->apply_in_reverse)
4658                        reverse_patches(patch);
4659                if (use_patch(state, patch)) {
4660                        patch_stats(state, patch);
4661                        *listp = patch;
4662                        listp = &patch->next;
4663                }
4664                else {
4665                        if (state->apply_verbosity > verbosity_normal)
4666                                say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4667                        free_patch(patch);
4668                        skipped_patch++;
4669                }
4670                offset += nr;
4671        }
4672
4673        if (!list && !skipped_patch) {
4674                error(_("unrecognized input"));
4675                res = -128;
4676                goto end;
4677        }
4678
4679        if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4680                state->apply = 0;
4681
4682        state->update_index = state->check_index && state->apply;
4683        if (state->update_index && state->newfd < 0) {
4684                if (state->index_file)
4685                        state->newfd = hold_lock_file_for_update(state->lock_file,
4686                                                                 state->index_file,
4687                                                                 LOCK_DIE_ON_ERROR);
4688                else
4689                        state->newfd = hold_locked_index(state->lock_file, LOCK_DIE_ON_ERROR);
4690        }
4691
4692        if (state->check_index && read_apply_cache(state) < 0) {
4693                error(_("unable to read index file"));
4694                res = -128;
4695                goto end;
4696        }
4697
4698        if (state->check || state->apply) {
4699                int r = check_patch_list(state, list);
4700                if (r == -128) {
4701                        res = -128;
4702                        goto end;
4703                }
4704                if (r < 0 && !state->apply_with_reject) {
4705                        res = -1;
4706                        goto end;
4707                }
4708        }
4709
4710        if (state->apply) {
4711                int write_res = write_out_results(state, list);
4712                if (write_res < 0) {
4713                        res = -128;
4714                        goto end;
4715                }
4716                if (write_res > 0) {
4717                        /* with --3way, we still need to write the index out */
4718                        res = state->apply_with_reject ? -1 : 1;
4719                        goto end;
4720                }
4721        }
4722
4723        if (state->fake_ancestor &&
4724            build_fake_ancestor(state, list)) {
4725                res = -128;
4726                goto end;
4727        }
4728
4729        if (state->diffstat && state->apply_verbosity > verbosity_silent)
4730                stat_patch_list(state, list);
4731
4732        if (state->numstat && state->apply_verbosity > verbosity_silent)
4733                numstat_patch_list(state, list);
4734
4735        if (state->summary && state->apply_verbosity > verbosity_silent)
4736                summary_patch_list(list);
4737
4738end:
4739        free_patch_list(list);
4740        strbuf_release(&buf);
4741        string_list_clear(&state->fn_table, 0);
4742        return res;
4743}
4744
4745static int apply_option_parse_exclude(const struct option *opt,
4746                                      const char *arg, int unset)
4747{
4748        struct apply_state *state = opt->value;
4749        add_name_limit(state, arg, 1);
4750        return 0;
4751}
4752
4753static int apply_option_parse_include(const struct option *opt,
4754                                      const char *arg, int unset)
4755{
4756        struct apply_state *state = opt->value;
4757        add_name_limit(state, arg, 0);
4758        state->has_include = 1;
4759        return 0;
4760}
4761
4762static int apply_option_parse_p(const struct option *opt,
4763                                const char *arg,
4764                                int unset)
4765{
4766        struct apply_state *state = opt->value;
4767        state->p_value = atoi(arg);
4768        state->p_value_known = 1;
4769        return 0;
4770}
4771
4772static int apply_option_parse_space_change(const struct option *opt,
4773                                           const char *arg, int unset)
4774{
4775        struct apply_state *state = opt->value;
4776        if (unset)
4777                state->ws_ignore_action = ignore_ws_none;
4778        else
4779                state->ws_ignore_action = ignore_ws_change;
4780        return 0;
4781}
4782
4783static int apply_option_parse_whitespace(const struct option *opt,
4784                                         const char *arg, int unset)
4785{
4786        struct apply_state *state = opt->value;
4787        state->whitespace_option = arg;
4788        if (parse_whitespace_option(state, arg))
4789                exit(1);
4790        return 0;
4791}
