builtin / apply.con commit Merge branch 'tb/connect-ipv6-parse-fix' into maint (ad34ad6)
   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
  24/*
  25 *  --check turns on checking that the working tree matches the
  26 *    files that are being modified, but doesn't apply the patch
  27 *  --stat does just a diffstat, and doesn't actually apply
  28 *  --numstat does numeric diffstat, and doesn't actually apply
  29 *  --index-info shows the old and new index info for paths if available.
  30 *  --index updates the cache as well.
  31 *  --cached updates only the cache without ever touching the working tree.
  32 */
  33static const char *prefix;
  34static int prefix_length = -1;
  35static int newfd = -1;
  36
  37static int unidiff_zero;
  38static int p_value = 1;
  39static int p_value_known;
  40static int check_index;
  41static int update_index;
  42static int cached;
  43static int diffstat;
  44static int numstat;
  45static int summary;
  46static int check;
  47static int apply = 1;
  48static int apply_in_reverse;
  49static int apply_with_reject;
  50static int apply_verbosely;
  51static int allow_overlap;
  52static int no_add;
  53static int threeway;
  54static int unsafe_paths;
  55static const char *fake_ancestor;
  56static int line_termination = '\n';
  57static unsigned int p_context = UINT_MAX;
  58static const char * const apply_usage[] = {
  59        N_("git apply [options] [<patch>...]"),
  60        NULL
  61};
  62
  63static enum ws_error_action {
  64        nowarn_ws_error,
  65        warn_on_ws_error,
  66        die_on_ws_error,
  67        correct_ws_error
  68} ws_error_action = warn_on_ws_error;
  69static int whitespace_error;
  70static int squelch_whitespace_errors = 5;
  71static int applied_after_fixing_ws;
  72
  73static enum ws_ignore {
  74        ignore_ws_none,
  75        ignore_ws_change
  76} ws_ignore_action = ignore_ws_none;
  77
  78
  79static const char *patch_input_file;
  80static const char *root;
  81static int root_len;
  82static int read_stdin = 1;
  83static int options;
  84
  85static void parse_whitespace_option(const char *option)
  86{
  87        if (!option) {
  88                ws_error_action = warn_on_ws_error;
  89                return;
  90        }
  91        if (!strcmp(option, "warn")) {
  92                ws_error_action = warn_on_ws_error;
  93                return;
  94        }
  95        if (!strcmp(option, "nowarn")) {
  96                ws_error_action = nowarn_ws_error;
  97                return;
  98        }
  99        if (!strcmp(option, "error")) {
 100                ws_error_action = die_on_ws_error;
 101                return;
 102        }
 103        if (!strcmp(option, "error-all")) {
 104                ws_error_action = die_on_ws_error;
 105                squelch_whitespace_errors = 0;
 106                return;
 107        }
 108        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
 109                ws_error_action = correct_ws_error;
 110                return;
 111        }
 112        die(_("unrecognized whitespace option '%s'"), option);
 113}
 114
 115static void parse_ignorewhitespace_option(const char *option)
 116{
 117        if (!option || !strcmp(option, "no") ||
 118            !strcmp(option, "false") || !strcmp(option, "never") ||
 119            !strcmp(option, "none")) {
 120                ws_ignore_action = ignore_ws_none;
 121                return;
 122        }
 123        if (!strcmp(option, "change")) {
 124                ws_ignore_action = ignore_ws_change;
 125                return;
 126        }
 127        die(_("unrecognized whitespace ignore option '%s'"), option);
 128}
 129
 130static void set_default_whitespace_mode(const char *whitespace_option)
 131{
 132        if (!whitespace_option && !apply_default_whitespace)
 133                ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
 134}
 135
 136/*
 137 * For "diff-stat" like behaviour, we keep track of the biggest change
 138 * we've seen, and the longest filename. That allows us to do simple
 139 * scaling.
 140 */
 141static int max_change, max_len;
 142
 143/*
 144 * Various "current state", notably line numbers and what
 145 * file (and how) we're patching right now.. The "is_xxxx"
 146 * things are flags, where -1 means "don't know yet".
 147 */
 148static int linenr = 1;
 149
 150/*
 151 * This represents one "hunk" from a patch, starting with
 152 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 153 * patch text is pointed at by patch, and its byte length
 154 * is stored in size.  leading and trailing are the number
 155 * of context lines.
 156 */
 157struct fragment {
 158        unsigned long leading, trailing;
 159        unsigned long oldpos, oldlines;
 160        unsigned long newpos, newlines;
 161        /*
 162         * 'patch' is usually borrowed from buf in apply_patch(),
 163         * but some codepaths store an allocated buffer.
 164         */
 165        const char *patch;
 166        unsigned free_patch:1,
 167                rejected:1;
 168        int size;
 169        int linenr;
 170        struct fragment *next;
 171};
 172
 173/*
 174 * When dealing with a binary patch, we reuse "leading" field
 175 * to store the type of the binary hunk, either deflated "delta"
 176 * or deflated "literal".
 177 */
 178#define binary_patch_method leading
 179#define BINARY_DELTA_DEFLATED   1
 180#define BINARY_LITERAL_DEFLATED 2
 181
 182/*
 183 * This represents a "patch" to a file, both metainfo changes
 184 * such as creation/deletion, filemode and content changes represented
 185 * as a series of fragments.
 186 */
 187struct patch {
 188        char *new_name, *old_name, *def_name;
 189        unsigned int old_mode, new_mode;
 190        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 191        int rejected;
 192        unsigned ws_rule;
 193        int lines_added, lines_deleted;
 194        int score;
 195        unsigned int is_toplevel_relative:1;
 196        unsigned int inaccurate_eof:1;
 197        unsigned int is_binary:1;
 198        unsigned int is_copy:1;
 199        unsigned int is_rename:1;
 200        unsigned int recount:1;
 201        unsigned int conflicted_threeway:1;
 202        unsigned int direct_to_threeway:1;
 203        struct fragment *fragments;
 204        char *result;
 205        size_t resultsize;
 206        char old_sha1_prefix[41];
 207        char new_sha1_prefix[41];
 208        struct patch *next;
 209
 210        /* three-way fallback result */
 211        unsigned char threeway_stage[3][20];
 212};
 213
 214static void free_fragment_list(struct fragment *list)
 215{
 216        while (list) {
 217                struct fragment *next = list->next;
 218                if (list->free_patch)
 219                        free((char *)list->patch);
 220                free(list);
 221                list = next;
 222        }
 223}
 224
 225static void free_patch(struct patch *patch)
 226{
 227        free_fragment_list(patch->fragments);
 228        free(patch->def_name);
 229        free(patch->old_name);
 230        free(patch->new_name);
 231        free(patch->result);
 232        free(patch);
 233}
 234
 235static void free_patch_list(struct patch *list)
 236{
 237        while (list) {
 238                struct patch *next = list->next;
 239                free_patch(list);
 240                list = next;
 241        }
 242}
 243
 244/*
 245 * A line in a file, len-bytes long (includes the terminating LF,
 246 * except for an incomplete line at the end if the file ends with
 247 * one), and its contents hashes to 'hash'.
