619b8fb1255e757d7646d5377405e5b8ae18ab3c
   1/*
   2 * apply.c
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This applies patches on top of some (arbitrary) version of the SCM.
   7 *
   8 */
   9#include "cache.h"
  10#include "lockfile.h"
  11#include "cache-tree.h"
  12#include "quote.h"
  13#include "blob.h"
  14#include "delta.h"
  15#include "builtin.h"
  16#include "string-list.h"
  17#include "dir.h"
  18#include "diff.h"
  19#include "parse-options.h"
  20#include "xdiff-interface.h"
  21#include "ll-merge.h"
  22#include "rerere.h"
  23
  24struct apply_state {
  25        const char *prefix;
  26        int prefix_length;
  27
  28        /* These control what gets looked at and modified */
  29        int apply; /* this is not a dry-run */
  30        int cached; /* apply to the index only */
  31        int check; /* preimage must match working tree, don't actually apply */
  32        int check_index; /* preimage must match the indexed version */
  33        int update_index; /* check_index && apply */
  34
  35        /* These control cosmetic aspect of the output */
  36        int diffstat; /* just show a diffstat, and don't actually apply */
  37        int numstat; /* just show a numeric diffstat, and don't actually apply */
  38        int summary; /* just report creation, deletion, etc, and don't actually apply */
  39
  40        /* These boolean parameters control how the apply is done */
  41        int allow_overlap;
  42        int apply_in_reverse;
  43        int apply_with_reject;
  44        int apply_verbosely;
  45        int no_add;
  46        int threeway;
  47        int unidiff_zero;
  48        int unsafe_paths;
  49
  50        /* Other non boolean parameters */
  51        const char *fake_ancestor;
  52        const char *patch_input_file;
  53        int line_termination;
  54        struct strbuf root;
  55        int p_value;
  56        int p_value_known;
  57        unsigned int p_context;
  58
  59        /* Exclude and include path parameters */
  60        struct string_list limit_by_name;
  61        int has_include;
  62
  63        /* These control whitespace errors */
  64        const char *whitespace_option;
  65        int whitespace_error;
  66        int squelch_whitespace_errors;
  67};
  68
  69static int newfd = -1;
  70
  71static const char * const apply_usage[] = {
  72        N_("git apply [<options>] [<patch>...]"),
  73        NULL
  74};
  75
  76static enum ws_error_action {
  77        nowarn_ws_error,
  78        warn_on_ws_error,
  79        die_on_ws_error,
  80        correct_ws_error
  81} ws_error_action = warn_on_ws_error;
  82static int applied_after_fixing_ws;
  83
  84static enum ws_ignore {
  85        ignore_ws_none,
  86        ignore_ws_change
  87} ws_ignore_action = ignore_ws_none;
  88
  89
  90static void parse_whitespace_option(struct apply_state *state, const char *option)
  91{
  92        if (!option) {
  93                ws_error_action = warn_on_ws_error;
  94                return;
  95        }
  96        if (!strcmp(option, "warn")) {
  97                ws_error_action = warn_on_ws_error;
  98                return;
  99        }
 100        if (!strcmp(option, "nowarn")) {
 101                ws_error_action = nowarn_ws_error;
 102                return;
 103        }
 104        if (!strcmp(option, "error")) {
 105                ws_error_action = die_on_ws_error;
 106                return;
 107        }
 108        if (!strcmp(option, "error-all")) {
 109                ws_error_action = die_on_ws_error;
 110                state->squelch_whitespace_errors = 0;
 111                return;
 112        }
 113        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
 114                ws_error_action = correct_ws_error;
 115                return;
 116        }
 117        die(_("unrecognized whitespace option '%s'"), option);
 118}
 119
 120static void parse_ignorewhitespace_option(const char *option)
 121{
 122        if (!option || !strcmp(option, "no") ||
 123            !strcmp(option, "false") || !strcmp(option, "never") ||
 124            !strcmp(option, "none")) {
 125                ws_ignore_action = ignore_ws_none;
 126                return;
 127        }
 128        if (!strcmp(option, "change")) {
 129                ws_ignore_action = ignore_ws_change;
 130                return;
 131        }
 132        die(_("unrecognized whitespace ignore option '%s'"), option);
 133}
 134
 135static void set_default_whitespace_mode(struct apply_state *state)
 136{
 137        if (!state->whitespace_option && !apply_default_whitespace)
 138                ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
 139}
 140
 141/*
 142 * For "diff-stat" like behaviour, we keep track of the biggest change
 143 * we've seen, and the longest filename. That allows us to do simple
 144 * scaling.
 145 */
 146static int max_change, max_len;
 147
 148/*
 149 * Various "current state", notably line numbers and what
 150 * file (and how) we're patching right now.. The "is_xxxx"
 151 * things are flags, where -1 means "don't know yet".
 152 */
 153static int state_linenr = 1;
 154
 155/*
 156 * This represents one "hunk" from a patch, starting with
 157 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 158 * patch text is pointed at by patch, and its byte length
 159 * is stored in size.  leading and trailing are the number
 160 * of context lines.
 161 */
 162struct fragment {
 163        unsigned long leading, trailing;
 164        unsigned long oldpos, oldlines;
 165        unsigned long newpos, newlines;
 166        /*
 167         * 'patch' is usually borrowed from buf in apply_patch(),
 168         * but some codepaths store an allocated buffer.
 169         */
 170        const char *patch;
 171        unsigned free_patch:1,
 172                rejected:1;
 173        int size;
 174        int linenr;
 175        struct fragment *next;
 176};
 177
 178/*
 179 * When dealing with a binary patch, we reuse "leading" field
 180 * to store the type of the binary hunk, either deflated "delta"
 181 * or deflated "literal".
 182 */
 183#define binary_patch_method leading
 184#define BINARY_DELTA_DEFLATED   1
 185#define BINARY_LITERAL_DEFLATED 2
 186
 187/*
 188 * This represents a "patch" to a file, both metainfo changes
 189 * such as creation/deletion, filemode and content changes represented
 190 * as a series of fragments.
 191 */
 192struct patch {
 193        char *new_name, *old_name, *def_name;
 194        unsigned int old_mode, new_mode;
 195        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 196        int rejected;
 197        unsigned ws_rule;
 198        int lines_added, lines_deleted;
 199        int score;
 200        unsigned int is_toplevel_relative:1;
 201        unsigned int inaccurate_eof:1;
 202        unsigned int is_binary:1;
 203        unsigned int is_copy:1;
 204        unsigned int is_rename:1;
 205        unsigned int recount:1;
 206        unsigned int conflicted_threeway:1;
 207        unsigned int direct_to_threeway:1;
 208        struct fragment *fragments;
 209        char *result;
 210        size_t resultsize;
 211        char old_sha1_prefix[41];
 212        char new_sha1_prefix[41];
 213        struct patch *next;
 214
 215        /* three-way fallback result */
 216        struct object_id threeway_stage[3];
 217};
 218
 219static void free_fragment_list(struct fragment *list)
 220{
 221        while (list) {
 222                struct fragment *next = list->next;
 223                if (list->free_patch)
 224                        free((char *)list->patch);
 225                free(list);
 226                list = next;
 227        }
 228}
 229
 230static void free_patch(struct patch *patch)
 231{
 232        free_fragment_list(patch->fragments);
 233        free(patch->def_name);
 234        free(patch->old_name);
 235        free(patch->new_name);
 236        free(patch->result);
 237        free(patch);
 238}
 239
 240static void free_patch_list(struct patch *list)
 241{
 242        while (list) {
 243                struct patch *next = list->next;
 244                free_patch(list);
 245                list = next;
 246        }
 247}
 248
 249/*
 250 * A line in a file, len-bytes long (includes the terminating LF,
 251 * except for an incomplete line at the end if the file ends with
 252 * one), and its contents hashes to 'hash'.
 253 */
 254struct line {
 255        size_t len;
 256        unsigned hash : 24;
 257        unsigned flag : 8;
 258#define LINE_COMMON     1
 259#define LINE_PATCHED    2
 260};
 261
 262/*
 263 * This represents a "file", which is an array of "lines".
 264 */
 265struct image {
 266        char *buf;
 267        size_t len;
 268        size_t nr;
 269        size_t alloc;
 270        struct line *line_allocated;
 271        struct line *line;
 272};
 273
 274/*
 275 * Records filenames that have been touched, in order to handle
 276 * the case where more than one patches touch the same file.
 277 */
 278
 279static struct string_list fn_table;
 280
 281static uint32_t hash_line(const char *cp, size_t len)
 282{
 283        size_t i;
 284        uint32_t h;
 285        for (i = 0, h = 0; i < len; i++) {
 286                if (!isspace(cp[i])) {
 287                        h = h * 3 + (cp[i] & 0xff);
 288                }
 289        }
 290        return h;
 291}
 292
 293/*
 294 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 295 * whitespace difference. Returns 1 if they match, 0 otherwise
 296 */
 297static int fuzzy_matchlines(const char *s1, size_t n1,
 298                            const char *s2, size_t n2)
 299{
 300        const char *last1 = s1 + n1 - 1;
 301        const char *last2 = s2 + n2 - 1;
 302        int result = 0;
 303
 304        /* ignore line endings */
 305        while ((*last1 == '\r') || (*last1 == '\n'))
 306                last1--;
 307        while ((*last2 == '\r') || (*last2 == '\n'))
 308                last2--;
 309
 310        /* skip leading whitespaces, if both begin with whitespace */
 311        if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
 312                while (isspace(*s1) && (s1 <= last1))
 313                        s1++;
 314                while (isspace(*s2) && (s2 <= last2))
 315                        s2++;
 316        }
 317        /* early return if both lines are empty */
 318        if ((s1 > last1) && (s2 > last2))
 319                return 1;
 320        while (!result) {
 321                result = *s1++ - *s2++;
 322                /*
 323                 * Skip whitespace inside. We check for whitespace on
 324                 * both buffers because we don't want "a b" to match
 325                 * "ab"
 326                 */
 327                if (isspace(*s1) && isspace(*s2)) {
 328                        while (isspace(*s1) && s1 <= last1)
 329                                s1++;
 330                        while (isspace(*s2) && s2 <= last2)
 331                                s2++;
 332                }
 333                /*
 334                 * If we reached the end on one side only,
 335                 * lines don't match
 336                 */
 337                if (
 338                    ((s2 > last2) && (s1 <= last1)) ||
 339                    ((s1 > last1) && (s2 <= last2)))
 340                        return 0;
 341                if ((s1 > last1) && (s2 > last2))
 342                        break;
 343        }
 344
 345        return !result;
 346}
 347
 348static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 349{
 350        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 351        img->line_allocated[img->nr].len = len;
 352        img->line_allocated[img->nr].hash = hash_line(bol, len);
 353        img->line_allocated[img->nr].flag = flag;
 354        img->nr++;
 355}
 356
 357/*
 358 * "buf" has the file contents to be patched (read from various sources).
 359 * attach it to "image" and add line-based index to it.
 360 * "image" now owns the "buf".
 361 */
 362static void prepare_image(struct image *image, char *buf, size_t len,
 363                          int prepare_linetable)
 364{
 365        const char *cp, *ep;
 366
 367        memset(image, 0, sizeof(*image));
 368        image->buf = buf;
 369        image->len = len;
 370
 371        if (!prepare_linetable)
 372                return;
 373
 374        ep = image->buf + image->len;
 375        cp = image->buf;
 376        while (cp < ep) {
 377                const char *next;
 378                for (next = cp; next < ep && *next != '\n'; next++)
 379                        ;
 380                if (next < ep)
 381                        next++;
 382                add_line_info(image, cp, next - cp, 0);
 383                cp = next;
 384        }
 385        image->line = image->line_allocated;
 386}
 387
 388static void clear_image(struct image *image)
 389{
 390        free(image->buf);
 391        free(image->line_allocated);
 392        memset(image, 0, sizeof(*image));
 393}
 394
 395/* fmt must contain _one_ %s and no other substitution */
 396static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 397{
 398        struct strbuf sb = STRBUF_INIT;
 399
 400        if (patch->old_name && patch->new_name &&
 401            strcmp(patch->old_name, patch->new_name)) {
 402                quote_c_style(patch->old_name, &sb, NULL, 0);
 403                strbuf_addstr(&sb, " => ");
 404                quote_c_style(patch->new_name, &sb, NULL, 0);
 405        } else {
 406                const char *n = patch->new_name;
 407                if (!n)
 408                        n = patch->old_name;
 409                quote_c_style(n, &sb, NULL, 0);
 410        }
 411        fprintf(output, fmt, sb.buf);
 412        fputc('\n', output);
 413        strbuf_release(&sb);
 414}
 415
 416#define SLOP (16)
 417
 418static void read_patch_file(struct strbuf *sb, int fd)
 419{
 420        if (strbuf_read(sb, fd, 0) < 0)
 421                die_errno("git apply: failed to read");
 422
 423        /*
 424         * Make sure that we have some slop in the buffer
 425         * so that we can do speculative "memcmp" etc, and
 426         * see to it that it is NUL-filled.
 427         */
 428        strbuf_grow(sb, SLOP);
 429        memset(sb->buf + sb->len, 0, SLOP);
 430}
 431
 432static unsigned long linelen(const char *buffer, unsigned long size)
 433{
 434        unsigned long len = 0;
 435        while (size--) {
 436                len++;
 437                if (*buffer++ == '\n')
 438                        break;
 439        }
 440        return len;
 441}
 442
 443static int is_dev_null(const char *str)
 444{
 445        return skip_prefix(str, "/dev/null", &str) && isspace(*str);
 446}
 447
 448#define TERM_SPACE      1
 449#define TERM_TAB        2
 450
 451static int name_terminate(const char *name, int namelen, int c, int terminate)
 452{
 453        if (c == ' ' && !(terminate & TERM_SPACE))
 454                return 0;
 455        if (c == '\t' && !(terminate & TERM_TAB))
 456                return 0;
 457
 458        return 1;
 459}
 460
 461/* remove double slashes to make --index work with such filenames */
 462static char *squash_slash(char *name)
 463{
 464        int i = 0, j = 0;
 465
 466        if (!name)
 467                return NULL;
 468
 469        while (name[i]) {
 470                if ((name[j++] = name[i++]) == '/')
 471                        while (name[i] == '/')
 472                                i++;
 473        }
 474        name[j] = '\0';
 475        return name;
 476}
 477
 478static char *find_name_gnu(struct apply_state *state,
 479                           const char *line,
 480                           const char *def,
 481                           int p_value)
 482{
 483        struct strbuf name = STRBUF_INIT;
 484        char *cp;
 485
 486        /*
 487         * Proposed "new-style" GNU patch/diff format; see
 488         * http://marc.info/?l=git&m=112927316408690&w=2
 489         */
 490        if (unquote_c_style(&name, line, NULL)) {
 491                strbuf_release(&name);
 492                return NULL;
 493        }
 494
 495        for (cp = name.buf; p_value; p_value--) {
 496                cp = strchr(cp, '/');
 497                if (!cp) {
 498                        strbuf_release(&name);
 499                        return NULL;
 500                }
 501                cp++;
 502        }
 503
 504        strbuf_remove(&name, 0, cp - name.buf);
 505        if (state->root.len)
 506                strbuf_insert(&name, 0, state->root.buf, state->root.len);
 507        return squash_slash(strbuf_detach(&name, NULL));
 508}
 509
 510static size_t sane_tz_len(const char *line, size_t len)
 511{
 512        const char *tz, *p;
 513
 514        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 515                return 0;
 516        tz = line + len - strlen(" +0500");
 517
 518        if (tz[1] != '+' && tz[1] != '-')
 519                return 0;
 520
 521        for (p = tz + 2; p != line + len; p++)
 522                if (!isdigit(*p))
 523                        return 0;
 524
 525        return line + len - tz;
 526}
 527
 528static size_t tz_with_colon_len(const char *line, size_t len)
 529{
 530        const char *tz, *p;
 531
 532        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 533                return 0;
 534        tz = line + len - strlen(" +08:00");
 535
 536        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 537                return 0;
 538        p = tz + 2;
 539        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 540            !isdigit(*p++) || !isdigit(*p++))
 541                return 0;
 542
 543        return line + len - tz;
 544}
 545
 546static size_t date_len(const char *line, size_t len)
 547{
 548        const char *date, *p;
 549
 550        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 551                return 0;
 552        p = date = line + len - strlen("72-02-05");
 553
 554        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 555            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 556            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 557                return 0;
 558
 559        if (date - line >= strlen("19") &&
 560            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 561                date -= strlen("19");
 562
 563        return line + len - date;
 564}
 565
 566static size_t short_time_len(const char *line, size_t len)
 567{
 568        const char *time, *p;
 569
 570        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 571                return 0;
 572        p = time = line + len - strlen(" 07:01:32");
 573
 574        /* Permit 1-digit hours? */
 575        if (*p++ != ' ' ||
 576            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 577            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 578            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 579                return 0;
 580
 581        return line + len - time;
 582}
 583
 584static size_t fractional_time_len(const char *line, size_t len)
 585{
 586        const char *p;
 587        size_t n;
 588
 589        /* Expected format: 19:41:17.620000023 */
 590        if (!len || !isdigit(line[len - 1]))
 591                return 0;
 592        p = line + len - 1;
 593
 594        /* Fractional seconds. */
 595        while (p > line && isdigit(*p))
 596                p--;
 597        if (*p != '.')
 598                return 0;
 599
 600        /* Hours, minutes, and whole seconds. */
 601        n = short_time_len(line, p - line);
 602        if (!n)
 603                return 0;
 604
 605        return line + len - p + n;
 606}
 607
 608static size_t trailing_spaces_len(const char *line, size_t len)
 609{
 610        const char *p;
 611
 612        /* Expected format: ' ' x (1 or more)  */
 613        if (!len || line[len - 1] != ' ')
 614                return 0;
 615
 616        p = line + len;
 617        while (p != line) {
 618                p--;
 619                if (*p != ' ')
 620                        return line + len - (p + 1);
 621        }
 622
 623        /* All spaces! */
 624        return len;
 625}
 626
 627static size_t diff_timestamp_len(const char *line, size_t len)
 628{
 629        const char *end = line + len;
 630        size_t n;
 631
 632        /*
 633         * Posix: 2010-07-05 19:41:17
 634         * GNU: 2010-07-05 19:41:17.620000023 -0500
 635         */
 636
 637        if (!isdigit(end[-1]))
 638                return 0;
 639
 640        n = sane_tz_len(line, end - line);
 641        if (!n)
 642                n = tz_with_colon_len(line, end - line);
 643        end -= n;
 644
 645        n = short_time_len(line, end - line);
 646        if (!n)
 647                n = fractional_time_len(line, end - line);
 648        end -= n;
 649
 650        n = date_len(line, end - line);
 651        if (!n) /* No date.  Too bad. */
 652                return 0;
 653        end -= n;
 654
 655        if (end == line)        /* No space before date. */
 656                return 0;
 657        if (end[-1] == '\t') {  /* Success! */
 658                end--;
 659                return line + len - end;
 660        }
 661        if (end[-1] != ' ')     /* No space before date. */
 662                return 0;
 663
 664        /* Whitespace damage. */
 665        end -= trailing_spaces_len(line, end - line);
 666        return line + len - end;
 667}
 668
 669static char *find_name_common(struct apply_state *state,
 670                              const char *line,
 671                              const char *def,
 672                              int p_value,
 673                              const char *end,
 674                              int terminate)
 675{
 676        int len;
 677        const char *start = NULL;
 678
 679        if (p_value == 0)
 680                start = line;
 681        while (line != end) {
 682                char c = *line;
 683
 684                if (!end && isspace(c)) {
 685                        if (c == '\n')
 686                                break;
 687                        if (name_terminate(start, line-start, c, terminate))
 688                                break;
 689                }
 690                line++;
 691                if (c == '/' && !--p_value)
 692                        start = line;
 693        }
 694        if (!start)
 695                return squash_slash(xstrdup_or_null(def));
 696        len = line - start;
 697        if (!len)
 698                return squash_slash(xstrdup_or_null(def));
 699
 700        /*
 701         * Generally we prefer the shorter name, especially
 702         * if the other one is just a variation of that with
 703         * something else tacked on to the end (ie "file.orig"
 704         * or "file~").