4792
4793static int apply_option_parse_directory(const struct option *opt,
4794                                        const char *arg, int unset)
4795{
4796        struct apply_state *state = opt->value;
4797        strbuf_reset(&state->root);
4798        strbuf_addstr(&state->root, arg);
4799        strbuf_complete(&state->root, '/');
4800        return 0;
4801}
4802
4803int apply_all_patches(struct apply_state *state,
4804                      int argc,
4805                      const char **argv,
4806                      int options)
4807{
4808        int i;
4809        int res;
4810        int errs = 0;
4811        int read_stdin = 1;
4812
4813        for (i = 0; i < argc; i++) {
4814                const char *arg = argv[i];
4815                char *to_free = NULL;
4816                int fd;
4817
4818                if (!strcmp(arg, "-")) {
4819                        res = apply_patch(state, 0, "<stdin>", options);
4820                        if (res < 0)
4821                                goto end;
4822                        errs |= res;
4823                        read_stdin = 0;
4824                        continue;
4825                } else
4826                        arg = to_free = prefix_filename(state->prefix, arg);
4827
4828                fd = open(arg, O_RDONLY);
4829                if (fd < 0) {
4830                        error(_("can't open patch '%s': %s"), arg, strerror(errno));
4831                        res = -128;
4832                        free(to_free);
4833                        goto end;
4834                }
4835                read_stdin = 0;
4836                set_default_whitespace_mode(state);
4837                res = apply_patch(state, fd, arg, options);
4838                close(fd);
4839                free(to_free);
4840                if (res < 0)
4841                        goto end;
4842                errs |= res;
4843        }
4844        set_default_whitespace_mode(state);
4845        if (read_stdin) {
4846                res = apply_patch(state, 0, "<stdin>", options);
4847                if (res < 0)
4848                        goto end;
4849                errs |= res;
4850        }
4851
4852        if (state->whitespace_error) {
4853                if (state->squelch_whitespace_errors &&
4854                    state->squelch_whitespace_errors < state->whitespace_error) {
4855                        int squelched =
4856                                state->whitespace_error - state->squelch_whitespace_errors;
4857                        warning(Q_("squelched %d whitespace error",
4858                                   "squelched %d whitespace errors",
4859                                   squelched),
4860                                squelched);
4861                }
4862                if (state->ws_error_action == die_on_ws_error) {
4863                        error(Q_("%d line adds whitespace errors.",
4864                                 "%d lines add whitespace errors.",
4865                                 state->whitespace_error),
4866                              state->whitespace_error);
4867                        res = -128;
4868                        goto end;
4869                }
4870                if (state->applied_after_fixing_ws && state->apply)
4871                        warning(Q_("%d line applied after"
4872                                   " fixing whitespace errors.",
4873                                   "%d lines applied after"
4874                                   " fixing whitespace errors.",
4875                                   state->applied_after_fixing_ws),
4876                                state->applied_after_fixing_ws);
4877                else if (state->whitespace_error)
4878                        warning(Q_("%d line adds whitespace errors.",
4879                                   "%d lines add whitespace errors.",
4880                                   state->whitespace_error),
4881                                state->whitespace_error);
4882        }
4883
4884        if (state->update_index) {
4885                res = write_locked_index(&the_index, state->lock_file, COMMIT_LOCK);
4886                if (res) {
4887                        error(_("Unable to write new index file"));
4888                        res = -128;
4889                        goto end;
4890                }
4891                state->newfd = -1;
4892        }
4893
4894        res = !!errs;
4895
4896end:
4897        if (state->newfd >= 0) {
4898                rollback_lock_file(state->lock_file);
4899                state->newfd = -1;
4900        }
4901
4902        if (state->apply_verbosity <= verbosity_silent) {
4903                set_error_routine(state->saved_error_routine);
4904                set_warn_routine(state->saved_warn_routine);
4905        }
4906
4907        if (res > -1)
4908                return res;
4909        return (res == -1 ? 1 : 128);
4910}
4911
4912int apply_parse_options(int argc, const char **argv,
4913                        struct apply_state *state,