 248 */
 249struct line {
 250        size_t len;
 251        unsigned hash : 24;
 252        unsigned flag : 8;
 253#define LINE_COMMON     1
 254#define LINE_PATCHED    2
 255};
 256
 257/*
 258 * This represents a "file", which is an array of "lines".
 259 */
 260struct image {
 261        char *buf;
 262        size_t len;
 263        size_t nr;
 264        size_t alloc;
 265        struct line *line_allocated;
 266        struct line *line;
 267};
 268
 269/*
 270 * Records filenames that have been touched, in order to handle
 271 * the case where more than one patches touch the same file.
 272 */
 273
 274static struct string_list fn_table;
 275
 276static uint32_t hash_line(const char *cp, size_t len)
 277{
 278        size_t i;
 279        uint32_t h;
 280        for (i = 0, h = 0; i < len; i++) {
 281                if (!isspace(cp[i])) {
 282                        h = h * 3 + (cp[i] & 0xff);
 283                }
 284        }
 285        return h;
 286}
 287
 288/*
 289 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 290 * whitespace difference. Returns 1 if they match, 0 otherwise
 291 */
 292static int fuzzy_matchlines(const char *s1, size_t n1,
 293                            const char *s2, size_t n2)
 294{
 295        const char *last1 = s1 + n1 - 1;
 296        const char *last2 = s2 + n2 - 1;
 297        int result = 0;
 298
 299        /* ignore line endings */
 300        while ((*last1 == '\r') || (*last1 == '\n'))
 301                last1--;
 302        while ((*last2 == '\r') || (*last2 == '\n'))
 303                last2--;
 304
 305        /* skip leading whitespaces, if both begin with whitespace */
 306        if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
 307                while (isspace(*s1) && (s1 <= last1))
 308                        s1++;
 309                while (isspace(*s2) && (s2 <= last2))
 310                        s2++;
 311        }
 312        /* early return if both lines are empty */
 313        if ((s1 > last1) && (s2 > last2))
 314                return 1;
 315        while (!result) {
 316                result = *s1++ - *s2++;
 317                /*
 318                 * Skip whitespace inside. We check for whitespace on
 319                 * both buffers because we don't want "a b" to match
 320                 * "ab"
 321                 */
 322                if (isspace(*s1) && isspace(*s2)) {
 323                        while (isspace(*s1) && s1 <= last1)
 324                                s1++;
 325                        while (isspace(*s2) && s2 <= last2)
 326                                s2++;
 327                }
 328                /*
 329                 * If we reached the end on one side only,
 330                 * lines don't match
 331                 */
 332                if (
 333                    ((s2 > last2) && (s1 <= last1)) ||
 334                    ((s1 > last1) && (s2 <= last2)))
 335                        return 0;
 336                if ((s1 > last1) && (s2 > last2))
 337                        break;
 338        }
 339
 340        return !result;
 341}
 342
 343static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 344{
 345        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 346        img->line_allocated[img->nr].len = len;
 347        img->line_allocated[img->nr].hash = hash_line(bol, len);
 348        img->line_allocated[img->nr].flag = flag;
 349        img->nr++;
 350}
 351
 352/*
 353 * "buf" has the file contents to be patched (read from various sources).
 354 * attach it to "image" and add line-based index to it.
 355 * "image" now owns the "buf".
 356 */
 357static void prepare_image(struct image *image, char *buf, size_t len,
 358                          int prepare_linetable)
 359{
 360        const char *cp, *ep;
 361
 362        memset(image, 0, sizeof(*image));
 363        image->buf = buf;
 364        image->len = len;
 365
 366        if (!prepare_linetable)
 367                return;
 368
 369        ep = image->buf + image->len;
 370        cp = image->buf;
 371        while (cp < ep) {
 372                const char *next;
 373                for (next = cp; next < ep && *next != '\n'; next++)
 374                        ;
 375                if (next < ep)
 376                        next++;
 377                add_line_info(image, cp, next - cp, 0);
 378                cp = next;
 379        }
 380        image->line = image->line_allocated;
 381}
 382
 383static void clear_image(struct image *image)
 384{
 385        free(image->buf);
 386        free(image->line_allocated);
 387        memset(image, 0, sizeof(*image));
 388}
 389
 390/* fmt must contain _one_ %s and no other substitution */
 391static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 392{
 393        struct strbuf sb = STRBUF_INIT;
 394
 395        if (patch->old_name && patch->new_name &&
 396            strcmp(patch->old_name, patch->new_name)) {
 397                quote_c_style(patch->old_name, &sb, NULL, 0);
 398                strbuf_addstr(&sb, " => ");
 399                quote_c_style(patch->new_name, &sb, NULL, 0);
 400        } else {
 401                const char *n = patch->new_name;
 402                if (!n)
 403                        n = patch->old_name;
 404                quote_c_style(n, &sb, NULL, 0);
 405        }
 406        fprintf(output, fmt, sb.buf);
 407        fputc('\n', output);
 408        strbuf_release(&sb);
 409}
 410
 411#define SLOP (16)
 412
 413static void read_patch_file(struct strbuf *sb, int fd)
 414{
 415        if (strbuf_read(sb, fd, 0) < 0)
 416                die_errno("git apply: failed to read");
 417
 418        /*
 419         * Make sure that we have some slop in the buffer
 420         * so that we can do speculative "memcmp" etc, and
 421         * see to it that it is NUL-filled.