 705         */
 706        if (def) {
 707                int deflen = strlen(def);
 708                if (deflen < len && !strncmp(start, def, deflen))
 709                        return squash_slash(xstrdup(def));
 710        }
 711
 712        if (state->root.len) {
 713                char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
 714                return squash_slash(ret);
 715        }
 716
 717        return squash_slash(xmemdupz(start, len));
 718}
 719
 720static char *find_name(struct apply_state *state,
 721                       const char *line,
 722                       char *def,
 723                       int p_value,
 724                       int terminate)
 725{
 726        if (*line == '"') {
 727                char *name = find_name_gnu(state, line, def, p_value);
 728                if (name)
 729                        return name;
 730        }
 731
 732        return find_name_common(state, line, def, p_value, NULL, terminate);
 733}
 734
 735static char *find_name_traditional(struct apply_state *state,
 736                                   const char *line,
 737                                   char *def,
 738                                   int p_value)
 739{
 740        size_t len;
 741        size_t date_len;
 742
 743        if (*line == '"') {
 744                char *name = find_name_gnu(state, line, def, p_value);
 745                if (name)
 746                        return name;
 747        }
 748
 749        len = strchrnul(line, '\n') - line;
 750        date_len = diff_timestamp_len(line, len);
 751        if (!date_len)
 752                return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
 753        len -= date_len;
 754
 755        return find_name_common(state, line, def, p_value, line + len, 0);
 756}
 757
 758static int count_slashes(const char *cp)
 759{
 760        int cnt = 0;
 761        char ch;
 762
 763        while ((ch = *cp++))
 764                if (ch == '/')
 765                        cnt++;
 766        return cnt;
 767}
 768
 769/*
 770 * Given the string after "--- " or "+++ ", guess the appropriate
 771 * p_value for the given patch.
 772 */
 773static int guess_p_value(struct apply_state *state, const char *nameline)
 774{
 775        char *name, *cp;
 776        int val = -1;
 777
 778        if (is_dev_null(nameline))
 779                return -1;
 780        name = find_name_traditional(state, nameline, NULL, 0);
 781        if (!name)
 782                return -1;
 783        cp = strchr(name, '/');
 784        if (!cp)
 785                val = 0;
 786        else if (state->prefix) {
 787                /*
 788                 * Does it begin with "a/$our-prefix" and such?  Then this is
 789                 * very likely to apply to our directory.
 790                 */
 791                if (!strncmp(name, state->prefix, state->prefix_length))
 792                        val = count_slashes(state->prefix);
 793                else {
 794                        cp++;
 795                        if (!strncmp(cp, state->prefix, state->prefix_length))
 796                                val = count_slashes(state->prefix) + 1;
 797                }
 798        }
 799        free(name);
 800        return val;
 801}
 802
 803/*
 804 * Does the ---/+++ line have the POSIX timestamp after the last HT?
 805 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 806 * this such a timestamp?
 807 */
 808static int has_epoch_timestamp(const char *nameline)
 809{
 810        /*
 811         * We are only interested in epoch timestamp; any non-zero
 812         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 813         * For the same reason, the date must be either 1969-12-31 or
 814         * 1970-01-01, and the seconds part must be "00".
 815         */
 816        const char stamp_regexp[] =
 817                "^(1969-12-31|1970-01-01)"
 818                " "
 819                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 820                " "
 821                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 822        const char *timestamp = NULL, *cp, *colon;
 823        static regex_t *stamp;
 824        regmatch_t m[10];
 825        int zoneoffset;
 826        int hourminute;
 827        int status;
 828
 829        for (cp = nameline; *cp != '\n'; cp++) {
 830                if (*cp == '\t')
 831                        timestamp = cp + 1;
 832        }
 833        if (!timestamp)
 834                return 0;
 835        if (!stamp) {
 836                stamp = xmalloc(sizeof(*stamp));
 837                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 838                        warning(_("Cannot prepare timestamp regexp %s"),
 839                                stamp_regexp);
 840                        return 0;
 841                }
 842        }
 843
 844        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 845        if (status) {
 846                if (status != REG_NOMATCH)
 847                        warning(_("regexec returned %d for input: %s"),
 848                                status, timestamp);
 849                return 0;
 850        }
 851
 852        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 853        if (*colon == ':')
 854                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 855        else
 856                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 857        if (timestamp[m[3].rm_so] == '-')
 858                zoneoffset = -zoneoffset;
 859
 860        /*
 861         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 862         * (west of GMT) or 1970-01-01 (east of GMT)
 863         */
 864        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 865            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 866                return 0;
 867
 868        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 869                      strtol(timestamp + 14, NULL, 10) -
 870                      zoneoffset);
 871
 872        return ((zoneoffset < 0 && hourminute == 1440) ||
 873                (0 <= zoneoffset && !hourminute));
 874}
 875
 876/*
 877 * Get the name etc info from the ---/+++ lines of a traditional patch header
 878 *
 879 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 880 * files, we can happily check the index for a match, but for creating a
 881 * new file we should try to match whatever "patch" does. I have no idea.
 882 */
 883static void parse_traditional_patch(struct apply_state *state,
 884                                    const char *first,
 885                                    const char *second,
 886                                    struct patch *patch)
 887{
 888        char *name;
 889
 890        first += 4;     /* skip "--- " */
 891        second += 4;    /* skip "+++ " */
 892        if (!state->p_value_known) {
 893                int p, q;
 894                p = guess_p_value(state, first);
 895                q = guess_p_value(state, second);
 896                if (p < 0) p = q;
 897                if (0 <= p && p == q) {
 898                        state->p_value = p;
 899                        state->p_value_known = 1;
 900                }
 901        }
 902        if (is_dev_null(first)) {
 903                patch->is_new = 1;
 904                patch->is_delete = 0;
 905                name = find_name_traditional(state, second, NULL, state->p_value);
 906                patch->new_name = name;
 907        } else if (is_dev_null(second)) {
 908                patch->is_new = 0;
 909                patch->is_delete = 1;
 910                name = find_name_traditional(state, first, NULL, state->p_value);
 911                patch->old_name = name;
 912        } else {
 913                char *first_name;
 914                first_name = find_name_traditional(state, first, NULL, state->p_value);
 915                name = find_name_traditional(state, second, first_name, state->p_value);
 916                free(first_name);
 917                if (has_epoch_timestamp(first)) {
 918                        patch->is_new = 1;
 919                        patch->is_delete = 0;
 920                        patch->new_name = name;
 921                } else if (has_epoch_timestamp(second)) {
 922                        patch->is_new = 0;
 923                        patch->is_delete = 1;
 924                        patch->old_name = name;
 925                } else {
 926                        patch->old_name = name;
 927                        patch->new_name = xstrdup_or_null(name);
 928                }
 929        }
 930        if (!name)
 931                die(_("unable to find filename in patch at line %d"), state_linenr);
 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 void 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;
 962        }
 963
 964        if (*name) {
 965                int len = strlen(*name);
 966                char *another;
 967                if (isnull)
 968                        die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
 969                            *name, state_linenr);
 970                another = find_name(state, line, NULL, state->p_value, TERM_TAB);
 971                if (!another || memcmp(another, *name, len + 1))
 972                        die((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                free(another);
 976        } else {
 977                /* expect "/dev/null" */
 978                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 979                        die(_("git apply: bad git-diff - expected /dev/null on line %d"), state_linenr);
 980        }
 981}
 982
 983static int gitdiff_oldname(struct apply_state *state,
 984                           const char *line,
 985                           struct patch *patch)
 986{
 987        gitdiff_verify_name(state, line,
 988                            patch->is_new, &patch->old_name,
 989                            DIFF_OLD_NAME);
 990        return 0;
 991}
 992
 993static int gitdiff_newname(struct apply_state *state,
 994                           const char *line,
 995                           struct patch *patch)
 996{
 997        gitdiff_verify_name(state, line,
 998                            patch->is_delete, &patch->new_name,
 999                            DIFF_NEW_NAME);
1000        return 0;
1001}
1002
1003static int gitdiff_oldmode(struct apply_state *state,
1004                           const char *line,
1005                           struct patch *patch)
1006{
1007        patch->old_mode = strtoul(line, NULL, 8);
1008        return 0;
1009}
1010
1011static int gitdiff_newmode(struct apply_state *state,
1012                           const char *line,
1013                           struct patch *patch)
1014{
1015        patch->new_mode = strtoul(line, NULL, 8);
1016        return 0;
1017}
1018
1019static int gitdiff_delete(struct apply_state *state,
1020                          const char *line,
1021                          struct patch *patch)
1022{
1023        patch->is_delete = 1;
1024        free(patch->old_name);
1025        patch->old_name = xstrdup_or_null(patch->def_name);
1026        return gitdiff_oldmode(state, line, patch);
1027}
1028
1029static int gitdiff_newfile(struct apply_state *state,
1030                           const char *line,
1031                           struct patch *patch)
1032{
1033        patch->is_new = 1;
1034        free(patch->new_name);
1035        patch->new_name = xstrdup_or_null(patch->def_name);
1036        return gitdiff_newmode(state, line, patch);
1037}
1038
1039static int gitdiff_copysrc(struct apply_state *state,
1040                           const char *line,
1041                           struct patch *patch)
1042{
1043        patch->is_copy = 1;
1044        free(patch->old_name);
1045        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1046        return 0;
1047}
1048
1049static int gitdiff_copydst(struct apply_state *state,
1050                           const char *line,
1051                           struct patch *patch)
1052{
1053        patch->is_copy = 1;
1054        free(patch->new_name);
1055        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1056        return 0;
1057}
1058
1059static int gitdiff_renamesrc(struct apply_state *state,
1060                             const char *line,
1061                             struct patch *patch)
1062{
1063        patch->is_rename = 1;
1064        free(patch->old_name);
1065        patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1066        return 0;
1067}
1068
1069static int gitdiff_renamedst(struct apply_state *state,
1070                             const char *line,
1071                             struct patch *patch)
1072{
1073        patch->is_rename = 1;
1074        free(patch->new_name);
1075        patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1076        return 0;
1077}
1078
1079static int gitdiff_similarity(struct apply_state *state,
1080                              const char *line,
1081                              struct patch *patch)
1082{
1083        unsigned long val = strtoul(line, NULL, 10);
1084        if (val <= 100)
1085                patch->score = val;
1086        return 0;
1087}
1088
1089static int gitdiff_dissimilarity(struct apply_state *state,
1090                                 const char *line,
1091                                 struct patch *patch)
1092{
1093        unsigned long val = strtoul(line, NULL, 10);
1094        if (val <= 100)
1095                patch->score = val;
1096        return 0;
1097}
1098
1099static int gitdiff_index(struct apply_state *state,
1100                         const char *line,
1101                         struct patch *patch)
1102{
1103        /*
1104         * index line is N hexadecimal, "..", N hexadecimal,
1105         * and optional space with octal mode.
1106         */
1107        const char *ptr, *eol;
1108        int len;
1109
1110        ptr = strchr(line, '.');
1111        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1112                return 0;
1113        len = ptr - line;
1114        memcpy(patch->old_sha1_prefix, line, len);
1115        patch->old_sha1_prefix[len] = 0;
1116
1117        line = ptr + 2;
1118        ptr = strchr(line, ' ');
1119        eol = strchrnul(line, '\n');
1120
1121        if (!ptr || eol < ptr)
1122                ptr = eol;
1123        len = ptr - line;
1124
1125        if (40 < len)
1126                return 0;
1127        memcpy(patch->new_sha1_prefix, line, len);
1128        patch->new_sha1_prefix[len] = 0;
1129        if (*ptr == ' ')
1130                patch->old_mode = strtoul(ptr+1, NULL, 8);
1131        return 0;
1132}
1133
1134/*
1135 * This is normal for a diff that doesn't change anything: we'll fall through
1136 * into the next diff. Tell the parser to break out.
1137 */
1138static int gitdiff_unrecognized(struct apply_state *state,
1139                                const char *line,
1140                                struct patch *patch)
1141{
1142        return -1;
1143}
1144
1145/*
1146 * Skip p_value leading components from "line"; as we do not accept
1147 * absolute paths, return NULL in that case.
1148 */
1149static const char *skip_tree_prefix(struct apply_state *state,
1150                                    const char *line,
1151                                    int llen)
1152{
1153        int nslash;
1154        int i;
1155
1156        if (!state->p_value)
1157                return (llen && line[0] == '/') ? NULL : line;
1158
1159        nslash = state->p_value;
1160        for (i = 0; i < llen; i++) {
1161                int ch = line[i];
1162                if (ch == '/' && --nslash <= 0)
1163                        return (i == 0) ? NULL : &line[i + 1];
1164        }
1165        return NULL;
1166}
1167
1168/*
1169 * This is to extract the same name that appears on "diff --git"
1170 * line.  We do not find and return anything if it is a rename
1171 * patch, and it is OK because we will find the name elsewhere.
1172 * We need to reliably find name only when it is mode-change only,
1173 * creation or deletion of an empty file.  In any of these cases,
1174 * both sides are the same name under a/ and b/ respectively.
1175 */
1176static char *git_header_name(struct apply_state *state,
1177                             const char *line,
1178                             int llen)
1179{
1180        const char *name;
1181        const char *second = NULL;
1182        size_t len, line_len;
1183
1184        line += strlen("diff --git ");
1185        llen -= strlen("diff --git ");
1186
1187        if (*line == '"') {
1188                const char *cp;
1189                struct strbuf first = STRBUF_INIT;
1190                struct strbuf sp = STRBUF_INIT;
1191
1192                if (unquote_c_style(&first, line, &second))
1193                        goto free_and_fail1;
1194
1195                /* strip the a/b prefix including trailing slash */
1196                cp = skip_tree_prefix(state, first.buf, first.len);
1197                if (!cp)
1198                        goto free_and_fail1;
1199                strbuf_remove(&first, 0, cp - first.buf);
1200
1201                /*
1202                 * second points at one past closing dq of name.
1203                 * find the second name.
1204                 */
1205                while ((second < line + llen) && isspace(*second))
1206                        second++;
1207
1208                if (line + llen <= second)
1209                        goto free_and_fail1;
1210                if (*second == '"') {
1211                        if (unquote_c_style(&sp, second, NULL))
1212                                goto free_and_fail1;
1213                        cp = skip_tree_prefix(state, sp.buf, sp.len);
1214                        if (!cp)
1215                                goto free_and_fail1;
1216                        /* They must match, otherwise ignore */
1217                        if (strcmp(cp, first.buf))
1218                                goto free_and_fail1;
1219                        strbuf_release(&sp);
1220                        return strbuf_detach(&first, NULL);
1221                }
1222
1223                /* unquoted second */
1224                cp = skip_tree_prefix(state, second, line + llen - second);
1225                if (!cp)
1226                        goto free_and_fail1;
1227                if (line + llen - cp != first.len ||
1228                    memcmp(first.buf, cp, first.len))
1229                        goto free_and_fail1;
1230                return strbuf_detach(&first, NULL);
1231
1232        free_and_fail1:
1233                strbuf_release(&first);
1234                strbuf_release(&sp);
1235                return NULL;
1236        }
1237
1238        /* unquoted first name */
1239        name = skip_tree_prefix(state, line, llen);
1240        if (!name)
1241                return NULL;
1242
1243        /*
1244         * since the first name is unquoted, a dq if exists must be
1245         * the beginning of the second name.
1246         */
1247        for (second = name; second < line + llen; second++) {
1248                if (*second == '"') {
1249                        struct strbuf sp = STRBUF_INIT;
1250                        const char *np;
1251
1252                        if (unquote_c_style(&sp, second, NULL))
1253                                goto free_and_fail2;
1254
1255                        np = skip_tree_prefix(state, sp.buf, sp.len);
1256                        if (!np)
1257                                goto free_and_fail2;
1258
1259                        len = sp.buf + sp.len - np;
1260                        if (len < second - name &&
1261                            !strncmp(np, name, len) &&
1262                            isspace(name[len])) {
1263                                /* Good */
1264                                strbuf_remove(&sp, 0, np - sp.buf);
1265                                return strbuf_detach(&sp, NULL);
1266                        }
1267
1268                free_and_fail2:
1269                        strbuf_release(&sp);
1270                        return NULL;
1271                }
1272        }
1273
1274        /*
1275         * Accept a name only if it shows up twice, exactly the same
1276         * form.
1277         */
1278        second = strchr(name, '\n');
1279        if (!second)
1280                return NULL;
1281        line_len = second - name;
1282        for (len = 0 ; ; len++) {
1283                switch (name[len]) {
1284                default:
1285                        continue;
1286                case '\n':
1287                        return NULL;
1288                case '\t': case ' ':
1289                        /*
1290                         * Is this the separator between the preimage
1291                         * and the postimage pathname?  Again, we are
1292                         * only interested in the case where there is
1293                         * no rename, as this is only to set def_name
1294                         * and a rename patch has the names elsewhere
1295                         * in an unambiguous form.
1296                         */
1297                        if (!name[len + 1])
1298                                return NULL; /* no postimage name */
1299                        second = skip_tree_prefix(state, name + len + 1,
1300                                                  line_len - (len + 1));
1301                        if (!second)
1302                                return NULL;
1303                        /*
1304                         * Does len bytes starting at "name" and "second"
1305                         * (that are separated by one HT or SP we just
1306                         * found) exactly match?
1307                         */
1308                        if (second[len] == '\n' && !strncmp(name, second, len))
1309                                return xmemdupz(name, len);
1310                }
1311        }
1312}
1313
1314/* Verify that we recognize the lines following a git header */
1315static int parse_git_header(struct apply_state *state,
1316                            const char *line,
1317                            int len,
1318                            unsigned int size,
1319                            struct patch *patch)
1320{
1321        unsigned long offset;
1322
1323        /* A git diff has explicit new/delete information, so we don't guess */
1324        patch->is_new = 0;
1325        patch->is_delete = 0;
1326
1327        /*
1328         * Some things may not have the old name in the
1329         * rest of the headers anywhere (pure mode changes,
1330         * or removing or adding empty files), so we get
1331         * the default name from the header.