4914                        int *force_apply, int *options,
4915                        const char * const *apply_usage)
4916{
4917        struct option builtin_apply_options[] = {
4918                { OPTION_CALLBACK, 0, "exclude", state, N_("path"),
4919                        N_("don't apply changes matching the given path"),
4920                        0, apply_option_parse_exclude },
4921                { OPTION_CALLBACK, 0, "include", state, N_("path"),
4922                        N_("apply changes matching the given path"),
4923                        0, apply_option_parse_include },
4924                { OPTION_CALLBACK, 'p', NULL, state, N_("num"),
4925                        N_("remove <num> leading slashes from traditional diff paths"),
4926                        0, apply_option_parse_p },
4927                OPT_BOOL(0, "no-add", &state->no_add,
4928                        N_("ignore additions made by the patch")),
4929                OPT_BOOL(0, "stat", &state->diffstat,
4930                        N_("instead of applying the patch, output diffstat for the input")),
4931                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4932                OPT_NOOP_NOARG(0, "binary"),
4933                OPT_BOOL(0, "numstat", &state->numstat,
4934                        N_("show number of added and deleted lines in decimal notation")),
4935                OPT_BOOL(0, "summary", &state->summary,
4936                        N_("instead of applying the patch, output a summary for the input")),
4937                OPT_BOOL(0, "check", &state->check,
4938                        N_("instead of applying the patch, see if the patch is applicable")),
4939                OPT_BOOL(0, "index", &state->check_index,
4940                        N_("make sure the patch is applicable to the current index")),
4941                OPT_BOOL(0, "cached", &state->cached,
4942                        N_("apply a patch without touching the working tree")),
4943                OPT_BOOL(0, "unsafe-paths", &state->unsafe_paths,
4944                        N_("accept a patch that touches outside the working area")),
4945                OPT_BOOL(0, "apply", force_apply,
4946                        N_("also apply the patch (use with --stat/--summary/--check)")),
4947                OPT_BOOL('3', "3way", &state->threeway,
4948                         N_( "attempt three-way merge if a patch does not apply")),
4949                OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
4950                        N_("build a temporary index based on embedded index information")),
4951                /* Think twice before adding "--nul" synonym to this */
4952                OPT_SET_INT('z', NULL, &state->line_termination,
4953                        N_("paths are separated with NUL character"), '\0'),
4954                OPT_INTEGER('C', NULL, &state->p_context,
4955                                N_("ensure at least <n> lines of context match")),
4956                { OPTION_CALLBACK, 0, "whitespace", state, N_("action"),
4957                        N_("detect new or modified lines that have whitespace errors"),
4958                        0, apply_option_parse_whitespace },
4959                { OPTION_CALLBACK, 0, "ignore-space-change", state, NULL,
4960                        N_("ignore changes in whitespace when finding context"),
4961                        PARSE_OPT_NOARG, apply_option_parse_space_change },
4962                { OPTION_CALLBACK, 0, "ignore-whitespace", state, NULL,
4963                        N_("ignore changes in whitespace when finding context"),
4964                        PARSE_OPT_NOARG, apply_option_parse_space_change },
4965                OPT_BOOL('R', "reverse", &state->apply_in_reverse,
4966                        N_("apply the patch in reverse")),
4967                OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
4968                        N_("don't expect at least one line of context")),
4969                OPT_BOOL(0, "reject", &state->apply_with_reject,
4970                        N_("leave the rejected hunks in corresponding *.rej files")),
4971                OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
4972                        N_("allow overlapping hunks")),
4973                OPT__VERBOSE(&state->apply_verbosity, N_("be verbose")),
4974                OPT_BIT(0, "inaccurate-eof", options,
4975                        N_("tolerate incorrectly detected missing new-line at the end of file"),
4976                        APPLY_OPT_INACCURATE_EOF),
4977                OPT_BIT(0, "recount", options,
4978                        N_("do not trust the line counts in the hunk headers"),
4979                        APPLY_OPT_RECOUNT),
4980                { OPTION_CALLBACK, 0, "directory", state, N_("root"),
4981                        N_("prepend <root> to all filenames"),
4982                        0, apply_option_parse_directory },
4983                OPT_END()
4984        };
4985
4986        return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);
4987}