 422         */
 423        strbuf_grow(sb, SLOP);
 424        memset(sb->buf + sb->len, 0, SLOP);
 425}
 426
 427static unsigned long linelen(const char *buffer, unsigned long size)
 428{
 429        unsigned long len = 0;
 430        while (size--) {
 431                len++;
 432                if (*buffer++ == '\n')
 433                        break;
 434        }
 435        return len;
 436}
 437
 438static int is_dev_null(const char *str)
 439{
 440        return skip_prefix(str, "/dev/null", &str) && isspace(*str);
 441}
 442
 443#define TERM_SPACE      1
 444#define TERM_TAB        2
 445
 446static int name_terminate(const char *name, int namelen, int c, int terminate)
 447{
 448        if (c == ' ' && !(terminate & TERM_SPACE))
 449                return 0;
 450        if (c == '\t' && !(terminate & TERM_TAB))
 451                return 0;
 452
 453        return 1;
 454}
 455
 456/* remove double slashes to make --index work with such filenames */
 457static char *squash_slash(char *name)
 458{
 459        int i = 0, j = 0;
 460
 461        if (!name)
 462                return NULL;
 463
 464        while (name[i]) {
 465                if ((name[j++] = name[i++]) == '/')
 466                        while (name[i] == '/')
 467                                i++;
 468        }
 469        name[j] = '\0';
 470        return name;
 471}
 472
 473static char *find_name_gnu(const char *line, const char *def, int p_value)
 474{
 475        struct strbuf name = STRBUF_INIT;
 476        char *cp;
 477
 478        /*
 479         * Proposed "new-style" GNU patch/diff format; see
 480         * http://marc.info/?l=git&m=112927316408690&w=2
 481         */
 482        if (unquote_c_style(&name, line, NULL)) {
 483                strbuf_release(&name);
 484                return NULL;
 485        }
 486
 487        for (cp = name.buf; p_value; p_value--) {
 488                cp = strchr(cp, '/');
 489                if (!cp) {
 490                        strbuf_release(&name);
 491                        return NULL;
 492                }
 493                cp++;
 494        }
 495
 496        strbuf_remove(&name, 0, cp - name.buf);
 497        if (root)
 498                strbuf_insert(&name, 0, root, root_len);
 499        return squash_slash(strbuf_detach(&name, NULL));
 500}
 501
 502static size_t sane_tz_len(const char *line, size_t len)
 503{
 504        const char *tz, *p;
 505
 506        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 507                return 0;
 508        tz = line + len - strlen(" +0500");
 509
 510        if (tz[1] != '+' && tz[1] != '-')
 511                return 0;
 512
 513        for (p = tz + 2; p != line + len; p++)
 514                if (!isdigit(*p))
 515                        return 0;
 516
 517        return line + len - tz;
 518}
 519
 520static size_t tz_with_colon_len(const char *line, size_t len)
 521{
 522        const char *tz, *p;
 523
 524        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 525                return 0;
 526        tz = line + len - strlen(" +08:00");
 527
 528        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 529                return 0;
 530        p = tz + 2;
 531        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 532            !isdigit(*p++) || !isdigit(*p++))
 533                return 0;
 534
 535        return line + len - tz;
 536}
 537
 538static size_t date_len(const char *line, size_t len)
 539{
 540        const char *date, *p;
 541
 542        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 543                return 0;
 544        p = date = line + len - strlen("72-02-05");
 545
 546        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 547            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 548            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 549                return 0;
 550
 551        if (date - line >= strlen("19") &&
 552            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 553                date -= strlen("19");
 554
 555        return line + len - date;
 556}
 557
 558static size_t short_time_len(const char *line, size_t len)
 559{
 560        const char *time, *p;
 561
 562        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 563                return 0;
 564        p = time = line + len - strlen(" 07:01:32");
 565
 566        /* Permit 1-digit hours? */
 567        if (*p++ != ' ' ||
 568            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 569            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 570            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 571                return 0;
 572
 573        return line + len - time;
 574}
 575
 576static size_t fractional_time_len(const char *line, size_t len)
 577{
 578        const char *p;
 579        size_t n;
 580
 581        /* Expected format: 19:41:17.620000023 */
 582        if (!len || !isdigit(line[len - 1]))
 583                return 0;
 584        p = line + len - 1;
 585
 586        /* Fractional seconds. */
 587        while (p > line && isdigit(*p))
 588                p--;
 589        if (*p != '.')
 590                return 0;
 591
 592        /* Hours, minutes, and whole seconds. */
 593        n = short_time_len(line, p - line);
 594        if (!n)
 595                return 0;
 596
 597        return line + len - p + n;
 598}
 599
 600static size_t trailing_spaces_len(const char *line, size_t len)
 601{
 602        const char *p;
 603
 604        /* Expected format: ' ' x (1 or more)  */
 605        if (!len || line[len - 1] != ' ')
 606                return 0;
 607
 608        p = line + len;
 609        while (p != line) {
 610                p--;
 611                if (*p != ' ')
 612                        return line + len - (p + 1);
 613        }
 614
 615        /* All spaces! */
 616        return len;
 617}
 618
 619static size_t diff_timestamp_len(const char *line, size_t len)
 620{
 621        const char *end = line + len;
 622        size_t n;
 623
 624        /*
 625         * Posix: 2010-07-05 19:41:17
 626         * GNU: 2010-07-05 19:41:17.620000023 -0500
 627         */
 628
 629        if (!isdigit(end[-1]))
 630                return 0;
 631
 632        n = sane_tz_len(line, end - line);
 633        if (!n)
 634                n = tz_with_colon_len(line, end - line);
 635        end -= n;
 636
 637        n = short_time_len(line, end - line);
 638        if (!n)
 639                n = fractional_time_len(line, end - line);
 640        end -= n;
 641
 642        n = date_len(line, end - line);
 643        if (!n) /* No date.  Too bad. */
 644                return 0;
 645        end -= n;
 646
 647        if (end == line)        /* No space before date. */
 648                return 0;
 649        if (end[-1] == '\t') {  /* Success! */
 650                end--;
 651                return line + len - end;
 652        }
 653        if (end[-1] != ' ')     /* No space before date. */
 654                return 0;
 655
 656        /* Whitespace damage. */
 657        end -= trailing_spaces_len(line, end - line);
 658        return line + len - end;
 659}
 660
 661static char *find_name_common(const char *line, const char *def,
 662                              int p_value, const char *end, int terminate)
 663{
 664        int len;
 665        const char *start = NULL;
 666
 667        if (p_value == 0)
 668                start = line;
 669        while (line != end) {
 670                char c = *line;
 671
 672                if (!end && isspace(c)) {
 673                        if (c == '\n')
 674                                break;
 675                        if (name_terminate(start, line-start, c, terminate))
 676                                break;
 677                }
 678                line++;
 679                if (c == '/' && !--p_value)
 680                        start = line;
 681        }
 682        if (!start)
 683                return squash_slash(xstrdup_or_null(def));
 684        len = line - start;
 685        if (!len)
 686                return squash_slash(xstrdup_or_null(def));
 687
 688        /*
 689         * Generally we prefer the shorter name, especially
 690         * if the other one is just a variation of that with
 691         * something else tacked on to the end (ie "file.orig"
 692         * or "file~").
 693         */
 694        if (def) {
 695                int deflen = strlen(def);
 696                if (deflen < len && !strncmp(start, def, deflen))
 697                        return squash_slash(xstrdup(def));
 698        }
 699
 700        if (root) {
 701                char *ret = xmalloc(root_len + len + 1);
 702                strcpy(ret, root);
 703                memcpy(ret + root_len, start, len);
 704                ret[root_len + len] = '\0';
 705                return squash_slash(ret);
 706        }
 707
 708        return squash_slash(xmemdupz(start, len));
 709}
 710
 711static char *find_name(const char *line, char *def, int p_value, int terminate)
 712{
 713        if (*line == '"') {
 714                char *name = find_name_gnu(line, def, p_value);
 715                if (name)
 716                        return name;
 717        }
 718
 719        return find_name_common(line, def, p_value, NULL, terminate);
 720}
 721
 722static char *find_name_traditional(const char *line, char *def, int p_value)
 723{
 724        size_t len;
 725        size_t date_len;
 726
 727        if (*line == '"') {
 728                char *name = find_name_gnu(line, def, p_value);
 729                if (name)
 730                        return name;
 731        }
 732
 733        len = strchrnul(line, '\n') - line;
 734        date_len = diff_timestamp_len(line, len);
 735        if (!date_len)
 736                return find_name_common(line, def, p_value, NULL, TERM_TAB);
 737        len -= date_len;
 738
 739        return find_name_common(line, def, p_value, line + len, 0);
 740}
 741
 742static int count_slashes(const char *cp)
 743{
 744        int cnt = 0;
 745        char ch;
 746
 747        while ((ch = *cp++))
 748                if (ch == '/')
 749                        cnt++;
 750        return cnt;
 751}
 752
 753/*
 754 * Given the string after "--- " or "+++ ", guess the appropriate
 755 * p_value for the given patch.