1332         */
1333        patch->def_name = git_header_name(state, line, len);
1334        if (patch->def_name && state->root.len) {
1335                char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1336                free(patch->def_name);
1337                patch->def_name = s;
1338        }
1339
1340        line += len;
1341        size -= len;
1342        state_linenr++;
1343        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state_linenr++) {
1344                static const struct opentry {
1345                        const char *str;
1346                        int (*fn)(struct apply_state *, const char *, struct patch *);
1347                } optable[] = {
1348                        { "@@ -", gitdiff_hdrend },
1349                        { "--- ", gitdiff_oldname },
1350                        { "+++ ", gitdiff_newname },
1351                        { "old mode ", gitdiff_oldmode },
1352                        { "new mode ", gitdiff_newmode },
1353                        { "deleted file mode ", gitdiff_delete },
1354                        { "new file mode ", gitdiff_newfile },
1355                        { "copy from ", gitdiff_copysrc },
1356                        { "copy to ", gitdiff_copydst },
1357                        { "rename old ", gitdiff_renamesrc },
1358                        { "rename new ", gitdiff_renamedst },
1359                        { "rename from ", gitdiff_renamesrc },
1360                        { "rename to ", gitdiff_renamedst },
1361                        { "similarity index ", gitdiff_similarity },
1362                        { "dissimilarity index ", gitdiff_dissimilarity },
1363                        { "index ", gitdiff_index },
1364                        { "", gitdiff_unrecognized },
1365                };
1366                int i;
1367
1368                len = linelen(line, size);
1369                if (!len || line[len-1] != '\n')
1370                        break;
1371                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1372                        const struct opentry *p = optable + i;
1373                        int oplen = strlen(p->str);
1374                        if (len < oplen || memcmp(p->str, line, oplen))
1375                                continue;
1376                        if (p->fn(state, line + oplen, patch) < 0)
1377                                return offset;
1378                        break;
1379                }
1380        }
1381
1382        return offset;
1383}
1384
1385static int parse_num(const char *line, unsigned long *p)
1386{
1387        char *ptr;
1388
1389        if (!isdigit(*line))
1390                return 0;
1391        *p = strtoul(line, &ptr, 10);
1392        return ptr - line;
1393}
1394
1395static int parse_range(const char *line, int len, int offset, const char *expect,
1396                       unsigned long *p1, unsigned long *p2)
1397{
1398        int digits, ex;
1399
1400        if (offset < 0 || offset >= len)
1401                return -1;
1402        line += offset;
1403        len -= offset;
1404
1405        digits = parse_num(line, p1);
1406        if (!digits)
1407                return -1;
1408
1409        offset += digits;
1410        line += digits;
1411        len -= digits;
1412
1413        *p2 = 1;
1414        if (*line == ',') {
1415                digits = parse_num(line+1, p2);
1416                if (!digits)
1417                        return -1;
1418
1419                offset += digits+1;
1420                line += digits+1;
1421                len -= digits+1;
1422        }
1423
1424        ex = strlen(expect);
1425        if (ex > len)
1426                return -1;
1427        if (memcmp(line, expect, ex))
1428                return -1;
1429
1430        return offset + ex;
1431}
1432
1433static void recount_diff(const char *line, int size, struct fragment *fragment)
1434{
1435        int oldlines = 0, newlines = 0, ret = 0;
1436
1437        if (size < 1) {
1438                warning("recount: ignore empty hunk");
1439                return;
1440        }
1441
1442        for (;;) {
1443                int len = linelen(line, size);
1444                size -= len;
1445                line += len;
1446
1447                if (size < 1)
1448                        break;
1449
1450                switch (*line) {
1451                case ' ': case '\n':
1452                        newlines++;
1453                        /* fall through */
1454                case '-':
1455                        oldlines++;
1456                        continue;
1457                case '+':
1458                        newlines++;
1459                        continue;
1460                case '\\':
1461                        continue;
1462                case '@':
1463                        ret = size < 3 || !starts_with(line, "@@ ");
1464                        break;
1465                case 'd':
1466                        ret = size < 5 || !starts_with(line, "diff ");
1467                        break;
1468                default:
1469                        ret = -1;
1470                        break;
1471                }
1472                if (ret) {
1473                        warning(_("recount: unexpected line: %.*s"),
1474                                (int)linelen(line, size), line);
1475                        return;
1476                }
1477                break;
1478        }
1479        fragment->oldlines = oldlines;
1480        fragment->newlines = newlines;
1481}
1482
1483/*
1484 * Parse a unified diff fragment header of the
1485 * form "@@ -a,b +c,d @@"
1486 */
1487static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1488{
1489        int offset;
1490
1491        if (!len || line[len-1] != '\n')
1492                return -1;
1493
1494        /* Figure out the number of lines in a fragment */
1495        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1496        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1497
1498        return offset;
1499}
1500
1501static int find_header(struct apply_state *state,
1502                       const char *line,
1503                       unsigned long size,
1504                       int *hdrsize,
1505                       struct patch *patch)
1506{
1507        unsigned long offset, len;
1508
1509        patch->is_toplevel_relative = 0;
1510        patch->is_rename = patch->is_copy = 0;
1511        patch->is_new = patch->is_delete = -1;
1512        patch->old_mode = patch->new_mode = 0;
1513        patch->old_name = patch->new_name = NULL;
1514        for (offset = 0; size > 0; offset += len, size -= len, line += len, state_linenr++) {
1515                unsigned long nextlen;
1516
1517                len = linelen(line, size);
1518                if (!len)
1519                        break;
1520
1521                /* Testing this early allows us to take a few shortcuts.. */
1522                if (len < 6)
1523                        continue;
1524
1525                /*
1526                 * Make sure we don't find any unconnected patch fragments.
1527                 * That's a sign that we didn't find a header, and that a
1528                 * patch has become corrupted/broken up.
1529                 */
1530                if (!memcmp("@@ -", line, 4)) {
1531                        struct fragment dummy;
1532                        if (parse_fragment_header(line, len, &dummy) < 0)
1533                                continue;
1534                        die(_("patch fragment without header at line %d: %.*s"),
1535                            state_linenr, (int)len-1, line);
1536                }
1537
1538                if (size < len + 6)
1539                        break;
1540
1541                /*
1542                 * Git patch? It might not have a real patch, just a rename
1543                 * or mode change, so we handle that specially
1544                 */
1545                if (!memcmp("diff --git ", line, 11)) {
1546                        int git_hdr_len = parse_git_header(state, line, len, size, patch);
1547                        if (git_hdr_len <= len)
1548                                continue;
1549                        if (!patch->old_name && !patch->new_name) {
1550                                if (!patch->def_name)
1551                                        die(Q_("git diff header lacks filename information when removing "
1552                                               "%d leading pathname component (line %d)",
1553                                               "git diff header lacks filename information when removing "
1554                                               "%d leading pathname components (line %d)",
1555                                               state->p_value),
1556                                            state->p_value, state_linenr);
1557                                patch->old_name = xstrdup(patch->def_name);
1558                                patch->new_name = xstrdup(patch->def_name);
1559                        }
1560                        if (!patch->is_delete && !patch->new_name)
1561                                die("git diff header lacks filename information "
1562                                    "(line %d)", state_linenr);
1563                        patch->is_toplevel_relative = 1;
1564                        *hdrsize = git_hdr_len;
1565                        return offset;
1566                }
1567
1568                /* --- followed by +++ ? */
1569                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1570                        continue;
1571
1572                /*
1573                 * We only accept unified patches, so we want it to
1574                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1575                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1576                 */
1577                nextlen = linelen(line + len, size - len);
1578                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1579                        continue;
1580
1581                /* Ok, we'll consider it a patch */
1582                parse_traditional_patch(state, line, line+len, patch);
1583                *hdrsize = len + nextlen;
1584                state_linenr += 2;
1585                return offset;
1586        }
1587        return -1;
1588}
1589
1590static void record_ws_error(struct apply_state *state,
1591                            unsigned result,
1592                            const char *line,
1593                            int len,
1594                            int linenr)
1595{
1596        char *err;
1597
1598        if (!result)
1599                return;
1600
1601        state->whitespace_error++;
1602        if (state->squelch_whitespace_errors &&
1603            state->squelch_whitespace_errors < state->whitespace_error)
1604                return;
1605
1606        err = whitespace_error_string(result);
1607        fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1608                state->patch_input_file, linenr, err, len, line);
1609        free(err);
1610}
1611
1612static void check_whitespace(struct apply_state *state,
1613                             const char *line,
1614                             int len,
1615                             unsigned ws_rule)
1616{
1617        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1618
1619        record_ws_error(state, result, line + 1, len - 2, state_linenr);
1620}
1621
1622/*
1623 * Parse a unified diff. Note that this really needs to parse each
1624 * fragment separately, since the only way to know the difference
1625 * between a "---" that is part of a patch, and a "---" that starts
1626 * the next patch is to look at the line counts..
1627 */
1628static int parse_fragment(struct apply_state *state,
1629                          const char *line,
1630                          unsigned long size,
1631                          struct patch *patch,
1632                          struct fragment *fragment)
1633{
1634        int added, deleted;
1635        int len = linelen(line, size), offset;
1636        unsigned long oldlines, newlines;
1637        unsigned long leading, trailing;
1638
1639        offset = parse_fragment_header(line, len, fragment);
1640        if (offset < 0)
1641                return -1;
1642        if (offset > 0 && patch->recount)
1643                recount_diff(line + offset, size - offset, fragment);
1644        oldlines = fragment->oldlines;
1645        newlines = fragment->newlines;
1646        leading = 0;
1647        trailing = 0;
1648
1649        /* Parse the thing.. */
1650        line += len;
1651        size -= len;
1652        state_linenr++;
1653        added = deleted = 0;
1654        for (offset = len;
1655             0 < size;
1656             offset += len, size -= len, line += len, state_linenr++) {
1657                if (!oldlines && !newlines)
1658                        break;
1659                len = linelen(line, size);
1660                if (!len || line[len-1] != '\n')
1661                        return -1;
1662                switch (*line) {
1663                default:
1664                        return -1;
1665                case '\n': /* newer GNU diff, an empty context line */
1666                case ' ':
1667                        oldlines--;
1668                        newlines--;
1669                        if (!deleted && !added)
1670                                leading++;
1671                        trailing++;
1672                        if (!state->apply_in_reverse &&
1673                            ws_error_action == correct_ws_error)
1674                                check_whitespace(state, line, len, patch->ws_rule);
1675                        break;
1676                case '-':
1677                        if (state->apply_in_reverse &&
1678                            ws_error_action != nowarn_ws_error)
1679                                check_whitespace(state, line, len, patch->ws_rule);
1680                        deleted++;
1681                        oldlines--;
1682                        trailing = 0;
1683                        break;
1684                case '+':
1685                        if (!state->apply_in_reverse &&
1686                            ws_error_action != nowarn_ws_error)
1687                                check_whitespace(state, line, len, patch->ws_rule);
1688                        added++;
1689                        newlines--;
1690                        trailing = 0;
1691                        break;
1692
1693                /*
1694                 * We allow "\ No newline at end of file". Depending
1695                 * on locale settings when the patch was produced we
1696                 * don't know what this line looks like. The only
1697                 * thing we do know is that it begins with "\ ".
1698                 * Checking for 12 is just for sanity check -- any
1699                 * l10n of "\ No newline..." is at least that long.
1700                 */
1701                case '\\':
1702                        if (len < 12 || memcmp(line, "\\ ", 2))
1703                                return -1;
1704                        break;
1705                }
1706        }
1707        if (oldlines || newlines)
1708                return -1;
1709        if (!deleted && !added)
1710                return -1;
1711
1712        fragment->leading = leading;
1713        fragment->trailing = trailing;
1714
1715        /*
1716         * If a fragment ends with an incomplete line, we failed to include
1717         * it in the above loop because we hit oldlines == newlines == 0
1718         * before seeing it.
1719         */
1720        if (12 < size && !memcmp(line, "\\ ", 2))
1721                offset += linelen(line, size);
1722
1723        patch->lines_added += added;
1724        patch->lines_deleted += deleted;
1725
1726        if (0 < patch->is_new && oldlines)
1727                return error(_("new file depends on old contents"));
1728        if (0 < patch->is_delete && newlines)
1729                return error(_("deleted file still has contents"));
1730        return offset;
1731}
1732
1733/*
1734 * We have seen "diff --git a/... b/..." header (or a traditional patch
1735 * header).  Read hunks that belong to this patch into fragments and hang
1736 * them to the given patch structure.
1737 *
1738 * The (fragment->patch, fragment->size) pair points into the memory given
1739 * by the caller, not a copy, when we return.
1740 */
1741static int parse_single_patch(struct apply_state *state,
1742                              const char *line,
1743                              unsigned long size,
1744                              struct patch *patch)
1745{
1746        unsigned long offset = 0;
1747        unsigned long oldlines = 0, newlines = 0, context = 0;
1748        struct fragment **fragp = &patch->fragments;
1749
1750        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1751                struct fragment *fragment;
1752                int len;
1753
1754                fragment = xcalloc(1, sizeof(*fragment));
1755                fragment->linenr = state_linenr;
1756                len = parse_fragment(state, line, size, patch, fragment);
1757                if (len <= 0)
1758                        die(_("corrupt patch at line %d"), state_linenr);
1759                fragment->patch = line;
1760                fragment->size = len;
1761                oldlines += fragment->oldlines;
1762                newlines += fragment->newlines;
1763                context += fragment->leading + fragment->trailing;
1764
1765                *fragp = fragment;
1766                fragp = &fragment->next;
1767
1768                offset += len;
1769                line += len;
1770                size -= len;
1771        }
1772
1773        /*
1774         * If something was removed (i.e. we have old-lines) it cannot
1775         * be creation, and if something was added it cannot be
1776         * deletion.  However, the reverse is not true; --unified=0
1777         * patches that only add are not necessarily creation even
1778         * though they do not have any old lines, and ones that only
1779         * delete are not necessarily deletion.
1780         *
1781         * Unfortunately, a real creation/deletion patch do _not_ have
1782         * any context line by definition, so we cannot safely tell it
1783         * apart with --unified=0 insanity.  At least if the patch has
1784         * more than one hunk it is not creation or deletion.
1785         */
1786        if (patch->is_new < 0 &&
1787            (oldlines || (patch->fragments && patch->fragments->next)))
1788                patch->is_new = 0;
1789        if (patch->is_delete < 0 &&
1790            (newlines || (patch->fragments && patch->fragments->next)))
1791                patch->is_delete = 0;
1792
1793        if (0 < patch->is_new && oldlines)
1794                die(_("new file %s depends on old contents"), patch->new_name);
1795        if (0 < patch->is_delete && newlines)
1796                die(_("deleted file %s still has contents"), patch->old_name);
1797        if (!patch->is_delete && !newlines && context)
1798                fprintf_ln(stderr,
1799                           _("** warning: "
1800                             "file %s becomes empty but is not deleted"),
1801                           patch->new_name);
1802
1803        return offset;
1804}
1805
1806static inline int metadata_changes(struct patch *patch)
1807{
1808        return  patch->is_rename > 0 ||
1809                patch->is_copy > 0 ||
1810                patch->is_new > 0 ||
1811                patch->is_delete ||
1812                (patch->old_mode && patch->new_mode &&
1813                 patch->old_mode != patch->new_mode);
1814}
1815
1816static char *inflate_it(const void *data, unsigned long size,
1817                        unsigned long inflated_size)
1818{
1819        git_zstream stream;
1820        void *out;
1821        int st;
1822
1823        memset(&stream, 0, sizeof(stream));
1824
1825        stream.next_in = (unsigned char *)data;
1826        stream.avail_in = size;
1827        stream.next_out = out = xmalloc(inflated_size);
1828        stream.avail_out = inflated_size;
1829        git_inflate_init(&stream);
1830        st = git_inflate(&stream, Z_FINISH);
1831        git_inflate_end(&stream);
1832        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1833                free(out);
1834                return NULL;
1835        }
1836        return out;
1837}
1838
1839/*
1840 * Read a binary hunk and return a new fragment; fragment->patch
1841 * points at an allocated memory that the caller must free, so
1842 * it is marked as "->free_patch = 1".
1843 */
1844static struct fragment *parse_binary_hunk(char **buf_p,
1845                                          unsigned long *sz_p,
1846                                          int *status_p,
1847                                          int *used_p)
1848{
1849        /*
1850         * Expect a line that begins with binary patch method ("literal"
1851         * or "delta"), followed by the length of data before deflating.
1852         * a sequence of 'length-byte' followed by base-85 encoded data
1853         * should follow, terminated by a newline.
1854         *
1855         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1856         * and we would limit the patch line to 66 characters,
1857         * so one line can fit up to 13 groups that would decode
1858         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1859         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1860         */
1861        int llen, used;
1862        unsigned long size = *sz_p;
1863        char *buffer = *buf_p;
1864        int patch_method;
1865        unsigned long origlen;
1866        char *data = NULL;
1867        int hunk_size = 0;
1868        struct fragment *frag;
1869
1870        llen = linelen(buffer, size);
1871        used = llen;
1872
1873        *status_p = 0;
1874
1875        if (starts_with(buffer, "delta ")) {
1876                patch_method = BINARY_DELTA_DEFLATED;
1877                origlen = strtoul(buffer + 6, NULL, 10);
1878        }
1879        else if (starts_with(buffer, "literal ")) {
1880                patch_method = BINARY_LITERAL_DEFLATED;
1881                origlen = strtoul(buffer + 8, NULL, 10);
1882        }
1883        else
1884                return NULL;
1885
1886        state_linenr++;
1887        buffer += llen;
1888        while (1) {
1889                int byte_length, max_byte_length, newsize;
1890                llen = linelen(buffer, size);
1891                used += llen;
1892                state_linenr++;
1893                if (llen == 1) {
1894                        /* consume the blank line */
1895                        buffer++;
1896                        size--;
1897                        break;
1898                }
1899                /*
1900                 * Minimum line is "A00000\n" which is 7-byte long,
1901                 * and the line length must be multiple of 5 plus 2.