 756 */
 757static int guess_p_value(const char *nameline)
 758{
 759        char *name, *cp;
 760        int val = -1;
 761
 762        if (is_dev_null(nameline))
 763                return -1;
 764        name = find_name_traditional(nameline, NULL, 0);
 765        if (!name)
 766                return -1;
 767        cp = strchr(name, '/');
 768        if (!cp)
 769                val = 0;
 770        else if (prefix) {
 771                /*
 772                 * Does it begin with "a/$our-prefix" and such?  Then this is
 773                 * very likely to apply to our directory.
 774                 */
 775                if (!strncmp(name, prefix, prefix_length))
 776                        val = count_slashes(prefix);
 777                else {
 778                        cp++;
 779                        if (!strncmp(cp, prefix, prefix_length))
 780                                val = count_slashes(prefix) + 1;
 781                }
 782        }
 783        free(name);
 784        return val;
 785}
 786
 787/*
 788 * Does the ---/+++ line has the POSIX timestamp after the last HT?
 789 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 790 * this such a timestamp?
 791 */
 792static int has_epoch_timestamp(const char *nameline)
 793{
 794        /*
 795         * We are only interested in epoch timestamp; any non-zero
 796         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 797         * For the same reason, the date must be either 1969-12-31 or
 798         * 1970-01-01, and the seconds part must be "00".
 799         */
 800        const char stamp_regexp[] =
 801                "^(1969-12-31|1970-01-01)"
 802                " "
 803                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 804                " "
 805                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 806        const char *timestamp = NULL, *cp, *colon;
 807        static regex_t *stamp;
 808        regmatch_t m[10];
 809        int zoneoffset;
 810        int hourminute;
 811        int status;
 812
 813        for (cp = nameline; *cp != '\n'; cp++) {
 814                if (*cp == '\t')
 815                        timestamp = cp + 1;
 816        }
 817        if (!timestamp)
 818                return 0;
 819        if (!stamp) {
 820                stamp = xmalloc(sizeof(*stamp));
 821                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 822                        warning(_("Cannot prepare timestamp regexp %s"),
 823                                stamp_regexp);
 824                        return 0;
 825                }
 826        }
 827
 828        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 829        if (status) {
 830                if (status != REG_NOMATCH)
 831                        warning(_("regexec returned %d for input: %s"),
 832                                status, timestamp);
 833                return 0;
 834        }
 835
 836        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 837        if (*colon == ':')
 838                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 839        else
 840                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 841        if (timestamp[m[3].rm_so] == '-')
 842                zoneoffset = -zoneoffset;
 843
 844        /*
 845         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 846         * (west of GMT) or 1970-01-01 (east of GMT)
 847         */
 848        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 849            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 850                return 0;
 851
 852        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 853                      strtol(timestamp + 14, NULL, 10) -
 854                      zoneoffset);
 855
 856        return ((zoneoffset < 0 && hourminute == 1440) ||
 857                (0 <= zoneoffset && !hourminute));
 858}
 859
 860/*
 861 * Get the name etc info from the ---/+++ lines of a traditional patch header
 862 *
 863 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 864 * files, we can happily check the index for a match, but for creating a
 865 * new file we should try to match whatever "patch" does. I have no idea.
 866 */
 867static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 868{
 869        char *name;
 870
 871        first += 4;     /* skip "--- " */
 872        second += 4;    /* skip "+++ " */
 873        if (!p_value_known) {
 874                int p, q;
 875                p = guess_p_value(first);
 876                q = guess_p_value(second);
 877                if (p < 0) p = q;
 878                if (0 <= p && p == q) {
 879                        p_value = p;
 880                        p_value_known = 1;
 881                }
 882        }
 883        if (is_dev_null(first)) {
 884                patch->is_new = 1;
 885                patch->is_delete = 0;
 886                name = find_name_traditional(second, NULL, p_value);
 887                patch->new_name = name;
 888        } else if (is_dev_null(second)) {
 889                patch->is_new = 0;
 890                patch->is_delete = 1;
 891                name = find_name_traditional(first, NULL, p_value);
 892                patch->old_name = name;
 893        } else {
 894                char *first_name;
 895                first_name = find_name_traditional(first, NULL, p_value);
 896                name = find_name_traditional(second, first_name, p_value);
 897                free(first_name);
 898                if (has_epoch_timestamp(first)) {
 899                        patch->is_new = 1;
 900                        patch->is_delete = 0;
 901                        patch->new_name = name;
 902                } else if (has_epoch_timestamp(second)) {
 903                        patch->is_new = 0;
 904                        patch->is_delete = 1;
 905                        patch->old_name = name;
 906                } else {
 907                        patch->old_name = name;
 908                        patch->new_name = xstrdup_or_null(name);
 909                }
 910        }
 911        if (!name)
 912                die(_("unable to find filename in patch at line %d"), linenr);
 913}
 914
 915static int gitdiff_hdrend(const char *line, struct patch *patch)
 916{
 917        return -1;
 918}
 919
 920/*
 921 * We're anal about diff header consistency, to make
 922 * sure that we don't end up having strange ambiguous
 923 * patches floating around.
 924 *
 925 * As a result, gitdiff_{old|new}name() will check
 926 * their names against any previous information, just
 927 * to make sure..
 928 */
 929#define DIFF_OLD_NAME 0
 930#define DIFF_NEW_NAME 1
 931
 932static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
 933{
 934        if (!orig_name && !isnull)
 935                return find_name(line, NULL, p_value, TERM_TAB);
 936
 937        if (orig_name) {
 938                int len;
 939                const char *name;
 940                char *another;
 941                name = orig_name;
 942                len = strlen(name);
 943                if (isnull)
 944                        die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
 945                another = find_name(line, NULL, p_value, TERM_TAB);
 946                if (!another || memcmp(another, name, len + 1))
 947                        die((side == DIFF_NEW_NAME) ?