1902                 */
1903                if ((llen < 7) || (llen-2) % 5)
1904                        goto corrupt;
1905                max_byte_length = (llen - 2) / 5 * 4;
1906                byte_length = *buffer;
1907                if ('A' <= byte_length && byte_length <= 'Z')
1908                        byte_length = byte_length - 'A' + 1;
1909                else if ('a' <= byte_length && byte_length <= 'z')
1910                        byte_length = byte_length - 'a' + 27;
1911                else
1912                        goto corrupt;
1913                /* if the input length was not multiple of 4, we would
1914                 * have filler at the end but the filler should never
1915                 * exceed 3 bytes
1916                 */
1917                if (max_byte_length < byte_length ||
1918                    byte_length <= max_byte_length - 4)
1919                        goto corrupt;
1920                newsize = hunk_size + byte_length;
1921                data = xrealloc(data, newsize);
1922                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1923                        goto corrupt;
1924                hunk_size = newsize;
1925                buffer += llen;
1926                size -= llen;
1927        }
1928
1929        frag = xcalloc(1, sizeof(*frag));
1930        frag->patch = inflate_it(data, hunk_size, origlen);
1931        frag->free_patch = 1;
1932        if (!frag->patch)
1933                goto corrupt;
1934        free(data);
1935        frag->size = origlen;
1936        *buf_p = buffer;
1937        *sz_p = size;
1938        *used_p = used;
1939        frag->binary_patch_method = patch_method;
1940        return frag;
1941
1942 corrupt:
1943        free(data);
1944        *status_p = -1;
1945        error(_("corrupt binary patch at line %d: %.*s"),
1946              state_linenr-1, llen-1, buffer);
1947        return NULL;
1948}
1949
1950/*
1951 * Returns:
1952 *   -1 in case of error,
1953 *   the length of the parsed binary patch otherwise
1954 */
1955static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1956{
1957        /*
1958         * We have read "GIT binary patch\n"; what follows is a line
1959         * that says the patch method (currently, either "literal" or
1960         * "delta") and the length of data before deflating; a
1961         * sequence of 'length-byte' followed by base-85 encoded data
1962         * follows.
1963         *
1964         * When a binary patch is reversible, there is another binary
1965         * hunk in the same format, starting with patch method (either
1966         * "literal" or "delta") with the length of data, and a sequence
1967         * of length-byte + base-85 encoded data, terminated with another
1968         * empty line.  This data, when applied to the postimage, produces
1969         * the preimage.
1970         */
1971        struct fragment *forward;
1972        struct fragment *reverse;
1973        int status;
1974        int used, used_1;
1975
1976        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1977        if (!forward && !status)
1978                /* there has to be one hunk (forward hunk) */
1979                return error(_("unrecognized binary patch at line %d"), state_linenr-1);
1980        if (status)
1981                /* otherwise we already gave an error message */
1982                return status;
1983
1984        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1985        if (reverse)
1986                used += used_1;
1987        else if (status) {
1988                /*
1989                 * Not having reverse hunk is not an error, but having
1990                 * a corrupt reverse hunk is.
1991                 */
1992                free((void*) forward->patch);
1993                free(forward);
1994                return status;
1995        }
1996        forward->next = reverse;
1997        patch->fragments = forward;
1998        patch->is_binary = 1;
1999        return used;
2000}
2001
2002static void prefix_one(struct apply_state *state, char **name)
2003{
2004        char *old_name = *name;
2005        if (!old_name)
2006                return;
2007        *name = xstrdup(prefix_filename(state->prefix, state->prefix_length, *name));
2008        free(old_name);
2009}
2010
2011static void prefix_patch(struct apply_state *state, struct patch *p)
2012{
2013        if (!state->prefix || p->is_toplevel_relative)
2014                return;
2015        prefix_one(state, &p->new_name);
2016        prefix_one(state, &p->old_name);
2017}
2018
2019/*
2020 * include/exclude
2021 */
2022
2023static void add_name_limit(struct apply_state *state,
2024                           const char *name,
2025                           int exclude)
2026{
2027        struct string_list_item *it;
2028
2029        it = string_list_append(&state->limit_by_name, name);
2030        it->util = exclude ? NULL : (void *) 1;
2031}
2032
2033static int use_patch(struct apply_state *state, struct patch *p)
2034{
2035        const char *pathname = p->new_name ? p->new_name : p->old_name;
2036        int i;
2037
2038        /* Paths outside are not touched regardless of "--include" */
2039        if (0 < state->prefix_length) {
2040                int pathlen = strlen(pathname);
2041                if (pathlen <= state->prefix_length ||
2042                    memcmp(state->prefix, pathname, state->prefix_length))
2043                        return 0;
2044        }
2045
2046        /* See if it matches any of exclude/include rule */
2047        for (i = 0; i < state->limit_by_name.nr; i++) {
2048                struct string_list_item *it = &state->limit_by_name.items[i];
2049                if (!wildmatch(it->string, pathname, 0, NULL))
2050                        return (it->util != NULL);
2051        }
2052
2053        /*
2054         * If we had any include, a path that does not match any rule is
2055         * not used.  Otherwise, we saw bunch of exclude rules (or none)
2056         * and such a path is used.
2057         */
2058        return !state->has_include;
2059}
2060
2061
2062/*
2063 * Read the patch text in "buffer" that extends for "size" bytes; stop
2064 * reading after seeing a single patch (i.e. changes to a single file).
2065 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2066 * Return the number of bytes consumed, so that the caller can call us
2067 * again for the next patch.
2068 */
2069static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2070{
2071        int hdrsize, patchsize;
2072        int offset = find_header(state, buffer, size, &hdrsize, patch);
2073
2074        if (offset < 0)
2075                return offset;
2076
2077        prefix_patch(state, patch);
2078
2079        if (!use_patch(state, patch))
2080                patch->ws_rule = 0;
2081        else
2082                patch->ws_rule = whitespace_rule(patch->new_name
2083                                                 ? patch->new_name
2084                                                 : patch->old_name);
2085
2086        patchsize = parse_single_patch(state,
2087                                       buffer + offset + hdrsize,
2088                                       size - offset - hdrsize,
2089                                       patch);
2090
2091        if (!patchsize) {
2092                static const char git_binary[] = "GIT binary patch\n";
2093                int hd = hdrsize + offset;
2094                unsigned long llen = linelen(buffer + hd, size - hd);
2095
2096                if (llen == sizeof(git_binary) - 1 &&
2097                    !memcmp(git_binary, buffer + hd, llen)) {
2098                        int used;
2099                        state_linenr++;
2100                        used = parse_binary(buffer + hd + llen,
2101                                            size - hd - llen, patch);
2102                        if (used < 0)
2103                                return -1;
2104                        if (used)
2105                                patchsize = used + llen;
2106                        else
2107                                patchsize = 0;
2108                }
2109                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2110                        static const char *binhdr[] = {
2111                                "Binary files ",
2112                                "Files ",
2113                                NULL,
2114                        };
2115                        int i;
2116                        for (i = 0; binhdr[i]; i++) {
2117                                int len = strlen(binhdr[i]);
2118                                if (len < size - hd &&
2119                                    !memcmp(binhdr[i], buffer + hd, len)) {
2120                                        state_linenr++;
2121                                        patch->is_binary = 1;
2122                                        patchsize = llen;
2123                                        break;
2124                                }
2125                        }
2126                }
2127
2128                /* Empty patch cannot be applied if it is a text patch
2129                 * without metadata change.  A binary patch appears
2130                 * empty to us here.
2131                 */
2132                if ((state->apply || state->check) &&
2133                    (!patch->is_binary && !metadata_changes(patch)))
2134                        die(_("patch with only garbage at line %d"), state_linenr);
2135        }
2136
2137        return offset + hdrsize + patchsize;
2138}
2139
2140#define swap(a,b) myswap((a),(b),sizeof(a))
2141
2142#define myswap(a, b, size) do {         \
2143        unsigned char mytmp[size];      \
2144        memcpy(mytmp, &a, size);                \
2145        memcpy(&a, &b, size);           \
2146        memcpy(&b, mytmp, size);                \
2147} while (0)
2148
2149static void reverse_patches(struct patch *p)
2150{
2151        for (; p; p = p->next) {
2152                struct fragment *frag = p->fragments;
2153
2154                swap(p->new_name, p->old_name);
2155                swap(p->new_mode, p->old_mode);
2156                swap(p->is_new, p->is_delete);
2157                swap(p->lines_added, p->lines_deleted);
2158                swap(p->old_sha1_prefix, p->new_sha1_prefix);
2159
2160                for (; frag; frag = frag->next) {
2161                        swap(frag->newpos, frag->oldpos);
2162                        swap(frag->newlines, frag->oldlines);
2163                }
2164        }
2165}
2166
2167static const char pluses[] =
2168"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2169static const char minuses[]=
2170"----------------------------------------------------------------------";
2171
2172static void show_stats(struct patch *patch)
2173{
2174        struct strbuf qname = STRBUF_INIT;
2175        char *cp = patch->new_name ? patch->new_name : patch->old_name;
2176        int max, add, del;
2177
2178        quote_c_style(cp, &qname, NULL, 0);
2179
2180        /*
2181         * "scale" the filename
2182         */
2183        max = max_len;
2184        if (max > 50)
2185                max = 50;
2186
2187        if (qname.len > max) {
2188                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2189                if (!cp)
2190                        cp = qname.buf + qname.len + 3 - max;
2191                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2192        }
2193
2194        if (patch->is_binary) {
2195                printf(" %-*s |  Bin\n", max, qname.buf);
2196                strbuf_release(&qname);
2197                return;
2198        }
2199
2200        printf(" %-*s |", max, qname.buf);
2201        strbuf_release(&qname);
2202
2203        /*
2204         * scale the add/delete
2205         */
2206        max = max + max_change > 70 ? 70 - max : max_change;
2207        add = patch->lines_added;
2208        del = patch->lines_deleted;
2209
2210        if (max_change > 0) {
2211                int total = ((add + del) * max + max_change / 2) / max_change;
2212                add = (add * max + max_change / 2) / max_change;
2213                del = total - add;
2214        }
2215        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2216                add, pluses, del, minuses);
2217}
2218
2219static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2220{
2221        switch (st->st_mode & S_IFMT) {
2222        case S_IFLNK:
2223                if (strbuf_readlink(buf, path, st->st_size) < 0)
2224                        return error(_("unable to read symlink %s"), path);
2225                return 0;
2226        case S_IFREG:
2227                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2228                        return error(_("unable to open or read %s"), path);
2229                convert_to_git(path, buf->buf, buf->len, buf, 0);
2230                return 0;
2231        default:
2232                return -1;
2233        }
2234}
2235
2236/*
2237 * Update the preimage, and the common lines in postimage,
2238 * from buffer buf of length len. If postlen is 0 the postimage
2239 * is updated in place, otherwise it's updated on a new buffer
2240 * of length postlen
2241 */
2242
2243static void update_pre_post_images(struct image *preimage,
2244                                   struct image *postimage,
2245                                   char *buf,
2246                                   size_t len, size_t postlen)
2247{
2248        int i, ctx, reduced;
2249        char *new, *old, *fixed;
2250        struct image fixed_preimage;
2251
2252        /*
2253         * Update the preimage with whitespace fixes.  Note that we
2254         * are not losing preimage->buf -- apply_one_fragment() will
2255         * free "oldlines".
2256         */
2257        prepare_image(&fixed_preimage, buf, len, 1);
2258        assert(postlen
2259               ? fixed_preimage.nr == preimage->nr
2260               : fixed_preimage.nr <= preimage->nr);
2261        for (i = 0; i < fixed_preimage.nr; i++)
2262                fixed_preimage.line[i].flag = preimage->line[i].flag;
2263        free(preimage->line_allocated);
2264        *preimage = fixed_preimage;
2265
2266        /*
2267         * Adjust the common context lines in postimage. This can be
2268         * done in-place when we are shrinking it with whitespace
2269         * fixing, but needs a new buffer when ignoring whitespace or
2270         * expanding leading tabs to spaces.
2271         *
2272         * We trust the caller to tell us if the update can be done
2273         * in place (postlen==0) or not.
2274         */
2275        old = postimage->buf;
2276        if (postlen)
2277                new = postimage->buf = xmalloc(postlen);
2278        else
2279                new = old;
2280        fixed = preimage->buf;
2281
2282        for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2283                size_t l_len = postimage->line[i].len;
2284                if (!(postimage->line[i].flag & LINE_COMMON)) {
2285                        /* an added line -- no counterparts in preimage */
2286                        memmove(new, old, l_len);
2287                        old += l_len;
2288                        new += l_len;
2289                        continue;
2290                }
2291
2292                /* a common context -- skip it in the original postimage */
2293                old += l_len;
2294
2295                /* and find the corresponding one in the fixed preimage */
2296                while (ctx < preimage->nr &&
2297                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2298                        fixed += preimage->line[ctx].len;
2299                        ctx++;
2300                }
2301
2302                /*
2303                 * preimage is expected to run out, if the caller
2304                 * fixed addition of trailing blank lines.
2305                 */
2306                if (preimage->nr <= ctx) {
2307                        reduced++;
2308                        continue;
2309                }
2310
2311                /* and copy it in, while fixing the line length */
2312                l_len = preimage->line[ctx].len;
2313                memcpy(new, fixed, l_len);
2314                new += l_len;
2315                fixed += l_len;
2316                postimage->line[i].len = l_len;
2317                ctx++;
2318        }
2319
2320        if (postlen
2321            ? postlen < new - postimage->buf
2322            : postimage->len < new - postimage->buf)
2323                die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2324                    (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2325
2326        /* Fix the length of the whole thing */
2327        postimage->len = new - postimage->buf;
2328        postimage->nr -= reduced;
2329}
2330
2331static int line_by_line_fuzzy_match(struct image *img,
2332                                    struct image *preimage,
2333                                    struct image *postimage,
2334                                    unsigned long try,
2335                                    int try_lno,
2336                                    int preimage_limit)
2337{
2338        int i;
2339        size_t imgoff = 0;
2340        size_t preoff = 0;
2341        size_t postlen = postimage->len;
2342        size_t extra_chars;
2343        char *buf;
2344        char *preimage_eof;
2345        char *preimage_end;
2346        struct strbuf fixed;
2347        char *fixed_buf;
2348        size_t fixed_len;
2349
2350        for (i = 0; i < preimage_limit; i++) {
2351                size_t prelen = preimage->line[i].len;
2352                size_t imglen = img->line[try_lno+i].len;
2353
2354                if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2355                                      preimage->buf + preoff, prelen))
2356                        return 0;
2357                if (preimage->line[i].flag & LINE_COMMON)
2358                        postlen += imglen - prelen;
2359                imgoff += imglen;
2360                preoff += prelen;
2361        }
2362
2363        /*
2364         * Ok, the preimage matches with whitespace fuzz.
2365         *
2366         * imgoff now holds the true length of the target that
2367         * matches the preimage before the end of the file.
2368         *
2369         * Count the number of characters in the preimage that fall
2370         * beyond the end of the file and make sure that all of them
2371         * are whitespace characters. (This can only happen if
2372         * we are removing blank lines at the end of the file.)
2373         */
2374        buf = preimage_eof = preimage->buf + preoff;
2375        for ( ; i < preimage->nr; i++)
2376                preoff += preimage->line[i].len;
2377        preimage_end = preimage->buf + preoff;
2378        for ( ; buf < preimage_end; buf++)
2379                if (!isspace(*buf))
2380                        return 0;
2381
2382        /*
2383         * Update the preimage and the common postimage context
2384         * lines to use the same whitespace as the target.
2385         * If whitespace is missing in the target (i.e.
2386         * if the preimage extends beyond the end of the file),
2387         * use the whitespace from the preimage.
2388         */
2389        extra_chars = preimage_end - preimage_eof;
2390        strbuf_init(&fixed, imgoff + extra_chars);
2391        strbuf_add(&fixed, img->buf + try, imgoff);
2392        strbuf_add(&fixed, preimage_eof, extra_chars);
2393        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2394        update_pre_post_images(preimage, postimage,
2395                               fixed_buf, fixed_len, postlen);
2396        return 1;
2397}
2398
2399static int match_fragment(struct image *img,
2400                          struct image *preimage,
2401                          struct image *postimage,
2402                          unsigned long try,
2403                          int try_lno,
2404                          unsigned ws_rule,
2405                          int match_beginning, int match_end)
2406{
2407        int i;
2408        char *fixed_buf, *buf, *orig, *target;
2409        struct strbuf fixed;
2410        size_t fixed_len, postlen;
2411        int preimage_limit;
2412
2413        if (preimage->nr + try_lno <= img->nr) {
2414                /*
2415                 * The hunk falls within the boundaries of img.
2416                 */
2417                preimage_limit = preimage->nr;
2418                if (match_end && (preimage->nr + try_lno != img->nr))
2419                        return 0;
2420        } else if (ws_error_action == correct_ws_error &&
2421                   (ws_rule & WS_BLANK_AT_EOF)) {
2422                /*
2423                 * This hunk extends beyond the end of img, and we are
2424                 * removing blank lines at the end of the file.  This
2425                 * many lines from the beginning of the preimage must
2426                 * match with img, and the remainder of the preimage
2427                 * must be blank.
2428                 */
2429                preimage_limit = img->nr - try_lno;
2430        } else {
2431                /*
2432                 * The hunk extends beyond the end of the img and
2433                 * we are not removing blanks at the end, so we
2434                 * should reject the hunk at this position.
2435                 */
2436                return 0;
2437        }
2438
2439        if (match_beginning && try_lno)
2440                return 0;
2441
2442        /* Quick hash check */
2443        for (i = 0; i < preimage_limit; i++)
2444                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2445                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2446                        return 0;
2447
2448        if (preimage_limit == preimage->nr) {
2449                /*
2450                 * Do we have an exact match?  If we were told to match
2451                 * at the end, size must be exactly at try+fragsize,
2452                 * otherwise try+fragsize must be still within the preimage,
2453                 * and either case, the old piece should match the preimage
2454                 * exactly.
2455                 */
2456                if ((match_end
2457                     ? (try + preimage->len == img->len)
2458                     : (try + preimage->len <= img->len)) &&
2459                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2460                        return 1;
2461        } else {
2462                /*
2463                 * The preimage extends beyond the end of img, so
2464                 * there cannot be an exact match.
2465                 *
2466                 * There must be one non-blank context line that match
2467                 * a line before the end of img.
2468                 */
2469                char *buf_end;
2470
2471                buf = preimage->buf;
2472                buf_end = buf;
2473                for (i = 0; i < preimage_limit; i++)
2474                        buf_end += preimage->line[i].len;
2475
2476                for ( ; buf < buf_end; buf++)
2477                        if (!isspace(*buf))
2478                                break;
2479                if (buf == buf_end)
2480                        return 0;
2481        }
2482
2483        /*
2484         * No exact match. If we are ignoring whitespace, run a line-by-line
2485         * fuzzy matching. We collect all the line length information because
2486         * we need it to adjust whitespace if we match.
2487         */
2488        if (ws_ignore_action == ignore_ws_change)
2489                return line_by_line_fuzzy_match(img, preimage, postimage,
2490                                                try, try_lno, preimage_limit);
2491
2492        if (ws_error_action != correct_ws_error)
2493                return 0;
2494
2495        /*
2496         * The hunk does not apply byte-by-byte, but the hash says
2497         * it might with whitespace fuzz. We weren't asked to
2498         * ignore whitespace, we were asked to correct whitespace
2499         * errors, so let's try matching after whitespace correction.
2500         *
2501         * While checking the preimage against the target, whitespace
2502         * errors in both fixed, we count how large the corresponding
2503         * postimage needs to be.  The postimage prepared by
2504         * apply_one_fragment() has whitespace errors fixed on added
2505         * lines already, but the common lines were propagated as-is,
2506         * which may become longer when their whitespace errors are
2507         * fixed.
2508         */
2509
2510        /* First count added lines in postimage */
2511        postlen = 0;
2512        for (i = 0; i < postimage->nr; i++) {
2513                if (!(postimage->line[i].flag & LINE_COMMON))
2514                        postlen += postimage->line[i].len;
2515        }
2516
2517        /*
2518         * The preimage may extend beyond the end of the file,
2519         * but in this loop we will only handle the part of the
2520         * preimage that falls within the file.
2521         */
2522        strbuf_init(&fixed, preimage->len + 1);
2523        orig = preimage->buf;
2524        target = img->buf + try;
2525        for (i = 0; i < preimage_limit; i++) {
2526                size_t oldlen = preimage->line[i].len;
2527                size_t tgtlen = img->line[try_lno + i].len;
2528                size_t fixstart = fixed.len;
2529                struct strbuf tgtfix;
2530                int match;
2531
2532                /* Try fixing the line in the preimage */
2533                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2534
2535                /* Try fixing the line in the target */
2536                strbuf_init(&tgtfix, tgtlen);
2537                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2538
2539                /*
2540                 * If they match, either the preimage was based on
2541                 * a version before our tree fixed whitespace breakage,
2542                 * or we are lacking a whitespace-fix patch the tree
2543                 * the preimage was based on already had (i.e. target
2544                 * has whitespace breakage, the preimage doesn't).
2545                 * In either case, we are fixing the whitespace breakages
2546                 * so we might as well take the fix together with their
2547                 * real change.
2548                 */
2549                match = (tgtfix.len == fixed.len - fixstart &&
2550                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2551                                             fixed.len - fixstart));
2552
2553                /* Add the length if this is common with the postimage */
2554                if (preimage->line[i].flag & LINE_COMMON)
2555                        postlen += tgtfix.len;
2556
2557                strbuf_release(&tgtfix);
2558                if (!match)
2559                        goto unmatch_exit;
2560
2561                orig += oldlen;
2562                target += tgtlen;
2563        }
2564
2565
2566        /*
2567         * Now handle the lines in the preimage that falls beyond the
2568         * end of the file (if any). They will only match if they are
2569         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2570         * false).
2571         */
2572        for ( ; i < preimage->nr; i++) {
2573                size_t fixstart = fixed.len; /* start of the fixed preimage */
2574                size_t oldlen = preimage->line[i].len;
2575                int j;
2576
2577                /* Try fixing the line in the preimage */
2578                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2579
2580                for (j = fixstart; j < fixed.len; j++)
2581                        if (!isspace(fixed.buf[j]))
2582                                goto unmatch_exit;
2583
2584                orig += oldlen;
2585        }
2586
2587        /*
2588         * Yes, the preimage is based on an older version that still
2589         * has whitespace breakages unfixed, and fixing them makes the
2590         * hunk match.  Update the context lines in the postimage.
2591         */
2592        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2593        if (postlen < postimage->len)
2594                postlen = 0;
2595        update_pre_post_images(preimage, postimage,
2596                               fixed_buf, fixed_len, postlen);
2597        return 1;
2598
2599 unmatch_exit:
2600        strbuf_release(&fixed);
2601        return 0;
2602}
2603
2604static int find_pos(struct image *img,
2605                    struct image *preimage,
2606                    struct image *postimage,
2607                    int line,
2608                    unsigned ws_rule,
2609                    int match_beginning, int match_end)
2610{
2611        int i;
2612        unsigned long backwards, forwards, try;
2613        int backwards_lno, forwards_lno, try_lno;
2614
2615        /*
2616         * If match_beginning or match_end is specified, there is no
2617         * point starting from a wrong line that will never match and
2618         * wander around and wait for a match at the specified end.
2619         */
2620        if (match_beginning)
2621                line = 0;
2622        else if (match_end)
2623                line = img->nr - preimage->nr;
2624
2625        /*
2626         * Because the comparison is unsigned, the following test
2627         * will also take care of a negative line number that can
2628         * result when match_end and preimage is larger than the target.
2629         */
2630        if ((size_t) line > img->nr)
2631                line = img->nr;
2632
2633        try = 0;
2634        for (i = 0; i < line; i++)
2635                try += img->line[i].len;
2636
2637        /*
2638         * There's probably some smart way to do this, but I'll leave
2639         * that to the smart and beautiful people. I'm simple and stupid.
2640         */
2641        backwards = try;
2642        backwards_lno = line;
2643        forwards = try;
2644        forwards_lno = line;
2645        try_lno = line;
2646
2647        for (i = 0; ; i++) {
2648                if (match_fragment(img, preimage, postimage,
2649                                   try, try_lno, ws_rule,
2650                                   match_beginning, match_end))
2651                        return try_lno;
2652
2653        again:
2654                if (backwards_lno == 0 && forwards_lno == img->nr)
2655                        break;
2656
2657                if (i & 1) {
2658                        if (backwards_lno == 0) {
2659                                i++;
2660                                goto again;
2661                        }
2662                        backwards_lno--;
2663                        backwards -= img->line[backwards_lno].len;
2664                        try = backwards;
2665                        try_lno = backwards_lno;
2666                } else {
2667                        if (forwards_lno == img->nr) {
2668                                i++;
2669                                goto again;
2670                        }
2671                        forwards += img->line[forwards_lno].len;
2672                        forwards_lno++;
2673                        try = forwards;
2674                        try_lno = forwards_lno;
2675                }
2676
2677        }
2678        return -1;
2679}
2680
2681static void remove_first_line(struct image *img)
2682{
2683        img->buf += img->line[0].len;
2684        img->len -= img->line[0].len;
2685        img->line++;
2686        img->nr--;
2687}
2688
2689static void remove_last_line(struct image *img)
2690{
2691        img->len -= img->line[--img->nr].len;
2692}
2693
2694/*
2695 * The change from "preimage" and "postimage" has been found to
2696 * apply at applied_pos (counts in line numbers) in "img".
2697 * Update "img" to remove "preimage" and replace it with "postimage".
2698 */
2699static void update_image(struct apply_state *state,
2700                         struct image *img,
2701                         int applied_pos,
2702                         struct image *preimage,
2703                         struct image *postimage)
2704{
2705        /*
2706         * remove the copy of preimage at offset in img
2707         * and replace it with postimage
2708         */
2709        int i, nr;
2710        size_t remove_count, insert_count, applied_at = 0;
2711        char *result;
2712        int preimage_limit;
2713
2714        /*
2715         * If we are removing blank lines at the end of img,
2716         * the preimage may extend beyond the end.
2717         * If that is the case, we must be careful only to
2718         * remove the part of the preimage that falls within
2719         * the boundaries of img. Initialize preimage_limit
2720         * to the number of lines in the preimage that falls
2721         * within the boundaries.
2722         */
2723        preimage_limit = preimage->nr;
2724        if (preimage_limit > img->nr - applied_pos)
2725                preimage_limit = img->nr - applied_pos;
2726
2727        for (i = 0; i < applied_pos; i++)
2728                applied_at += img->line[i].len;
2729
2730        remove_count = 0;
2731        for (i = 0; i < preimage_limit; i++)
2732                remove_count += img->line[applied_pos + i].len;
2733        insert_count = postimage->len;
2734
2735        /* Adjust the contents */
2736        result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2737        memcpy(result, img->buf, applied_at);
2738        memcpy(result + applied_at, postimage->buf, postimage->len);
2739        memcpy(result + applied_at + postimage->len,
2740               img->buf + (applied_at + remove_count),
2741               img->len - (applied_at + remove_count));
2742        free(img->buf);
2743        img->buf = result;
2744        img->len += insert_count - remove_count;
2745        result[img->len] = '\0';
2746
2747        /* Adjust the line table */
2748        nr = img->nr + postimage->nr - preimage_limit;
2749        if (preimage_limit < postimage->nr) {
2750                /*
2751                 * NOTE: this knows that we never call remove_first_line()
2752                 * on anything other than pre/post image.
2753                 */
2754                REALLOC_ARRAY(img->line, nr);
2755                img->line_allocated = img->line;
2756        }
2757        if (preimage_limit != postimage->nr)
2758                memmove(img->line + applied_pos + postimage->nr,
2759                        img->line + applied_pos + preimage_limit,
2760                        (img->nr - (applied_pos + preimage_limit)) *
2761                        sizeof(*img->line));
2762        memcpy(img->line + applied_pos,
2763               postimage->line,
2764               postimage->nr * sizeof(*img->line));
2765        if (!state->allow_overlap)
2766                for (i = 0; i < postimage->nr; i++)
2767                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2768        img->nr = nr;
2769}
2770
2771/*
2772 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2773 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2774 * replace the part of "img" with "postimage" text.
2775 */
2776static int apply_one_fragment(struct apply_state *state,
2777                              struct image *img, struct fragment *frag,
2778                              int inaccurate_eof, unsigned ws_rule,
2779                              int nth_fragment)
2780{
2781        int match_beginning, match_end;
2782        const char *patch = frag->patch;
2783        int size = frag->size;
2784        char *old, *oldlines;
2785        struct strbuf newlines;
2786        int new_blank_lines_at_end = 0;
2787        int found_new_blank_lines_at_end = 0;
2788        int hunk_linenr = frag->linenr;
2789        unsigned long leading, trailing;
2790        int pos, applied_pos;
2791        struct image preimage;
2792        struct image postimage;
2793
2794        memset(&preimage, 0, sizeof(preimage));
2795        memset(&postimage, 0, sizeof(postimage));
2796        oldlines = xmalloc(size);
2797        strbuf_init(&newlines, size);
2798
2799        old = oldlines;
2800        while (size > 0) {
2801                char first;
2802                int len = linelen(patch, size);
2803                int plen;
2804                int added_blank_line = 0;
2805                int is_blank_context = 0;
2806                size_t start;
2807
2808                if (!len)
2809                        break;
2810
2811                /*
2812                 * "plen" is how much of the line we should use for
2813                 * the actual patch data. Normally we just remove the
2814                 * first character on the line, but if the line is
2815                 * followed by "\ No newline", then we also remove the
2816                 * last one (which is the newline, of course).
2817                 */
2818                plen = len - 1;
2819                if (len < size && patch[len] == '\\')
2820                        plen--;
2821                first = *patch;
2822                if (state->apply_in_reverse) {
2823                        if (first == '-')
2824                                first = '+';
2825                        else if (first == '+')
2826                                first = '-';
2827                }
2828
2829                switch (first) {
2830                case '\n':
2831                        /* Newer GNU diff, empty context line */
2832                        if (plen < 0)
2833                                /* ... followed by '\No newline'; nothing */
2834                                break;
2835                        *old++ = '\n';
2836                        strbuf_addch(&newlines, '\n');
2837                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2838                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2839                        is_blank_context = 1;
2840                        break;
2841                case ' ':
2842                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2843                            ws_blank_line(patch + 1, plen, ws_rule))
2844                                is_blank_context = 1;
2845                case '-':
2846                        memcpy(old, patch + 1, plen);
2847                        add_line_info(&preimage, old, plen,
2848                                      (first == ' ' ? LINE_COMMON : 0));
2849                        old += plen;
2850                        if (first == '-')
2851                                break;
2852                /* Fall-through for ' ' */
2853                case '+':
2854                        /* --no-add does not add new lines */
2855                        if (first == '+' && state->no_add)
2856                                break;
2857
2858                        start = newlines.len;
2859                        if (first != '+' ||
2860                            !state->whitespace_error ||
2861                            ws_error_action != correct_ws_error) {
2862                                strbuf_add(&newlines, patch + 1, plen);
2863                        }
2864                        else {
2865                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2866                        }
2867                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2868                                      (first == '+' ? 0 : LINE_COMMON));
2869                        if (first == '+' &&
2870                            (ws_rule & WS_BLANK_AT_EOF) &&
2871                            ws_blank_line(patch + 1, plen, ws_rule))
2872                                added_blank_line = 1;
2873                        break;
2874                case '@': case '\\':
2875                        /* Ignore it, we already handled it */
2876                        break;
2877                default:
2878                        if (state->apply_verbosely)
2879                                error(_("invalid start of line: '%c'"), first);
2880                        applied_pos = -1;
2881                        goto out;
2882                }
2883                if (added_blank_line) {
2884                        if (!new_blank_lines_at_end)
2885                                found_new_blank_lines_at_end = hunk_linenr;
2886                        new_blank_lines_at_end++;
2887                }
2888                else if (is_blank_context)
2889                        ;
2890                else
2891                        new_blank_lines_at_end = 0;
2892                patch += len;
2893                size -= len;
2894                hunk_linenr++;
2895        }
2896        if (inaccurate_eof &&
2897            old > oldlines && old[-1] == '\n' &&
2898            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2899                old--;
2900                strbuf_setlen(&newlines, newlines.len - 1);
2901        }
2902
2903        leading = frag->leading;
2904        trailing = frag->trailing;
2905
2906        /*
2907         * A hunk to change lines at the beginning would begin with
2908         * @@ -1,L +N,M @@
2909         * but we need to be careful.  -U0 that inserts before the second
2910         * line also has this pattern.
2911         *
2912         * And a hunk to add to an empty file would begin with
2913         * @@ -0,0 +N,M @@
2914         *
2915         * In other words, a hunk that is (frag->oldpos <= 1) with or
2916         * without leading context must match at the beginning.
2917         */
2918        match_beginning = (!frag->oldpos ||
2919                           (frag->oldpos == 1 && !state->unidiff_zero));
2920
2921        /*
2922         * A hunk without trailing lines must match at the end.
2923         * However, we simply cannot tell if a hunk must match end
2924         * from the lack of trailing lines if the patch was generated
2925         * with unidiff without any context.
2926         */
2927        match_end = !state->unidiff_zero && !trailing;
2928
2929        pos = frag->newpos ? (frag->newpos - 1) : 0;
2930        preimage.buf = oldlines;
2931        preimage.len = old - oldlines;
2932        postimage.buf = newlines.buf;
2933        postimage.len = newlines.len;
2934        preimage.line = preimage.line_allocated;
2935        postimage.line = postimage.line_allocated;
2936
2937        for (;;) {
2938
2939                applied_pos = find_pos(img, &preimage, &postimage, pos,
2940                                       ws_rule, match_beginning, match_end);
2941
2942                if (applied_pos >= 0)
2943                        break;
2944
2945                /* Am I at my context limits? */
2946                if ((leading <= state->p_context) && (trailing <= state->p_context))
2947                        break;
2948                if (match_beginning || match_end) {
2949                        match_beginning = match_end = 0;
2950                        continue;
2951                }
2952
2953                /*
2954                 * Reduce the number of context lines; reduce both
2955                 * leading and trailing if they are equal otherwise
2956                 * just reduce the larger context.
2957                 */
2958                if (leading >= trailing) {
2959                        remove_first_line(&preimage);
2960                        remove_first_line(&postimage);
2961                        pos--;
2962                        leading--;
2963                }
2964                if (trailing > leading) {
2965                        remove_last_line(&preimage);
2966                        remove_last_line(&postimage);
2967                        trailing--;
2968                }
2969        }
2970
2971        if (applied_pos >= 0) {
2972                if (new_blank_lines_at_end &&
2973                    preimage.nr + applied_pos >= img->nr &&
2974                    (ws_rule & WS_BLANK_AT_EOF) &&
2975                    ws_error_action != nowarn_ws_error) {
2976                        record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
2977                                        found_new_blank_lines_at_end);
2978                        if (ws_error_action == correct_ws_error) {
2979                                while (new_blank_lines_at_end--)
2980                                        remove_last_line(&postimage);
2981                        }
2982                        /*
2983                         * We would want to prevent write_out_results()
2984                         * from taking place in apply_patch() that follows
2985                         * the callchain led us here, which is:
2986                         * apply_patch->check_patch_list->check_patch->
2987                         * apply_data->apply_fragments->apply_one_fragment
2988                         */
2989                        if (ws_error_action == die_on_ws_error)
2990                                state->apply = 0;
2991                }
2992
2993                if (state->apply_verbosely && applied_pos != pos) {
2994                        int offset = applied_pos - pos;
2995                        if (state->apply_in_reverse)
2996                                offset = 0 - offset;
2997                        fprintf_ln(stderr,
2998                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
2999                                      "Hunk #%d succeeded at %d (offset %d lines).",
3000                                      offset),
3001                                   nth_fragment, applied_pos + 1, offset);
3002                }
3003
3004                /*
3005                 * Warn if it was necessary to reduce the number
3006                 * of context lines.
3007                 */
3008                if ((leading != frag->leading) ||
3009                    (trailing != frag->trailing))
3010                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3011                                             " to apply fragment at %d"),
3012                                   leading, trailing, applied_pos+1);
3013                update_image(state, img, applied_pos, &preimage, &postimage);
3014        } else {
3015                if (state->apply_verbosely)
3016                        error(_("while searching for:\n%.*s"),
3017                              (int)(old - oldlines), oldlines);
3018        }
3019
3020out:
3021        free(oldlines);
3022        strbuf_release(&newlines);
3023        free(preimage.line_allocated);
3024        free(postimage.line_allocated);
3025
3026        return (applied_pos < 0);
3027}
3028
3029static int apply_binary_fragment(struct apply_state *state,
3030                                 struct image *img,
3031                                 struct patch *patch)
3032{
3033        struct fragment *fragment = patch->fragments;
3034        unsigned long len;
3035        void *dst;
3036
3037        if (!fragment)
3038                return error(_("missing binary patch data for '%s'"),
3039                             patch->new_name ?
3040                             patch->new_name :
3041                             patch->old_name);
3042
3043        /* Binary patch is irreversible without the optional second hunk */
3044        if (state->apply_in_reverse) {
3045                if (!fragment->next)
3046                        return error("cannot reverse-apply a binary patch "
3047                                     "without the reverse hunk to '%s'",
3048                                     patch->new_name
3049                                     ? patch->new_name : patch->old_name);
3050                fragment = fragment->next;
3051        }
3052        switch (fragment->binary_patch_method) {
3053        case BINARY_DELTA_DEFLATED:
3054                dst = patch_delta(img->buf, img->len, fragment->patch,
3055                                  fragment->size, &len);
3056                if (!dst)
3057                        return -1;
3058                clear_image(img);
3059                img->buf = dst;
3060                img->len = len;
3061                return 0;
3062        case BINARY_LITERAL_DEFLATED:
3063                clear_image(img);
3064                img->len = fragment->size;
3065                img->buf = xmemdupz(fragment->patch, img->len);
3066                return 0;
3067        }
3068        return -1;
3069}
3070
3071/*
3072 * Replace "img" with the result of applying the binary patch.
3073 * The binary patch data itself in patch->fragment is still kept
3074 * but the preimage prepared by the caller in "img" is freed here
3075 * or in the helper function apply_binary_fragment() this calls.
3076 */
3077static int apply_binary(struct apply_state *state,
3078                        struct image *img,
3079                        struct patch *patch)
3080{
3081        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3082        unsigned char sha1[20];
3083
3084        /*
3085         * For safety, we require patch index line to contain
3086         * full 40-byte textual SHA1 for old and new, at least for now.