 948                            _("git apply: bad git-diff - inconsistent new filename on line %d") :
 949                            _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
 950                free(another);
 951                return orig_name;
 952        }
 953        else {
 954                /* expect "/dev/null" */
 955                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 956                        die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
 957                return NULL;
 958        }
 959}
 960
 961static int gitdiff_oldname(const char *line, struct patch *patch)
 962{
 963        char *orig = patch->old_name;
 964        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
 965                                              DIFF_OLD_NAME);
 966        if (orig != patch->old_name)
 967                free(orig);
 968        return 0;
 969}
 970
 971static int gitdiff_newname(const char *line, struct patch *patch)
 972{
 973        char *orig = patch->new_name;
 974        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
 975                                              DIFF_NEW_NAME);
 976        if (orig != patch->new_name)
 977                free(orig);
 978        return 0;
 979}
 980
 981static int gitdiff_oldmode(const char *line, struct patch *patch)
 982{
 983        patch->old_mode = strtoul(line, NULL, 8);
 984        return 0;
 985}
 986
 987static int gitdiff_newmode(const char *line, struct patch *patch)
 988{
 989        patch->new_mode = strtoul(line, NULL, 8);
 990        return 0;
 991}
 992
 993static int gitdiff_delete(const char *line, struct patch *patch)
 994{
 995        patch->is_delete = 1;
 996        free(patch->old_name);
 997        patch->old_name = xstrdup_or_null(patch->def_name);
 998        return gitdiff_oldmode(line, patch);
 999}
1000
1001static int gitdiff_newfile(const char *line, struct patch *patch)
1002{
1003        patch->is_new = 1;
1004        free(patch->new_name);
1005        patch->new_name = xstrdup_or_null(patch->def_name);
1006        return gitdiff_newmode(line, patch);
1007}
1008
1009static int gitdiff_copysrc(const char *line, struct patch *patch)
1010{
1011        patch->is_copy = 1;
1012        free(patch->old_name);
1013        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1014        return 0;
1015}
1016
1017static int gitdiff_copydst(const char *line, struct patch *patch)
1018{
1019        patch->is_copy = 1;
1020        free(patch->new_name);
1021        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1022        return 0;
1023}
1024
1025static int gitdiff_renamesrc(const char *line, struct patch *patch)
1026{
1027        patch->is_rename = 1;
1028        free(patch->old_name);
1029        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1030        return 0;
1031}
1032
1033static int gitdiff_renamedst(const char *line, struct patch *patch)
1034{
1035        patch->is_rename = 1;
1036        free(patch->new_name);
1037        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1038        return 0;
1039}
1040
1041static int gitdiff_similarity(const char *line, struct patch *patch)
1042{
1043        unsigned long val = strtoul(line, NULL, 10);
1044        if (val <= 100)
1045                patch->score = val;
1046        return 0;
1047}
1048
1049static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1050{
1051        unsigned long val = strtoul(line, NULL, 10);
1052        if (val <= 100)
1053                patch->score = val;
1054        return 0;
1055}
1056
1057static int gitdiff_index(const char *line, struct patch *patch)
1058{
1059        /*
1060         * index line is N hexadecimal, "..", N hexadecimal,
1061         * and optional space with octal mode.
1062         */
1063        const char *ptr, *eol;
1064        int len;
1065
1066        ptr = strchr(line, '.');
1067        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1068                return 0;
1069        len = ptr - line;
1070        memcpy(patch->old_sha1_prefix, line, len);
1071        patch->old_sha1_prefix[len] = 0;
1072
1073        line = ptr + 2;
1074        ptr = strchr(line, ' ');
1075        eol = strchrnul(line, '\n');
1076
1077        if (!ptr || eol < ptr)
1078                ptr = eol;
1079        len = ptr - line;
1080
1081        if (40 < len)
1082                return 0;
1083        memcpy(patch->new_sha1_prefix, line, len);
1084        patch->new_sha1_prefix[len] = 0;
1085        if (*ptr == ' ')
1086                patch->old_mode = strtoul(ptr+1, NULL, 8);
1087        return 0;
1088}
1089
1090/*
1091 * This is normal for a diff that doesn't change anything: we'll fall through
1092 * into the next diff. Tell the parser to break out.
1093 */
1094static int gitdiff_unrecognized(const char *line, struct patch *patch)
1095{
1096        return -1;
1097}
1098
1099/*
1100 * Skip p_value leading components from "line"; as we do not accept
1101 * absolute paths, return NULL in that case.
1102 */
1103static const char *skip_tree_prefix(const char *line, int llen)
1104{
1105        int nslash;
1106        int i;
1107
1108        if (!p_value)
1109                return (llen && line[0] == '/') ? NULL : line;
1110
1111        nslash = p_value;
1112        for (i = 0; i < llen; i++) {
1113                int ch = line[i];
1114                if (ch == '/' && --nslash <= 0)
1115                        return (i == 0) ? NULL : &line[i + 1];
1116        }
1117        return NULL;
1118}
1119
1120/*
1121 * This is to extract the same name that appears on "diff --git"
1122 * line.  We do not find and return anything if it is a rename
1123 * patch, and it is OK because we will find the name elsewhere.
1124 * We need to reliably find name only when it is mode-change only,
1125 * creation or deletion of an empty file.  In any of these cases,
1126 * both sides are the same name under a/ and b/ respectively.
1127 */
1128static char *git_header_name(const char *line, int llen)
1129{
1130        const char *name;
1131        const char *second = NULL;
1132        size_t len, line_len;
1133
1134        line += strlen("diff --git ");
1135        llen -= strlen("diff --git ");
1136
1137        if (*line == '"') {
1138                const char *cp;
1139                struct strbuf first = STRBUF_INIT;
1140                struct strbuf sp = STRBUF_INIT;
1141
1142                if (unquote_c_style(&first, line, &second))
1143                        goto free_and_fail1;
1144
1145                /* strip the a/b prefix including trailing slash */
1146                cp = skip_tree_prefix(first.buf, first.len);
1147                if (!cp)
1148                        goto free_and_fail1;
1149                strbuf_remove(&first, 0, cp - first.buf);
1150
1151                /*
1152                 * second points at one past closing dq of name.
1153                 * find the second name.
1154                 */
1155                while ((second < line + llen) && isspace(*second))
1156                        second++;
1157
1158                if (line + llen <= second)
1159                        goto free_and_fail1;
1160                if (*second == '"') {
1161                        if (unquote_c_style(&sp, second, NULL))
1162                                goto free_and_fail1;
1163                        cp = skip_tree_prefix(sp.buf, sp.len);
1164                        if (!cp)
1165                                goto free_and_fail1;
1166                        /* They must match, otherwise ignore */
1167                        if (strcmp(cp, first.buf))
1168                                goto free_and_fail1;
1169                        strbuf_release(&sp);
1170                        return strbuf_detach(&first, NULL);
1171                }
1172
1173                /* unquoted second */
1174                cp = skip_tree_prefix(second, line + llen - second);
1175                if (!cp)
1176                        goto free_and_fail1;
1177                if (line + llen - cp != first.len ||
1178                    memcmp(first.buf, cp, first.len))
1179                        goto free_and_fail1;
1180                return strbuf_detach(&first, NULL);
1181
1182        free_and_fail1:
1183                strbuf_release(&first);
1184                strbuf_release(&sp);
1185                return NULL;
1186        }
1187
1188        /* unquoted first name */
1189        name = skip_tree_prefix(line, llen);
1190        if (!name)
1191                return NULL;
1192
1193        /*
1194         * since the first name is unquoted, a dq if exists must be
1195         * the beginning of the second name.
1196         */
1197        for (second = name; second < line + llen; second++) {
1198                if (*second == '"') {
1199                        struct strbuf sp = STRBUF_INIT;
1200                        const char *np;
1201
1202                        if (unquote_c_style(&sp, second, NULL))
1203                                goto free_and_fail2;
1204
1205                        np = skip_tree_prefix(sp.buf, sp.len);
1206                        if (!np)
1207                                goto free_and_fail2;
1208
1209                        len = sp.buf + sp.len - np;
1210                        if (len < second - name &&
1211                            !strncmp(np, name, len) &&
1212                            isspace(name[len])) {
1213                                /* Good */
1214                                strbuf_remove(&sp, 0, np - sp.buf);
1215                                return strbuf_detach(&sp, NULL);
1216                        }
1217
1218                free_and_fail2:
1219                        strbuf_release(&sp);
1220                        return NULL;
1221                }
1222        }
1223
1224        /*
1225         * Accept a name only if it shows up twice, exactly the same
1226         * form.
1227         */
1228        second = strchr(name, '\n');
1229        if (!second)
1230                return NULL;
1231        line_len = second - name;
1232        for (len = 0 ; ; len++) {
1233                switch (name[len]) {
1234                default:
1235                        continue;
1236                case '\n':
1237                        return NULL;
1238                case '\t': case ' ':
1239                        /*
1240                         * Is this the separator between the preimage
1241                         * and the postimage pathname?  Again, we are
1242                         * only interested in the case where there is
1243                         * no rename, as this is only to set def_name
1244                         * and a rename patch has the names elsewhere
1245                         * in an unambiguous form.
1246                         */
1247                        if (!name[len + 1])
1248                                return NULL; /* no postimage name */
1249                        second = skip_tree_prefix(name + len + 1,
1250                                                  line_len - (len + 1));
1251                        if (!second)
1252                                return NULL;
1253                        /*
1254                         * Does len bytes starting at "name" and "second"
1255                         * (that are separated by one HT or SP we just
1256                         * found) exactly match?
1257                         */
1258                        if (second[len] == '\n' && !strncmp(name, second, len))
1259                                return xmemdupz(name, len);
1260                }
1261        }
1262}
1263
1264/* Verify that we recognize the lines following a git header */
1265static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1266{
1267        unsigned long offset;
1268
1269        /* A git diff has explicit new/delete information, so we don't guess */
1270        patch->is_new = 0;
1271        patch->is_delete = 0;
1272
1273        /*
1274         * Some things may not have the old name in the
1275         * rest of the headers anywhere (pure mode changes,
1276         * or removing or adding empty files), so we get
1277         * the default name from the header.
1278         */
1279        patch->def_name = git_header_name(line, len);
1280        if (patch->def_name && root) {
1281                char *s = xstrfmt("%s%s", root, patch->def_name);
1282                free(patch->def_name);
1283                patch->def_name = s;
1284        }
1285
1286        line += len;
1287        size -= len;
1288        linenr++;
1289        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1290                static const struct opentry {
1291                        const char *str;
1292                        int (*fn)(const char *, struct patch *);
1293                } optable[] = {
1294                        { "@@ -", gitdiff_hdrend },
1295                        { "--- ", gitdiff_oldname },
1296                        { "+++ ", gitdiff_newname },
1297                        { "old mode ", gitdiff_oldmode },
1298                        { "new mode ", gitdiff_newmode },
1299                        { "deleted file mode ", gitdiff_delete },
1300                        { "new file mode ", gitdiff_newfile },
1301                        { "copy from ", gitdiff_copysrc },
1302                        { "copy to ", gitdiff_copydst },
1303                        { "rename old ", gitdiff_renamesrc },
1304                        { "rename new ", gitdiff_renamedst },
1305                        { "rename from ", gitdiff_renamesrc },
1306                        { "rename to ", gitdiff_renamedst },
1307                        { "similarity index ", gitdiff_similarity },
1308                        { "dissimilarity index ", gitdiff_dissimilarity },
1309                        { "index ", gitdiff_index },
1310                        { "", gitdiff_unrecognized },
1311                };
1312                int i;
1313
1314                len = linelen(line, size);
1315                if (!len || line[len-1] != '\n')
1316                        break;
1317                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1318                        const struct opentry *p = optable + i;
1319                        int oplen = strlen(p->str);
1320                        if (len < oplen || memcmp(p->str, line, oplen))
1321                                continue;
1322                        if (p->fn(line + oplen, patch) < 0)
1323                                return offset;
1324                        break;
1325                }
1326        }
1327
1328        return offset;
1329}
1330
1331static int parse_num(const char *line, unsigned long *p)
1332{
1333        char *ptr;
1334
1335        if (!isdigit(*line))
1336                return 0;
1337        *p = strtoul(line, &ptr, 10);
1338        return ptr - line;
1339}
1340
1341static int parse_range(const char *line, int len, int offset, const char *expect,
1342                       unsigned long *p1, unsigned long *p2)
1343{
1344        int digits, ex;
1345
1346        if (offset < 0 || offset >= len)
1347                return -1;
1348        line += offset;
1349        len -= offset;
1350
1351        digits = parse_num(line, p1);
1352        if (!digits)
1353                return -1;
1354
1355        offset += digits;
1356        line += digits;
1357        len -= digits;
1358
1359        *p2 = 1;
1360        if (*line == ',') {
1361                digits = parse_num(line+1, p2);
1362                if (!digits)
1363                        return -1;
1364
1365                offset += digits+1;
1366                line += digits+1;
1367                len -= digits+1;
1368        }
1369
1370        ex = strlen(expect);
1371        if (ex > len)
1372                return -1;
1373        if (memcmp(line, expect, ex))
1374                return -1;
1375
1376        return offset + ex;
1377}
1378
1379static void recount_diff(const char *line, int size, struct fragment *fragment)
1380{
1381        int oldlines = 0, newlines = 0, ret = 0;
1382
1383        if (size < 1) {
1384                warning("recount: ignore empty hunk");
1385                return;
1386        }
1387
1388        for (;;) {
1389                int len = linelen(line, size);
1390                size -= len;
1391                line += len;
1392
1393                if (size < 1)
1394                        break;
1395
1396                switch (*line) {
1397                case ' ': case '\n':
1398                        newlines++;
1399                        /* fall through */
1400                case '-':
1401                        oldlines++;
1402                        continue;
1403                case '+':
1404                        newlines++;
1405                        continue;
1406                case '\\':
1407                        continue;
1408                case '@':
1409                        ret = size < 3 || !starts_with(line, "@@ ");
1410                        break;
1411                case 'd':
1412                        ret = size < 5 || !starts_with(line, "diff ");
1413                        break;
1414                default:
1415                        ret = -1;
1416                        break;
1417                }
1418                if (ret) {
1419                        warning(_("recount: unexpected line: %.*s"),
1420                                (int)linelen(line, size), line);
1421                        return;
1422                }
1423                break;
1424        }
1425        fragment->oldlines = oldlines;
1426        fragment->newlines = newlines;
1427}
1428
1429/*
1430 * Parse a unified diff fragment header of the
1431 * form "@@ -a,b +c,d @@"
1432 */
1433static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1434{
1435        int offset;
1436
1437        if (!len || line[len-1] != '\n')
1438                return -1;
1439
1440        /* Figure out the number of lines in a fragment */
1441        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1442        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1443
1444        return offset;
1445}
1446
1447static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1448{
1449        unsigned long offset, len;
1450
1451        patch->is_toplevel_relative = 0;
1452        patch->is_rename = patch->is_copy = 0;
1453        patch->is_new = patch->is_delete = -1;
1454        patch->old_mode = patch->new_mode = 0;
1455        patch->old_name = patch->new_name = NULL;
1456        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1457                unsigned long nextlen;
1458
1459                len = linelen(line, size);
1460                if (!len)
1461                        break;
1462
1463                /* Testing this early allows us to take a few shortcuts.. */
1464                if (len < 6)
1465                        continue;
1466
1467                /*
1468                 * Make sure we don't find any unconnected patch fragments.