3087         */
3088        if (strlen(patch->old_sha1_prefix) != 40 ||
3089            strlen(patch->new_sha1_prefix) != 40 ||
3090            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
3091            get_sha1_hex(patch->new_sha1_prefix, sha1))
3092                return error("cannot apply binary patch to '%s' "
3093                             "without full index line", name);
3094
3095        if (patch->old_name) {
3096                /*
3097                 * See if the old one matches what the patch
3098                 * applies to.
3099                 */
3100                hash_sha1_file(img->buf, img->len, blob_type, sha1);
3101                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
3102                        return error("the patch applies to '%s' (%s), "
3103                                     "which does not match the "
3104                                     "current contents.",
3105                                     name, sha1_to_hex(sha1));
3106        }
3107        else {
3108                /* Otherwise, the old one must be empty. */
3109                if (img->len)
3110                        return error("the patch applies to an empty "
3111                                     "'%s' but it is not empty", name);
3112        }
3113
3114        get_sha1_hex(patch->new_sha1_prefix, sha1);
3115        if (is_null_sha1(sha1)) {
3116                clear_image(img);
3117                return 0; /* deletion patch */
3118        }
3119
3120        if (has_sha1_file(sha1)) {
3121                /* We already have the postimage */
3122                enum object_type type;
3123                unsigned long size;
3124                char *result;
3125
3126                result = read_sha1_file(sha1, &type, &size);
3127                if (!result)
3128                        return error("the necessary postimage %s for "
3129                                     "'%s' cannot be read",
3130                                     patch->new_sha1_prefix, name);
3131                clear_image(img);
3132                img->buf = result;
3133                img->len = size;
3134        } else {
3135                /*
3136                 * We have verified buf matches the preimage;
3137                 * apply the patch data to it, which is stored
3138                 * in the patch->fragments->{patch,size}.
3139                 */
3140                if (apply_binary_fragment(state, img, patch))
3141                        return error(_("binary patch does not apply to '%s'"),
3142                                     name);
3143
3144                /* verify that the result matches */
3145                hash_sha1_file(img->buf, img->len, blob_type, sha1);
3146                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
3147                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3148                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
3149        }
3150
3151        return 0;
3152}
3153
3154static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3155{
3156        struct fragment *frag = patch->fragments;
3157        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3158        unsigned ws_rule = patch->ws_rule;
3159        unsigned inaccurate_eof = patch->inaccurate_eof;
3160        int nth = 0;
3161
3162        if (patch->is_binary)
3163                return apply_binary(state, img, patch);
3164
3165        while (frag) {
3166                nth++;
3167                if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3168                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
3169                        if (!state->apply_with_reject)
3170                                return -1;
3171                        frag->rejected = 1;
3172                }
3173                frag = frag->next;
3174        }
3175        return 0;
3176}
3177
3178static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3179{
3180        if (S_ISGITLINK(mode)) {
3181                strbuf_grow(buf, 100);
3182                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3183        } else {
3184                enum object_type type;
3185                unsigned long sz;
3186                char *result;
3187
3188                result = read_sha1_file(sha1, &type, &sz);
3189                if (!result)
3190                        return -1;
3191                /* XXX read_sha1_file NUL-terminates */
3192                strbuf_attach(buf, result, sz, sz + 1);
3193        }
3194        return 0;
3195}
3196
3197static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3198{
3199        if (!ce)
3200                return 0;
3201        return read_blob_object(buf, ce->sha1, ce->ce_mode);
3202}
3203
3204static struct patch *in_fn_table(const char *name)
3205{
3206        struct string_list_item *item;
3207
3208        if (name == NULL)
3209                return NULL;
3210
3211        item = string_list_lookup(&fn_table, name);
3212        if (item != NULL)
3213                return (struct patch *)item->util;
3214
3215        return NULL;
3216}
3217
3218/*
3219 * item->util in the filename table records the status of the path.
3220 * Usually it points at a patch (whose result records the contents
3221 * of it after applying it), but it could be PATH_WAS_DELETED for a
3222 * path that a previously applied patch has already removed, or
3223 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3224 *
3225 * The latter is needed to deal with a case where two paths A and B
3226 * are swapped by first renaming A to B and then renaming B to A;
3227 * moving A to B should not be prevented due to presence of B as we
3228 * will remove it in a later patch.
3229 */
3230#define PATH_TO_BE_DELETED ((struct patch *) -2)
3231#define PATH_WAS_DELETED ((struct patch *) -1)
3232
3233static int to_be_deleted(struct patch *patch)
3234{
3235        return patch == PATH_TO_BE_DELETED;
3236}
3237
3238static int was_deleted(struct patch *patch)
3239{
3240        return patch == PATH_WAS_DELETED;
3241}
3242
3243static void add_to_fn_table(struct patch *patch)
3244{
3245        struct string_list_item *item;
3246
3247        /*
3248         * Always add new_name unless patch is a deletion
3249         * This should cover the cases for normal diffs,
3250         * file creations and copies
3251         */
3252        if (patch->new_name != NULL) {
3253                item = string_list_insert(&fn_table, patch->new_name);
3254                item->util = patch;
3255        }
3256
3257        /*
3258         * store a failure on rename/deletion cases because
3259         * later chunks shouldn't patch old names
3260         */
3261        if ((patch->new_name == NULL) || (patch->is_rename)) {
3262                item = string_list_insert(&fn_table, patch->old_name);
3263                item->util = PATH_WAS_DELETED;
3264        }
3265}
3266
3267static void prepare_fn_table(struct patch *patch)
3268{
3269        /*
3270         * store information about incoming file deletion
3271         */
3272        while (patch) {
3273                if ((patch->new_name == NULL) || (patch->is_rename)) {
3274                        struct string_list_item *item;
3275                        item = string_list_insert(&fn_table, patch->old_name);
3276                        item->util = PATH_TO_BE_DELETED;
3277                }
3278                patch = patch->next;
3279        }
3280}
3281
3282static int checkout_target(struct index_state *istate,
3283                           struct cache_entry *ce, struct stat *st)
3284{
3285        struct checkout costate;
3286
3287        memset(&costate, 0, sizeof(costate));
3288        costate.base_dir = "";
3289        costate.refresh_cache = 1;
3290        costate.istate = istate;
3291        if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3292                return error(_("cannot checkout %s"), ce->name);
3293        return 0;
3294}
3295
3296static struct patch *previous_patch(struct patch *patch, int *gone)
3297{
3298        struct patch *previous;
3299
3300        *gone = 0;
3301        if (patch->is_copy || patch->is_rename)
3302                return NULL; /* "git" patches do not depend on the order */
3303
3304        previous = in_fn_table(patch->old_name);
3305        if (!previous)
3306                return NULL;
3307
3308        if (to_be_deleted(previous))
3309                return NULL; /* the deletion hasn't happened yet */
3310
3311        if (was_deleted(previous))
3312                *gone = 1;
3313
3314        return previous;
3315}
3316
3317static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3318{
3319        if (S_ISGITLINK(ce->ce_mode)) {
3320                if (!S_ISDIR(st->st_mode))
3321                        return -1;
3322                return 0;
3323        }
3324        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3325}
3326
3327#define SUBMODULE_PATCH_WITHOUT_INDEX 1
3328
3329static int load_patch_target(struct apply_state *state,
3330                             struct strbuf *buf,
3331                             const struct cache_entry *ce,
3332                             struct stat *st,
3333                             const char *name,
3334                             unsigned expected_mode)
3335{
3336        if (state->cached || state->check_index) {
3337                if (read_file_or_gitlink(ce, buf))
3338                        return error(_("read of %s failed"), name);
3339        } else if (name) {
3340                if (S_ISGITLINK(expected_mode)) {
3341                        if (ce)
3342                                return read_file_or_gitlink(ce, buf);
3343                        else
3344                                return SUBMODULE_PATCH_WITHOUT_INDEX;
3345                } else if (has_symlink_leading_path(name, strlen(name))) {
3346                        return error(_("reading from '%s' beyond a symbolic link"), name);
3347                } else {
3348                        if (read_old_data(st, name, buf))
3349                                return error(_("read of %s failed"), name);
3350                }
3351        }
3352        return 0;
3353}
3354
3355/*
3356 * We are about to apply "patch"; populate the "image" with the
3357 * current version we have, from the working tree or from the index,
3358 * depending on the situation e.g. --cached/--index.  If we are
3359 * applying a non-git patch that incrementally updates the tree,
3360 * we read from the result of a previous diff.
3361 */
3362static int load_preimage(struct apply_state *state,
3363                         struct image *image,
3364                         struct patch *patch, struct stat *st,
3365                         const struct cache_entry *ce)
3366{
3367        struct strbuf buf = STRBUF_INIT;
3368        size_t len;
3369        char *img;
3370        struct patch *previous;
3371        int status;
3372
3373        previous = previous_patch(patch, &status);
3374        if (status)
3375                return error(_("path %s has been renamed/deleted"),
3376                             patch->old_name);
3377        if (previous) {
3378                /* We have a patched copy in memory; use that. */
3379                strbuf_add(&buf, previous->result, previous->resultsize);
3380        } else {
3381                status = load_patch_target(state, &buf, ce, st,
3382                                           patch->old_name, patch->old_mode);
3383                if (status < 0)
3384                        return status;
3385                else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3386                        /*
3387                         * There is no way to apply subproject
3388                         * patch without looking at the index.
3389                         * NEEDSWORK: shouldn't this be flagged
3390                         * as an error???
3391                         */
3392                        free_fragment_list(patch->fragments);
3393                        patch->fragments = NULL;
3394                } else if (status) {
3395                        return error(_("read of %s failed"), patch->old_name);
3396                }
3397        }
3398
3399        img = strbuf_detach(&buf, &len);
3400        prepare_image(image, img, len, !patch->is_binary);
3401        return 0;
3402}
3403
3404static int three_way_merge(struct image *image,
3405                           char *path,
3406                           const unsigned char *base,
3407                           const unsigned char *ours,
3408                           const unsigned char *theirs)
3409{
3410        mmfile_t base_file, our_file, their_file;
3411        mmbuffer_t result = { NULL };
3412        int status;
3413
3414        read_mmblob(&base_file, base);
3415        read_mmblob(&our_file, ours);
3416        read_mmblob(&their_file, theirs);
3417        status = ll_merge(&result, path,
3418                          &base_file, "base",
3419                          &our_file, "ours",
3420                          &their_file, "theirs", NULL);
3421        free(base_file.ptr);
3422        free(our_file.ptr);
3423        free(their_file.ptr);
3424        if (status < 0 || !result.ptr) {
3425                free(result.ptr);
3426                return -1;
3427        }
3428        clear_image(image);
3429        image->buf = result.ptr;
3430        image->len = result.size;
3431
3432        return status;
3433}
3434
3435/*
3436 * When directly falling back to add/add three-way merge, we read from
3437 * the current contents of the new_name.  In no cases other than that
3438 * this function will be called.
3439 */
3440static int load_current(struct apply_state *state,
3441                        struct image *image,
3442                        struct patch *patch)
3443{
3444        struct strbuf buf = STRBUF_INIT;
3445        int status, pos;
3446        size_t len;
3447        char *img;
3448        struct stat st;
3449        struct cache_entry *ce;
3450        char *name = patch->new_name;
3451        unsigned mode = patch->new_mode;
3452
3453        if (!patch->is_new)
3454                die("BUG: patch to %s is not a creation", patch->old_name);
3455
3456        pos = cache_name_pos(name, strlen(name));
3457        if (pos < 0)
3458                return error(_("%s: does not exist in index"), name);
3459        ce = active_cache[pos];
3460        if (lstat(name, &st)) {
3461                if (errno != ENOENT)
3462                        return error(_("%s: %s"), name, strerror(errno));
3463                if (checkout_target(&the_index, ce, &st))
3464                        return -1;
3465        }
3466        if (verify_index_match(ce, &st))
3467                return error(_("%s: does not match index"), name);
3468
3469        status = load_patch_target(state, &buf, ce, &st, name, mode);
3470        if (status < 0)
3471                return status;
3472        else if (status)
3473                return -1;
3474        img = strbuf_detach(&buf, &len);
3475        prepare_image(image, img, len, !patch->is_binary);
3476        return 0;
3477}
3478
3479static int try_threeway(struct apply_state *state,
3480                        struct image *image,
3481                        struct patch *patch,
3482                        struct stat *st,
3483                        const struct cache_entry *ce)
3484{
3485        unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3486        struct strbuf buf = STRBUF_INIT;
3487        size_t len;
3488        int status;
3489        char *img;
3490        struct image tmp_image;
3491
3492        /* No point falling back to 3-way merge in these cases */
3493        if (patch->is_delete ||
3494            S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3495                return -1;
3496
3497        /* Preimage the patch was prepared for */
3498        if (patch->is_new)
3499                write_sha1_file("", 0, blob_type, pre_sha1);
3500        else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3501                 read_blob_object(&buf, pre_sha1, patch->old_mode))
3502                return error("repository lacks the necessary blob to fall back on 3-way merge.");
3503
3504        fprintf(stderr, "Falling back to three-way merge...\n");
3505
3506        img = strbuf_detach(&buf, &len);
3507        prepare_image(&tmp_image, img, len, 1);
3508        /* Apply the patch to get the post image */
3509        if (apply_fragments(state, &tmp_image, patch) < 0) {
3510                clear_image(&tmp_image);
3511                return -1;
3512        }
3513        /* post_sha1[] is theirs */
3514        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3515        clear_image(&tmp_image);
3516
3517        /* our_sha1[] is ours */
3518        if (patch->is_new) {
3519                if (load_current(state, &tmp_image, patch))
3520                        return error("cannot read the current contents of '%s'",
3521                                     patch->new_name);
3522        } else {
3523                if (load_preimage(state, &tmp_image, patch, st, ce))
3524                        return error("cannot read the current contents of '%s'",
3525                                     patch->old_name);
3526        }
3527        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3528        clear_image(&tmp_image);
3529
3530        /* in-core three-way merge between post and our using pre as base */
3531        status = three_way_merge(image, patch->new_name,
3532                                 pre_sha1, our_sha1, post_sha1);
3533        if (status < 0) {
3534                fprintf(stderr, "Failed to fall back on three-way merge...\n");
3535                return status;
3536        }
3537
3538        if (status) {
3539                patch->conflicted_threeway = 1;
3540                if (patch->is_new)
3541                        oidclr(&patch->threeway_stage[0]);
3542                else
3543                        hashcpy(patch->threeway_stage[0].hash, pre_sha1);
3544                hashcpy(patch->threeway_stage[1].hash, our_sha1);
3545                hashcpy(patch->threeway_stage[2].hash, post_sha1);
3546                fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3547        } else {
3548                fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3549        }
3550        return 0;
3551}
3552
3553static int apply_data(struct apply_state *state, struct patch *patch,
3554                      struct stat *st, const struct cache_entry *ce)
3555{
3556        struct image image;
3557
3558        if (load_preimage(state, &image, patch, st, ce) < 0)
3559                return -1;
3560
3561        if (patch->direct_to_threeway ||
3562            apply_fragments(state, &image, patch) < 0) {
3563                /* Note: with --reject, apply_fragments() returns 0 */
3564                if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3565                        return -1;
3566        }
3567        patch->result = image.buf;
3568        patch->resultsize = image.len;
3569        add_to_fn_table(patch);
3570        free(image.line_allocated);
3571
3572        if (0 < patch->is_delete && patch->resultsize)
3573                return error(_("removal patch leaves file contents"));
3574
3575        return 0;
3576}
3577
3578/*
3579 * If "patch" that we are looking at modifies or deletes what we have,
3580 * we would want it not to lose any local modification we have, either
3581 * in the working tree or in the index.
3582 *
3583 * This also decides if a non-git patch is a creation patch or a
3584 * modification to an existing empty file.  We do not check the state
3585 * of the current tree for a creation patch in this function; the caller
3586 * check_patch() separately makes sure (and errors out otherwise) that
3587 * the path the patch creates does not exist in the current tree.
3588 */
3589static int check_preimage(struct apply_state *state,
3590                          struct patch *patch,
3591                          struct cache_entry **ce,
3592                          struct stat *st)
3593{
3594        const char *old_name = patch->old_name;
3595        struct patch *previous = NULL;
3596        int stat_ret = 0, status;
3597        unsigned st_mode = 0;
3598
3599        if (!old_name)
3600                return 0;
3601
3602        assert(patch->is_new <= 0);
3603        previous = previous_patch(patch, &status);
3604
3605        if (status)
3606                return error(_("path %s has been renamed/deleted"), old_name);
3607        if (previous) {
3608                st_mode = previous->new_mode;
3609        } else if (!state->cached) {
3610                stat_ret = lstat(old_name, st);
3611                if (stat_ret && errno != ENOENT)
3612                        return error(_("%s: %s"), old_name, strerror(errno));
3613        }
3614
3615        if (state->check_index && !previous) {
3616                int pos = cache_name_pos(old_name, strlen(old_name));
3617                if (pos < 0) {
3618                        if (patch->is_new < 0)
3619                                goto is_new;
3620                        return error(_("%s: does not exist in index"), old_name);
3621                }
3622                *ce = active_cache[pos];
3623                if (stat_ret < 0) {
3624                        if (checkout_target(&the_index, *ce, st))
3625                                return -1;
3626                }
3627                if (!state->cached && verify_index_match(*ce, st))
3628                        return error(_("%s: does not match index"), old_name);
3629                if (state->cached)
3630                        st_mode = (*ce)->ce_mode;
3631        } else if (stat_ret < 0) {
3632                if (patch->is_new < 0)
3633                        goto is_new;
3634                return error(_("%s: %s"), old_name, strerror(errno));
3635        }
3636
3637        if (!state->cached && !previous)
3638                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3639
3640        if (patch->is_new < 0)
3641                patch->is_new = 0;
3642        if (!patch->old_mode)
3643                patch->old_mode = st_mode;
3644        if ((st_mode ^ patch->old_mode) & S_IFMT)
3645                return error(_("%s: wrong type"), old_name);
3646        if (st_mode != patch->old_mode)
3647                warning(_("%s has type %o, expected %o"),
3648                        old_name, st_mode, patch->old_mode);
3649        if (!patch->new_mode && !patch->is_delete)
3650                patch->new_mode = st_mode;
3651        return 0;
3652
3653 is_new:
3654        patch->is_new = 1;
3655        patch->is_delete = 0;
3656        free(patch->old_name);
3657        patch->old_name = NULL;
3658        return 0;
3659}
3660
3661
3662#define EXISTS_IN_INDEX 1
3663#define EXISTS_IN_WORKTREE 2
3664
3665static int check_to_create(struct apply_state *state,
3666                           const char *new_name,
3667                           int ok_if_exists)
3668{
3669        struct stat nst;
3670
3671        if (state->check_index &&
3672            cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3673            !ok_if_exists)
3674                return EXISTS_IN_INDEX;
3675        if (state->cached)
3676                return 0;
3677
3678        if (!lstat(new_name, &nst)) {
3679                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3680                        return 0;
3681                /*
3682                 * A leading component of new_name might be a symlink
3683                 * that is going to be removed with this patch, but
3684                 * still pointing at somewhere that has the path.