1469                 * That's a sign that we didn't find a header, and that a
1470                 * patch has become corrupted/broken up.
1471                 */
1472                if (!memcmp("@@ -", line, 4)) {
1473                        struct fragment dummy;
1474                        if (parse_fragment_header(line, len, &dummy) < 0)
1475                                continue;
1476                        die(_("patch fragment without header at line %d: %.*s"),
1477                            linenr, (int)len-1, line);
1478                }
1479
1480                if (size < len + 6)
1481                        break;
1482
1483                /*
1484                 * Git patch? It might not have a real patch, just a rename
1485                 * or mode change, so we handle that specially
1486                 */
1487                if (!memcmp("diff --git ", line, 11)) {
1488                        int git_hdr_len = parse_git_header(line, len, size, patch);
1489                        if (git_hdr_len <= len)
1490                                continue;
1491                        if (!patch->old_name && !patch->new_name) {
1492                                if (!patch->def_name)
1493                                        die(Q_("git diff header lacks filename information when removing "
1494                                               "%d leading pathname component (line %d)",
1495                                               "git diff header lacks filename information when removing "
1496                                               "%d leading pathname components (line %d)",
1497                                               p_value),
1498                                            p_value, linenr);
1499                                patch->old_name = xstrdup(patch->def_name);
1500                                patch->new_name = xstrdup(patch->def_name);
1501                        }
1502                        if (!patch->is_delete && !patch->new_name)
1503                                die("git diff header lacks filename information "
1504                                    "(line %d)", linenr);
1505                        patch->is_toplevel_relative = 1;
1506                        *hdrsize = git_hdr_len;
1507                        return offset;
1508                }
1509
1510                /* --- followed by +++ ? */
1511                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1512                        continue;
1513
1514                /*
1515                 * We only accept unified patches, so we want it to
1516                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1517                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1518                 */
1519                nextlen = linelen(line + len, size - len);
1520                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1521                        continue;
1522
1523                /* Ok, we'll consider it a patch */
1524                parse_traditional_patch(line, line+len, patch);
1525                *hdrsize = len + nextlen;
1526                linenr += 2;
1527                return offset;
1528        }
1529        return -1;
1530}
1531
1532static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1533{
1534        char *err;
1535
1536        if (!result)
1537                return;
1538
1539        whitespace_error++;
1540        if (squelch_whitespace_errors &&
1541            squelch_whitespace_errors < whitespace_error)
1542                return;
1543
1544        err = whitespace_error_string(result);
1545        fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1546                patch_input_file, linenr, err, len, line);
1547        free(err);
1548}
1549
1550static void check_whitespace(const char *line, int len, unsigned ws_rule)
1551{
1552        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1553
1554        record_ws_error(result, line + 1, len - 2, linenr);
1555}
1556
1557/*
1558 * Parse a unified diff. Note that this really needs to parse each
1559 * fragment separately, since the only way to know the difference
1560 * between a "---" that is part of a patch, and a "---" that starts
1561 * the next patch is to look at the line counts..
1562 */
1563static int parse_fragment(const char *line, unsigned long size,
1564                          struct patch *patch, struct fragment *fragment)
1565{
1566        int added, deleted;
1567        int len = linelen(line, size), offset;
1568        unsigned long oldlines, newlines;
1569        unsigned long leading, trailing;
1570
1571        offset = parse_fragment_header(line, len, fragment);
1572        if (offset < 0)
1573                return -1;
1574        if (offset > 0 && patch->recount)
1575                recount_diff(line + offset, size - offset, fragment);
1576        oldlines = fragment->oldlines;
1577        newlines = fragment->newlines;
1578        leading = 0;
1579        trailing = 0;
1580
1581        /* Parse the thing.. */
1582        line += len;
1583        size -= len;
1584        linenr++;
1585        added = deleted = 0;
1586        for (offset = len;
1587             0 < size;
1588             offset += len, size -= len, line += len, linenr++) {
1589                if (!oldlines && !newlines)
1590                        break;
1591                len = linelen(line, size);
1592                if (!len || line[len-1] != '\n')
1593                        return -1;
1594                switch (*line) {
1595                default:
1596                        return -1;
1597                case '\n': /* newer GNU diff, an empty context line */
1598                case ' ':
1599                        oldlines--;
1600                        newlines--;
1601                        if (!deleted && !added)
1602                                leading++;
1603                        trailing++;
1604                        break;
1605                case '-':
1606                        if (apply_in_reverse &&
1607                            ws_error_action != nowarn_ws_error)
1608                                check_whitespace(line, len, patch->ws_rule);
1609                        deleted++;
1610                        oldlines--;
1611                        trailing = 0;
1612                        break;
1613                case '+':
1614                        if (!apply_in_reverse &&
1615                            ws_error_action != nowarn_ws_error)
1616                                check_whitespace(line, len, patch->ws_rule);
1617                        added++;
1618                        newlines--;
1619                        trailing = 0;
1620                        break;
1621
1622                /*
1623                 * We allow "\ No newline at end of file". Depending
1624                 * on locale settings when the patch was produced we
1625                 * don't know what this line looks like. The only
1626                 * thing we do know is that it begins with "\ ".
1627                 * Checking for 12 is just for sanity check -- any
1628                 * l10n of "\ No newline..." is at least that long.
1629                 */
1630                case '\\':
1631                        if (len < 12 || memcmp(line, "\\ ", 2))
1632                                return -1;
1633                        break;
1634                }
1635        }
1636        if (oldlines || newlines)
1637                return -1;
1638        fragment->leading = leading;
1639        fragment->trailing = trailing;
1640
1641        /*
1642         * If a fragment ends with an incomplete line, we failed to include
1643         * it in the above loop because we hit oldlines == newlines == 0
1644         * before seeing it.
1645         */
1646        if (12 < size && !memcmp(line, "\\ ", 2))
1647                offset += linelen(line, size);
1648
1649        patch->lines_added += added;
1650        patch->lines_deleted += deleted;
1651
1652        if (0 < patch->is_new && oldlines)
1653                return error(_("new file depends on old contents"));
1654        if (0 < patch->is_delete && newlines)
1655                return error(_("deleted file still has contents"));
1656        return offset;
1657}
1658
1659/*
1660 * We have seen "diff --git a/... b/..." header (or a traditional patch
1661 * header).  Read hunks that belong to this patch into fragments and hang
1662 * them to the given patch structure.
1663 *
1664 * The (fragment->patch, fragment->size) pair points into the memory given
1665 * by the caller, not a copy, when we return.
1666 */
1667static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1668{
1669        unsigned long offset = 0;
1670        unsigned long oldlines = 0, newlines = 0, context = 0;
1671        struct fragment **fragp = &patch->fragments;
1672
1673        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1674                struct fragment *fragment;
1675                int len;
1676
1677                fragment = xcalloc(1, sizeof(*fragment));
1678                fragment->linenr = linenr;
1679                len = parse_fragment(line, size, patch, fragment);
1680                if (len <= 0)
1681                        die(_("corrupt patch at line %d"), linenr);
1682                fragment->patch = line;
1683                fragment->size = len;
1684                oldlines += fragment->oldlines;
1685                newlines += fragment->newlines;
1686                context += fragment->leading + fragment->trailing;
1687
1688                *fragp = fragment;
1689                fragp = &fragment->next;
1690
1691                offset += len;
1692                line += len;
1693                size -= len;
1694        }
1695
1696        /*
1697         * If something was removed (i.e. we have old-lines) it cannot
1698         * be creation, and if something was added it cannot be
1699         * deletion.  However, the reverse is not true; --unified=0
1700         * patches that only add are not necessarily creation even
1701         * though they do not have any old lines, and ones that only
1702         * delete are not necessarily deletion.