3685                 * In such a case, path "new_name" does not exist as
3686                 * far as git is concerned.
3687                 */
3688                if (has_symlink_leading_path(new_name, strlen(new_name)))
3689                        return 0;
3690
3691                return EXISTS_IN_WORKTREE;
3692        } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3693                return error("%s: %s", new_name, strerror(errno));
3694        }
3695        return 0;
3696}
3697
3698/*
3699 * We need to keep track of how symlinks in the preimage are
3700 * manipulated by the patches.  A patch to add a/b/c where a/b
3701 * is a symlink should not be allowed to affect the directory
3702 * the symlink points at, but if the same patch removes a/b,
3703 * it is perfectly fine, as the patch removes a/b to make room
3704 * to create a directory a/b so that a/b/c can be created.
3705 */
3706static struct string_list symlink_changes;
3707#define SYMLINK_GOES_AWAY 01
3708#define SYMLINK_IN_RESULT 02
3709
3710static uintptr_t register_symlink_changes(const char *path, uintptr_t what)
3711{
3712        struct string_list_item *ent;
3713
3714        ent = string_list_lookup(&symlink_changes, path);
3715        if (!ent) {
3716                ent = string_list_insert(&symlink_changes, path);
3717                ent->util = (void *)0;
3718        }
3719        ent->util = (void *)(what | ((uintptr_t)ent->util));
3720        return (uintptr_t)ent->util;
3721}
3722
3723static uintptr_t check_symlink_changes(const char *path)
3724{
3725        struct string_list_item *ent;
3726
3727        ent = string_list_lookup(&symlink_changes, path);
3728        if (!ent)
3729                return 0;
3730        return (uintptr_t)ent->util;
3731}
3732
3733static void prepare_symlink_changes(struct patch *patch)
3734{
3735        for ( ; patch; patch = patch->next) {
3736                if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3737                    (patch->is_rename || patch->is_delete))
3738                        /* the symlink at patch->old_name is removed */
3739                        register_symlink_changes(patch->old_name, SYMLINK_GOES_AWAY);
3740
3741                if (patch->new_name && S_ISLNK(patch->new_mode))
3742                        /* the symlink at patch->new_name is created or remains */
3743                        register_symlink_changes(patch->new_name, SYMLINK_IN_RESULT);
3744        }
3745}
3746
3747static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3748{
3749        do {
3750                unsigned int change;
3751
3752                while (--name->len && name->buf[name->len] != '/')
3753                        ; /* scan backwards */
3754                if (!name->len)
3755                        break;
3756                name->buf[name->len] = '\0';
3757                change = check_symlink_changes(name->buf);
3758                if (change & SYMLINK_IN_RESULT)
3759                        return 1;
3760                if (change & SYMLINK_GOES_AWAY)
3761                        /*
3762                         * This cannot be "return 0", because we may
3763                         * see a new one created at a higher level.
3764                         */
3765                        continue;
3766
3767                /* otherwise, check the preimage */
3768                if (state->check_index) {
3769                        struct cache_entry *ce;
3770
3771                        ce = cache_file_exists(name->buf, name->len, ignore_case);
3772                        if (ce && S_ISLNK(ce->ce_mode))
3773                                return 1;
3774                } else {
3775                        struct stat st;
3776                        if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3777                                return 1;
3778                }
3779        } while (1);
3780        return 0;
3781}
3782
3783static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3784{
3785        int ret;
3786        struct strbuf name = STRBUF_INIT;
3787
3788        assert(*name_ != '\0');
3789        strbuf_addstr(&name, name_);
3790        ret = path_is_beyond_symlink_1(state, &name);
3791        strbuf_release(&name);
3792
3793        return ret;
3794}
3795
3796static void die_on_unsafe_path(struct patch *patch)
3797{
3798        const char *old_name = NULL;
3799        const char *new_name = NULL;
3800        if (patch->is_delete)
3801                old_name = patch->old_name;
3802        else if (!patch->is_new && !patch->is_copy)
3803                old_name = patch->old_name;
3804        if (!patch->is_delete)
3805                new_name = patch->new_name;
3806
3807        if (old_name && !verify_path(old_name))
3808                die(_("invalid path '%s'"), old_name);
3809        if (new_name && !verify_path(new_name))
3810                die(_("invalid path '%s'"), new_name);
3811}
3812
3813/*
3814 * Check and apply the patch in-core; leave the result in patch->result
3815 * for the caller to write it out to the final destination.
3816 */
3817static int check_patch(struct apply_state *state, struct patch *patch)
3818{
3819        struct stat st;
3820        const char *old_name = patch->old_name;
3821        const char *new_name = patch->new_name;
3822        const char *name = old_name ? old_name : new_name;
3823        struct cache_entry *ce = NULL;
3824        struct patch *tpatch;
3825        int ok_if_exists;
3826        int status;
3827
3828        patch->rejected = 1; /* we will drop this after we succeed */
3829
3830        status = check_preimage(state, patch, &ce, &st);
3831        if (status)
3832                return status;
3833        old_name = patch->old_name;
3834
3835        /*
3836         * A type-change diff is always split into a patch to delete
3837         * old, immediately followed by a patch to create new (see
3838         * diff.c::run_diff()); in such a case it is Ok that the entry
3839         * to be deleted by the previous patch is still in the working
3840         * tree and in the index.
3841         *
3842         * A patch to swap-rename between A and B would first rename A
3843         * to B and then rename B to A.  While applying the first one,
3844         * the presence of B should not stop A from getting renamed to
3845         * B; ask to_be_deleted() about the later rename.  Removal of
3846         * B and rename from A to B is handled the same way by asking
3847         * was_deleted().
3848         */
3849        if ((tpatch = in_fn_table(new_name)) &&
3850            (was_deleted(tpatch) || to_be_deleted(tpatch)))
3851                ok_if_exists = 1;
3852        else
3853                ok_if_exists = 0;
3854
3855        if (new_name &&
3856            ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3857                int err = check_to_create(state, new_name, ok_if_exists);
3858
3859                if (err && state->threeway) {
3860                        patch->direct_to_threeway = 1;
3861                } else switch (err) {
3862                case 0:
3863                        break; /* happy */
3864                case EXISTS_IN_INDEX:
3865                        return error(_("%s: already exists in index"), new_name);
3866                        break;
3867                case EXISTS_IN_WORKTREE:
3868                        return error(_("%s: already exists in working directory"),
3869                                     new_name);
3870                default:
3871                        return err;
3872                }
3873
3874                if (!patch->new_mode) {
3875                        if (0 < patch->is_new)
3876                                patch->new_mode = S_IFREG | 0644;
3877                        else
3878                                patch->new_mode = patch->old_mode;
3879                }
3880        }
3881
3882        if (new_name && old_name) {
3883                int same = !strcmp(old_name, new_name);
3884                if (!patch->new_mode)
3885                        patch->new_mode = patch->old_mode;
3886                if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3887                        if (same)
3888                                return error(_("new mode (%o) of %s does not "
3889                                               "match old mode (%o)"),
3890                                        patch->new_mode, new_name,
3891                                        patch->old_mode);
3892                        else
3893                                return error(_("new mode (%o) of %s does not "
3894                                               "match old mode (%o) of %s"),
3895                                        patch->new_mode, new_name,
3896                                        patch->old_mode, old_name);
3897                }
3898        }
3899
3900        if (!state->unsafe_paths)
3901                die_on_unsafe_path(patch);
3902
3903        /*
3904         * An attempt to read from or delete a path that is beyond a
3905         * symbolic link will be prevented by load_patch_target() that
3906         * is called at the beginning of apply_data() so we do not
3907         * have to worry about a patch marked with "is_delete" bit
3908         * here.  We however need to make sure that the patch result
3909         * is not deposited to a path that is beyond a symbolic link
3910         * here.
3911         */
3912        if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3913                return error(_("affected file '%s' is beyond a symbolic link"),
3914                             patch->new_name);
3915
3916        if (apply_data(state, patch, &st, ce) < 0)
3917                return error(_("%s: patch does not apply"), name);
3918        patch->rejected = 0;
3919        return 0;
3920}
3921
3922static int check_patch_list(struct apply_state *state, struct patch *patch)
3923{
3924        int err = 0;
3925
3926        prepare_symlink_changes(patch);
3927        prepare_fn_table(patch);
3928        while (patch) {
3929                if (state->apply_verbosely)
3930                        say_patch_name(stderr,
3931                                       _("Checking patch %s..."), patch);
3932                err |= check_patch(state, patch);
3933                patch = patch->next;
3934        }
3935        return err;
3936}
3937
3938/* This function tries to read the sha1 from the current index */
3939static int get_current_sha1(const char *path, unsigned char *sha1)
3940{
3941        int pos;
3942
3943        if (read_cache() < 0)
3944                return -1;
3945        pos = cache_name_pos(path, strlen(path));
3946        if (pos < 0)
3947                return -1;
3948        hashcpy(sha1, active_cache[pos]->sha1);
3949        return 0;
3950}
3951
3952static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3953{
3954        /*
3955         * A usable gitlink patch has only one fragment (hunk) that looks like:
3956         * @@ -1 +1 @@
3957         * -Subproject commit <old sha1>
3958         * +Subproject commit <new sha1>
3959         * or
3960         * @@ -1 +0,0 @@
3961         * -Subproject commit <old sha1>
3962         * for a removal patch.
3963         */
3964        struct fragment *hunk = p->fragments;
3965        static const char heading[] = "-Subproject commit ";
3966        char *preimage;
3967
3968        if (/* does the patch have only one hunk? */
3969            hunk && !hunk->next &&
3970            /* is its preimage one line? */
3971            hunk->oldpos == 1 && hunk->oldlines == 1 &&
3972            /* does preimage begin with the heading? */
3973            (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3974            starts_with(++preimage, heading) &&
3975            /* does it record full SHA-1? */
3976            !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3977            preimage[sizeof(heading) + 40 - 1] == '\n' &&
3978            /* does the abbreviated name on the index line agree with it? */
3979            starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3980                return 0; /* it all looks fine */
3981
3982        /* we may have full object name on the index line */
3983        return get_sha1_hex(p->old_sha1_prefix, sha1);
3984}
3985
3986/* Build an index that contains the just the files needed for a 3way merge */
3987static void build_fake_ancestor(struct patch *list, const char *filename)
3988{
3989        struct patch *patch;
3990        struct index_state result = { NULL };
3991        static struct lock_file lock;
3992
3993        /* Once we start supporting the reverse patch, it may be
3994         * worth showing the new sha1 prefix, but until then...
3995         */
3996        for (patch = list; patch; patch = patch->next) {
3997                unsigned char sha1[20];
3998                struct cache_entry *ce;
3999                const char *name;
4000
4001                name = patch->old_name ? patch->old_name : patch->new_name;
4002                if (0 < patch->is_new)
4003                        continue;
4004
4005                if (S_ISGITLINK(patch->old_mode)) {
4006                        if (!preimage_sha1_in_gitlink_patch(patch, sha1))
4007                                ; /* ok, the textual part looks sane */
4008                        else
4009                                die("sha1 information is lacking or useless for submodule %s",
4010                                    name);
4011                } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
4012                        ; /* ok */
4013                } else if (!patch->lines_added && !patch->lines_deleted) {
4014                        /* mode-only change: update the current */
4015                        if (get_current_sha1(patch->old_name, sha1))
4016                                die("mode change for %s, which is not "
4017                                    "in current HEAD", name);
4018                } else
4019                        die("sha1 information is lacking or useless "
4020                            "(%s).", name);
4021
4022                ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
4023                if (!ce)
4024                        die(_("make_cache_entry failed for path '%s'"), name);
4025                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
4026                        die ("Could not add %s to temporary index", name);
4027        }
4028
4029        hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);
4030        if (write_locked_index(&result, &lock, COMMIT_LOCK))
4031                die ("Could not write temporary index to %s", filename);
4032
4033        discard_index(&result);
4034}
4035
4036static void stat_patch_list(struct patch *patch)
4037{
4038        int files, adds, dels;
4039
4040        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4041                files++;
4042                adds += patch->lines_added;
4043                dels += patch->lines_deleted;
4044                show_stats(patch);
4045        }
4046
4047        print_stat_summary(stdout, files, adds, dels);
4048}
4049
4050static void numstat_patch_list(struct apply_state *state,
4051                               struct patch *patch)
4052{
4053        for ( ; patch; patch = patch->next) {
4054                const char *name;
4055                name = patch->new_name ? patch->new_name : patch->old_name;
4056                if (patch->is_binary)
4057                        printf("-\t-\t");
4058                else
4059                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4060                write_name_quoted(name, stdout, state->line_termination);
4061        }
4062}
4063
4064static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4065{
4066        if (mode)
4067                printf(" %s mode %06o %s\n", newdelete, mode, name);
4068        else
4069                printf(" %s %s\n", newdelete, name);
4070}
4071
4072static void show_mode_change(struct patch *p, int show_name)
4073{
4074        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4075                if (show_name)
4076                        printf(" mode change %06o => %06o %s\n",
4077                               p->old_mode, p->new_mode, p->new_name);
4078                else
4079                        printf(" mode change %06o => %06o\n",
4080                               p->old_mode, p->new_mode);
4081        }
4082}
4083
4084static void show_rename_copy(struct patch *p)
4085{
4086        const char *renamecopy = p->is_rename ? "rename" : "copy";
4087        const char *old, *new;
4088
4089        /* Find common prefix */
4090        old = p->old_name;
4091        new = p->new_name;
4092        while (1) {
4093                const char *slash_old, *slash_new;
4094                slash_old = strchr(old, '/');
4095                slash_new = strchr(new, '/');
4096                if (!slash_old ||
4097                    !slash_new ||
4098                    slash_old - old != slash_new - new ||
4099                    memcmp(old, new, slash_new - new))
4100                        break;
4101                old = slash_old + 1;
4102                new = slash_new + 1;
4103        }
4104        /* p->old_name thru old is the common prefix, and old and new
4105         * through the end of names are renames
4106         */
4107        if (old != p->old_name)
4108                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4109                       (int)(old - p->old_name), p->old_name,
4110                       old, new, p->score);
4111        else
4112                printf(" %s %s => %s (%d%%)\n", renamecopy,
4113                       p->old_name, p->new_name, p->score);
4114        show_mode_change(p, 0);
4115}
4116
4117static void summary_patch_list(struct patch *patch)
4118{
4119        struct patch *p;
4120
4121        for (p = patch; p; p = p->next) {
4122                if (p->is_new)
4123                        show_file_mode_name("create", p->new_mode, p->new_name);
4124                else if (p->is_delete)
4125                        show_file_mode_name("delete", p->old_mode, p->old_name);
4126                else {
4127                        if (p->is_rename || p->is_copy)
4128                                show_rename_copy(p);
4129                        else {
4130                                if (p->score) {
4131                                        printf(" rewrite %s (%d%%)\n",
4132                                               p->new_name, p->score);
4133                                        show_mode_change(p, 0);
4134                                }
4135                                else
4136                                        show_mode_change(p, 1);
4137                        }
4138                }
4139        }
4140}
4141
4142static void patch_stats(struct patch *patch)
4143{
4144        int lines = patch->lines_added + patch->lines_deleted;
4145
4146        if (lines > max_change)
4147                max_change = lines;
4148        if (patch->old_name) {
4149                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4150                if (!len)
4151                        len = strlen(patch->old_name);
4152                if (len > max_len)
4153                        max_len = len;
4154        }
4155        if (patch->new_name) {
4156                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4157                if (!len)
4158                        len = strlen(patch->new_name);
4159                if (len > max_len)
4160                        max_len = len;
4161        }
4162}
4163
4164static void remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4165{
4166        if (state->update_index) {
4167                if (remove_file_from_cache(patch->old_name) < 0)
4168                        die(_("unable to remove %s from index"), patch->old_name);
4169        }
4170        if (!state->cached) {
4171                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4172                        remove_path(patch->old_name);
4173                }
4174        }
4175}
4176
4177static void add_index_file(struct apply_state *state,
4178                           const char *path,
4179                           unsigned mode,
4180                           void *buf,
4181                           unsigned long size)
4182{
4183        struct stat st;
4184        struct cache_entry *ce;
4185        int namelen = strlen(path);
4186        unsigned ce_size = cache_entry_size(namelen);
4187
4188        if (!state->update_index)
4189                return;
4190
4191        ce = xcalloc(1, ce_size);
4192        memcpy(ce->name, path, namelen);
4193        ce->ce_mode = create_ce_mode(mode);
4194        ce->ce_flags = create_ce_flags(0);
4195        ce->ce_namelen = namelen;
4196        if (S_ISGITLINK(mode)) {
4197                const char *s;
4198
4199                if (!skip_prefix(buf, "Subproject commit ", &s) ||
4200                    get_sha1_hex(s, ce->sha1))
4201                        die(_("corrupt patch for submodule %s"), path);
4202        } else {
4203                if (!state->cached) {
4204                        if (lstat(path, &st) < 0)
4205                                die_errno(_("unable to stat newly created file '%s'"),
4206                                          path);
4207                        fill_stat_cache_info(ce, &st);
4208                }
4209                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
4210                        die(_("unable to create backing store for newly created file %s"), path);
4211        }
4212        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4213                die(_("unable to add cache entry for %s"), path);
4214}
4215
4216static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4217{
4218        int fd;
4219        struct strbuf nbuf = STRBUF_INIT;
4220
4221        if (S_ISGITLINK(mode)) {
4222                struct stat st;
4223                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4224                        return 0;
4225                return mkdir(path, 0777);
4226        }
4227
4228        if (has_symlinks && S_ISLNK(mode))
4229                /* Although buf:size is counted string, it also is NUL
4230                 * terminated.
4231                 */
4232                return symlink(buf, path);
4233
4234        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4235        if (fd < 0)
4236                return -1;
4237
4238        if (convert_to_working_tree(path, buf, size, &nbuf)) {
4239                size = nbuf.len;
4240                buf  = nbuf.buf;
4241        }
4242        write_or_die(fd, buf, size);
4243        strbuf_release(&nbuf);
4244
4245        if (close(fd) < 0)
4246                die_errno(_("closing file '%s'"), path);
4247        return 0;
4248}
4249
4250/*
4251 * We optimistically assume that the directories exist,
4252 * which is true 99% of the time anyway. If they don't,
4253 * we create them and try again.
4254 */
4255static void create_one_file(struct apply_state *state,
4256                            char *path,
4257                            unsigned mode,
4258                            const char *buf,
4259                            unsigned long size)
4260{
4261        if (state->cached)
4262                return;
4263        if (!try_create_file(path, mode, buf, size))
4264                return;
4265
4266        if (errno == ENOENT) {
4267                if (safe_create_leading_directories(path))
4268                        return;
4269                if (!try_create_file(path, mode, buf, size))
4270                        return;
4271        }
4272
4273        if (errno == EEXIST || errno == EACCES) {
4274                /* We may be trying to create a file where a directory
4275                 * used to be.
4276                 */
4277                struct stat st;
4278                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4279                        errno = EEXIST;
4280        }
4281
4282        if (errno == EEXIST) {
4283                unsigned int nr = getpid();
4284
4285                for (;;) {
4286                        char newpath[PATH_MAX];
4287                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4288                        if (!try_create_file(newpath, mode, buf, size)) {
4289                                if (!rename(newpath, path))
4290                                        return;
4291                                unlink_or_warn(newpath);
4292                                break;
4293                        }
4294                        if (errno != EEXIST)
4295                                break;
4296                        ++nr;
4297                }
4298        }
4299        die_errno(_("unable to write file '%s' mode %o"), path, mode);
4300}
4301
4302static void add_conflicted_stages_file(struct apply_state *state,
4303                                       struct patch *patch)
4304{
4305        int stage, namelen;
4306        unsigned ce_size, mode;
4307        struct cache_entry *ce;
4308
4309        if (!state->update_index)
4310                return;
4311        namelen = strlen(patch->new_name);
4312        ce_size = cache_entry_size(namelen);
4313        mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4314
4315        remove_file_from_cache(patch->new_name);
4316        for (stage = 1; stage < 4; stage++) {
4317                if (is_null_oid(&patch->threeway_stage[stage - 1]))
4318                        continue;
4319                ce = xcalloc(1, ce_size);
4320                memcpy(ce->name, patch->new_name, namelen);
4321                ce->ce_mode = create_ce_mode(mode);
4322                ce->ce_flags = create_ce_flags(stage);
4323                ce->ce_namelen = namelen;
4324                hashcpy(ce->sha1, patch->threeway_stage[stage - 1].hash);
4325                if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4326                        die(_("unable to add cache entry for %s"), patch->new_name);
4327        }
4328}
4329
4330static void create_file(struct apply_state *state, struct patch *patch)
4331{
4332        char *path = patch->new_name;
4333        unsigned mode = patch->new_mode;
4334        unsigned long size = patch->resultsize;
4335        char *buf = patch->result;
4336
4337        if (!mode)
4338                mode = S_IFREG | 0644;
4339        create_one_file(state, path, mode, buf, size);
4340
4341        if (patch->conflicted_threeway)
4342                add_conflicted_stages_file(state, patch);
4343        else
4344                add_index_file(state, path, mode, buf, size);
4345}
4346
4347/* phase zero is to remove, phase one is to create */
4348static void write_out_one_result(struct apply_state *state,
4349                                 struct patch *patch,
4350                                 int phase)
4351{
4352        if (patch->is_delete > 0) {
4353                if (phase == 0)
4354                        remove_file(state, patch, 1);
4355                return;
4356        }
4357        if (patch->is_new > 0 || patch->is_copy) {
4358                if (phase == 1)
4359                        create_file(state, patch);
4360                return;
4361        }
4362        /*
4363         * Rename or modification boils down to the same
4364         * thing: remove the old, write the new
4365         */
4366        if (phase == 0)
4367                remove_file(state, patch, patch->is_rename);
4368        if (phase == 1)
4369                create_file(state, patch);
4370}
4371
4372static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4373{
4374        FILE *rej;
4375        char namebuf[PATH_MAX];
4376        struct fragment *frag;
4377        int cnt = 0;
4378        struct strbuf sb = STRBUF_INIT;
4379
4380        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4381                if (!frag->rejected)
4382                        continue;
4383                cnt++;
4384        }
4385
4386        if (!cnt) {
4387                if (state->apply_verbosely)
4388                        say_patch_name(stderr,
4389                                       _("Applied patch %s cleanly."), patch);
4390                return 0;
4391        }
4392
4393        /* This should not happen, because a removal patch that leaves
4394         * contents are marked "rejected" at the patch level.
4395         */
4396        if (!patch->new_name)
4397                die(_("internal error"));
4398
4399        /* Say this even without --verbose */
4400        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4401                            "Applying patch %%s with %d rejects...",
4402                            cnt),
4403                    cnt);
4404        say_patch_name(stderr, sb.buf, patch);
4405        strbuf_release(&sb);
4406
4407        cnt = strlen(patch->new_name);
4408        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4409                cnt = ARRAY_SIZE(namebuf) - 5;
4410                warning(_("truncating .rej filename to %.*s.rej"),
4411                        cnt - 1, patch->new_name);
4412        }
4413        memcpy(namebuf, patch->new_name, cnt);
4414        memcpy(namebuf + cnt, ".rej", 5);
4415
4416        rej = fopen(namebuf, "w");
4417        if (!rej)
4418                return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4419
4420        /* Normal git tools never deal with .rej, so do not pretend
4421         * this is a git patch by saying --git or giving extended
4422         * headers.  While at it, maybe please "kompare" that wants
4423         * the trailing TAB and some garbage at the end of line ;-).
4424         */
4425        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4426                patch->new_name, patch->new_name);
4427        for (cnt = 1, frag = patch->fragments;
4428             frag;
4429             cnt++, frag = frag->next) {
4430                if (!frag->rejected) {
4431                        fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4432                        continue;
4433                }
4434                fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4435                fprintf(rej, "%.*s", frag->size, frag->patch);
4436                if (frag->patch[frag->size-1] != '\n')
4437                        fputc('\n', rej);
4438        }
4439        fclose(rej);
4440        return -1;
4441}
4442
4443static int write_out_results(struct apply_state *state, struct patch *list)
4444{
4445        int phase;
4446        int errs = 0;
4447        struct patch *l;
4448        struct string_list cpath = STRING_LIST_INIT_DUP;
4449
4450        for (phase = 0; phase < 2; phase++) {
4451                l = list;
4452                while (l) {
4453                        if (l->rejected)
4454                                errs = 1;
4455                        else {
4456                                write_out_one_result(state, l, phase);
4457                                if (phase == 1) {
4458                                        if (write_out_one_reject(state, l))
4459                                                errs = 1;
4460                                        if (l->conflicted_threeway) {
4461                                                string_list_append(&cpath, l->new_name);
4462                                                errs = 1;
4463                                        }
4464                                }
4465                        }
4466                        l = l->next;
4467                }
4468        }
4469
4470        if (cpath.nr) {
4471                struct string_list_item *item;
4472
4473                string_list_sort(&cpath);
4474                for_each_string_list_item(item, &cpath)
4475                        fprintf(stderr, "U %s\n", item->string);
4476                string_list_clear(&cpath, 0);
4477
4478                rerere(0);
4479        }
4480
4481        return errs;
4482}
4483
4484static struct lock_file lock_file;
4485
4486#define INACCURATE_EOF  (1<<0)
4487#define RECOUNT         (1<<1)
4488
4489static int apply_patch(struct apply_state *state,
4490                       int fd,
4491                       const char *filename,
4492                       int options)
4493{
4494        size_t offset;
4495        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4496        struct patch *list = NULL, **listp = &list;
4497        int skipped_patch = 0;
4498
4499        state->patch_input_file = filename;
4500        read_patch_file(&buf, fd);
4501        offset = 0;
4502        while (offset < buf.len) {
4503                struct patch *patch;
4504                int nr;
4505
4506                patch = xcalloc(1, sizeof(*patch));
4507                patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4508                patch->recount =  !!(options & RECOUNT);
4509                nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4510                if (nr < 0) {
4511                        free_patch(patch);
4512                        break;
4513                }
4514                if (state->apply_in_reverse)
4515                        reverse_patches(patch);
4516                if (use_patch(state, patch)) {
4517                        patch_stats(patch);
4518                        *listp = patch;
4519                        listp = &patch->next;
4520                }
4521                else {
4522                        if (state->apply_verbosely)
4523                                say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4524                        free_patch(patch);
4525                        skipped_patch++;
4526                }
4527                offset += nr;
4528        }
4529
4530        if (!list && !skipped_patch)
4531                die(_("unrecognized input"));
4532
4533        if (state->whitespace_error && (ws_error_action == die_on_ws_error))
4534                state->apply = 0;
4535
4536        state->update_index = state->check_index && state->apply;
4537        if (state->update_index && newfd < 0)
4538                newfd = hold_locked_index(&lock_file, 1);
4539
4540        if (state->check_index) {
4541                if (read_cache() < 0)
4542                        die(_("unable to read index file"));
4543        }
4544
4545        if ((state->check || state->apply) &&
4546            check_patch_list(state, list) < 0 &&
4547            !state->apply_with_reject)
4548                exit(1);
4549
4550        if (state->apply && write_out_results(state, list)) {
4551                if (state->apply_with_reject)
4552                        exit(1);
4553                /* with --3way, we still need to write the index out */
4554                return 1;
4555        }
4556
4557        if (state->fake_ancestor)
4558                build_fake_ancestor(list, state->fake_ancestor);
4559
4560        if (state->diffstat)
4561                stat_patch_list(list);
4562
4563        if (state->numstat)
4564                numstat_patch_list(state, list);
4565
4566        if (state->summary)
4567                summary_patch_list(list);
4568
4569        free_patch_list(list);
4570        strbuf_release(&buf);
4571        string_list_clear(&fn_table, 0);
4572        return 0;
4573}
4574
4575static void git_apply_config(void)
4576{
4577        git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
4578        git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
4579        git_config(git_default_config, NULL);
4580}
4581
4582static int option_parse_exclude(const struct option *opt,
4583                                const char *arg, int unset)
4584{
4585        struct apply_state *state = opt->value;
4586        add_name_limit(state, arg, 1);
4587        return 0;
4588}
4589
4590static int option_parse_include(const struct option *opt,
4591                                const char *arg, int unset)
4592{
4593        struct apply_state *state = opt->value;
4594        add_name_limit(state, arg, 0);
4595        state->has_include = 1;
4596        return 0;
4597}
4598
4599static int option_parse_p(const struct option *opt,
4600                          const char *arg,
4601                          int unset)
4602{
4603        struct apply_state *state = opt->value;
4604        state->p_value = atoi(arg);
4605        state->p_value_known = 1;
4606        return 0;
4607}
4608
4609static int option_parse_space_change(const struct option *opt,
4610                          const char *arg, int unset)
4611{
4612        if (unset)
4613                ws_ignore_action = ignore_ws_none;
4614        else
4615                ws_ignore_action = ignore_ws_change;
4616        return 0;
4617}
4618
4619static int option_parse_whitespace(const struct option *opt,
4620                                   const char *arg, int unset)
4621{
4622        struct apply_state *state = opt->value;
4623        state->whitespace_option = arg;
4624        parse_whitespace_option(state, arg);
4625        return 0;
4626}
4627
4628static int option_parse_directory(const struct option *opt,
4629                                  const char *arg, int unset)
4630{
4631        struct apply_state *state = opt->value;
4632        strbuf_reset(&state->root);
4633        strbuf_addstr(&state->root, arg);
4634        strbuf_complete(&state->root, '/');
4635        return 0;
4636}
4637
4638static void init_apply_state(struct apply_state *state, const char *prefix)
4639{
4640        memset(state, 0, sizeof(*state));
4641        state->prefix = prefix;
4642        state->prefix_length = state->prefix ? strlen(state->prefix) : 0;
4643        state->apply = 1;
4644        state->line_termination = '\n';
4645        state->p_value = 1;
4646        state->p_context = UINT_MAX;
4647        state->squelch_whitespace_errors = 5;
4648        strbuf_init(&state->root, 0);
4649
4650        git_apply_config();
4651        if (apply_default_whitespace)
4652                parse_whitespace_option(state, apply_default_whitespace);
4653        if (apply_default_ignorewhitespace)
4654                parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4655}
4656
4657static void clear_apply_state(struct apply_state *state)
4658{
4659        string_list_clear(&state->limit_by_name, 0);
4660        strbuf_release(&state->root);
4661}
4662
4663int cmd_apply(int argc, const char **argv, const char *prefix)
4664{
4665        int i;
4666        int errs = 0;
4667        int is_not_gitdir = !startup_info->have_repository;
4668        int force_apply = 0;
4669        int options = 0;
4670        int read_stdin = 1;
4671        struct apply_state state;
4672
4673        struct option builtin_apply_options[] = {
4674                { OPTION_CALLBACK, 0, "exclude", &state, N_("path"),
4675                        N_("don't apply changes matching the given path"),
4676                        0, option_parse_exclude },
4677                { OPTION_CALLBACK, 0, "include", &state, N_("path"),
4678                        N_("apply changes matching the given path"),
4679                        0, option_parse_include },
4680                { OPTION_CALLBACK, 'p', NULL, &state, N_("num"),
4681                        N_("remove <num> leading slashes from traditional diff paths"),
4682                        0, option_parse_p },
4683                OPT_BOOL(0, "no-add", &state.no_add,
4684                        N_("ignore additions made by the patch")),
4685                OPT_BOOL(0, "stat", &state.diffstat,
4686                        N_("instead of applying the patch, output diffstat for the input")),
4687                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4688                OPT_NOOP_NOARG(0, "binary"),
4689                OPT_BOOL(0, "numstat", &state.numstat,
4690                        N_("show number of added and deleted lines in decimal notation")),
4691                OPT_BOOL(0, "summary", &state.summary,
4692                        N_("instead of applying the patch, output a summary for the input")),
4693                OPT_BOOL(0, "check", &state.check,
4694                        N_("instead of applying the patch, see if the patch is applicable")),
4695                OPT_BOOL(0, "index", &state.check_index,
4696                        N_("make sure the patch is applicable to the current index")),
4697                OPT_BOOL(0, "cached", &state.cached,
4698                        N_("apply a patch without touching the working tree")),
4699                OPT_BOOL(0, "unsafe-paths", &state.unsafe_paths,
4700                        N_("accept a patch that touches outside the working area")),
4701                OPT_BOOL(0, "apply", &force_apply,
4702                        N_("also apply the patch (use with --stat/--summary/--check)")),
4703                OPT_BOOL('3', "3way", &state.threeway,
4704                         N_( "attempt three-way merge if a patch does not apply")),
4705                OPT_FILENAME(0, "build-fake-ancestor", &state.fake_ancestor,
4706                        N_("build a temporary index based on embedded index information")),
4707                /* Think twice before adding "--nul" synonym to this */
4708                OPT_SET_INT('z', NULL, &state.line_termination,
4709                        N_("paths are separated with NUL character"), '\0'),
4710                OPT_INTEGER('C', NULL, &state.p_context,
4711                                N_("ensure at least <n> lines of context match")),
4712                { OPTION_CALLBACK, 0, "whitespace", &state, N_("action"),
4713                        N_("detect new or modified lines that have whitespace errors"),
4714                        0, option_parse_whitespace },
4715                { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4716                        N_("ignore changes in whitespace when finding context"),
4717                        PARSE_OPT_NOARG, option_parse_space_change },
4718                { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4719                        N_("ignore changes in whitespace when finding context"),
4720                        PARSE_OPT_NOARG, option_parse_space_change },
4721                OPT_BOOL('R', "reverse", &state.apply_in_reverse,
4722                        N_("apply the patch in reverse")),
4723                OPT_BOOL(0, "unidiff-zero", &state.unidiff_zero,
4724                        N_("don't expect at least one line of context")),
4725                OPT_BOOL(0, "reject", &state.apply_with_reject,
4726                        N_("leave the rejected hunks in corresponding *.rej files")),
4727                OPT_BOOL(0, "allow-overlap", &state.allow_overlap,
4728                        N_("allow overlapping hunks")),
4729                OPT__VERBOSE(&state.apply_verbosely, N_("be verbose")),
4730                OPT_BIT(0, "inaccurate-eof", &options,
4731                        N_("tolerate incorrectly detected missing new-line at the end of file"),
4732                        INACCURATE_EOF),
4733                OPT_BIT(0, "recount", &options,
4734                        N_("do not trust the line counts in the hunk headers"),
4735                        RECOUNT),
4736                { OPTION_CALLBACK, 0, "directory", &state, N_("root"),
4737                        N_("prepend <root> to all filenames"),
4738                        0, option_parse_directory },
4739                OPT_END()
4740        };
4741
4742        init_apply_state(&state, prefix);
4743
4744        argc = parse_options(argc, argv, state.prefix, builtin_apply_options,
4745                        apply_usage, 0);
4746
4747        if (state.apply_with_reject && state.threeway)
4748                die("--reject and --3way cannot be used together.");
4749        if (state.cached && state.threeway)
4750                die("--cached and --3way cannot be used together.");
4751        if (state.threeway) {
4752                if (is_not_gitdir)
4753                        die(_("--3way outside a repository"));
4754                state.check_index = 1;
4755        }
4756        if (state.apply_with_reject)
4757                state.apply = state.apply_verbosely = 1;
4758        if (!force_apply && (state.diffstat || state.numstat || state.summary || state.check || state.fake_ancestor))
4759                state.apply = 0;
4760        if (state.check_index && is_not_gitdir)
4761                die(_("--index outside a repository"));
4762        if (state.cached) {
4763                if (is_not_gitdir)
4764                        die(_("--cached outside a repository"));
4765                state.check_index = 1;
4766        }
4767        if (state.check_index)
4768                state.unsafe_paths = 0;
4769
4770        for (i = 0; i < argc; i++) {
4771                const char *arg = argv[i];
4772                int fd;
4773
4774                if (!strcmp(arg, "-")) {
4775                        errs |= apply_patch(&state, 0, "<stdin>", options);
4776                        read_stdin = 0;
4777                        continue;
4778                } else if (0 < state.prefix_length)
4779                        arg = prefix_filename(state.prefix,
4780                                              state.prefix_length,
4781                                              arg);
4782
4783                fd = open(arg, O_RDONLY);
4784                if (fd < 0)
4785                        die_errno(_("can't open patch '%s'"), arg);
4786                read_stdin = 0;
4787                set_default_whitespace_mode(&state);
4788                errs |= apply_patch(&state, fd, arg, options);
4789                close(fd);
4790        }
4791        set_default_whitespace_mode(&state);
4792        if (read_stdin)
4793                errs |= apply_patch(&state, 0, "<stdin>", options);
4794        if (state.whitespace_error) {
4795                if (state.squelch_whitespace_errors &&
4796                    state.squelch_whitespace_errors < state.whitespace_error) {
4797                        int squelched =
4798                                state.whitespace_error - state.squelch_whitespace_errors;
4799                        warning(Q_("squelched %d whitespace error",
4800                                   "squelched %d whitespace errors",
4801                                   squelched),
4802                                squelched);
4803                }
4804                if (ws_error_action == die_on_ws_error)
4805                        die(Q_("%d line adds whitespace errors.",
4806                               "%d lines add whitespace errors.",
4807                               state.whitespace_error),
4808                            state.whitespace_error);
4809                if (applied_after_fixing_ws && state.apply)
4810                        warning("%d line%s applied after"
4811                                " fixing whitespace errors.",
4812                                applied_after_fixing_ws,
4813                                applied_after_fixing_ws == 1 ? "" : "s");
4814                else if (state.whitespace_error)
4815                        warning(Q_("%d line adds whitespace errors.",
4816                                   "%d lines add whitespace errors.",
4817                                   state.whitespace_error),
4818                                state.whitespace_error);
4819        }
4820
4821        if (state.update_index) {
4822                if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))
4823                        die(_("Unable to write new index file"));
4824        }
4825
4826        clear_apply_state(&state);
4827
4828        return !!errs;
4829}