1703         *
1704         * Unfortunately, a real creation/deletion patch do _not_ have
1705         * any context line by definition, so we cannot safely tell it
1706         * apart with --unified=0 insanity.  At least if the patch has
1707         * more than one hunk it is not creation or deletion.
1708         */
1709        if (patch->is_new < 0 &&
1710            (oldlines || (patch->fragments && patch->fragments->next)))
1711                patch->is_new = 0;
1712        if (patch->is_delete < 0 &&
1713            (newlines || (patch->fragments && patch->fragments->next)))
1714                patch->is_delete = 0;
1715
1716        if (0 < patch->is_new && oldlines)
1717                die(_("new file %s depends on old contents"), patch->new_name);
1718        if (0 < patch->is_delete && newlines)
1719                die(_("deleted file %s still has contents"), patch->old_name);
1720        if (!patch->is_delete && !newlines && context)
1721                fprintf_ln(stderr,
1722                           _("** warning: "
1723                             "file %s becomes empty but is not deleted"),
1724                           patch->new_name);
1725
1726        return offset;
1727}
1728
1729static inline int metadata_changes(struct patch *patch)
1730{
1731        return  patch->is_rename > 0 ||
1732                patch->is_copy > 0 ||
1733                patch->is_new > 0 ||
1734                patch->is_delete ||
1735                (patch->old_mode && patch->new_mode &&
1736                 patch->old_mode != patch->new_mode);
1737}
1738
1739static char *inflate_it(const void *data, unsigned long size,
1740                        unsigned long inflated_size)
1741{
1742        git_zstream stream;
1743        void *out;
1744        int st;
1745
1746        memset(&stream, 0, sizeof(stream));
1747
1748        stream.next_in = (unsigned char *)data;
1749        stream.avail_in = size;
1750        stream.next_out = out = xmalloc(inflated_size);
1751        stream.avail_out = inflated_size;
1752        git_inflate_init(&stream);
1753        st = git_inflate(&stream, Z_FINISH);
1754        git_inflate_end(&stream);
1755        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1756                free(out);
1757                return NULL;
1758        }
1759        return out;
1760}
1761
1762/*
1763 * Read a binary hunk and return a new fragment; fragment->patch
1764 * points at an allocated memory that the caller must free, so
1765 * it is marked as "->free_patch = 1".
1766 */
1767static struct fragment *parse_binary_hunk(char **buf_p,
1768                                          unsigned long *sz_p,
1769                                          int *status_p,
1770                                          int *used_p)
1771{
1772        /*
1773         * Expect a line that begins with binary patch method ("literal"
1774         * or "delta"), followed by the length of data before deflating.
1775         * a sequence of 'length-byte' followed by base-85 encoded data
1776         * should follow, terminated by a newline.
1777         *
1778         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1779         * and we would limit the patch line to 66 characters,
1780         * so one line can fit up to 13 groups that would decode
1781         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1782         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1783         */
1784        int llen, used;
1785        unsigned long size = *sz_p;
1786        char *buffer = *buf_p;
1787        int patch_method;
1788        unsigned long origlen;
1789        char *data = NULL;
1790        int hunk_size = 0;
1791        struct fragment *frag;
1792
1793        llen = linelen(buffer, size);
1794        used = llen;
1795
1796        *status_p = 0;
1797
1798        if (starts_with(buffer, "delta ")) {
1799                patch_method = BINARY_DELTA_DEFLATED;
1800                origlen = strtoul(buffer + 6, NULL, 10);
1801        }
1802        else if (starts_with(buffer, "literal ")) {
1803                patch_method = BINARY_LITERAL_DEFLATED;
1804                origlen = strtoul(buffer + 8, NULL, 10);
1805        }
1806        else
1807                return NULL;
1808
1809        linenr++;
1810        buffer += llen;
1811        while (1) {
1812                int byte_length, max_byte_length, newsize;
1813                llen = linelen(buffer, size);
1814                used += llen;
1815                linenr++;
1816                if (llen == 1) {
1817                        /* consume the blank line */
1818                        buffer++;
1819                        size--;
1820                        break;
1821                }
1822                /*
1823                 * Minimum line is "A00000\n" which is 7-byte long,
1824                 * and the line length must be multiple of 5 plus 2.
1825                 */
1826                if ((llen < 7) || (llen-2) % 5)
1827                        goto corrupt;
1828                max_byte_length = (llen - 2) / 5 * 4;
1829                byte_length = *buffer;
1830                if ('A' <= byte_length && byte_length <= 'Z')
1831                        byte_length = byte_length - 'A' + 1;
1832                else if ('a' <= byte_length && byte_length <= 'z')
1833                        byte_length = byte_length - 'a' + 27;
1834                else
1835                        goto corrupt;
1836                /* if the input length was not multiple of 4, we would
1837                 * have filler at the end but the filler should never
1838                 * exceed 3 bytes
1839                 */
1840                if (max_byte_length < byte_length ||
1841                    byte_length <= max_byte_length - 4)
1842                        goto corrupt;
1843                newsize = hunk_size + byte_length;
1844                data = xrealloc(data, newsize);
1845                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1846                        goto corrupt;
1847                hunk_size = newsize;
1848                buffer += llen;
1849                size -= llen;
1850        }
1851
1852        frag = xcalloc(1, sizeof(*frag));
1853        frag->patch = inflate_it(data, hunk_size, origlen);
1854        frag->free_patch = 1;
1855        if (!frag->patch)
1856                goto corrupt;
1857        free(data);
1858        frag->size = origlen;
1859        *buf_p = buffer;
1860        *sz_p = size;
1861        *used_p = used;
1862        frag->binary_patch_method = patch_method;
1863        return frag;
1864
1865 corrupt:
1866        free(data);
1867        *status_p = -1;
1868        error(_("corrupt binary patch at line %d: %.*s"),
1869              linenr-1, llen-1, buffer);
1870        return NULL;
1871}
1872
1873static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1874{
1875        /*
1876         * We have read "GIT binary patch\n"; what follows is a line
1877         * that says the patch method (currently, either "literal" or
1878         * "delta") and the length of data before deflating; a
1879         * sequence of 'length-byte' followed by base-85 encoded data
1880         * follows.
1881         *
1882         * When a binary patch is reversible, there is another binary
1883         * hunk in the same format, starting with patch method (either
1884         * "literal" or "delta") with the length of data, and a sequence
1885         * of length-byte + base-85 encoded data, terminated with another
1886         * empty line.  This data, when applied to the postimage, produces
1887         * the preimage.
1888         */
1889        struct fragment *forward;
1890        struct fragment *reverse;
1891        int status;
1892        int used, used_1;
1893
1894        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1895        if (!forward && !status)
1896                /* there has to be one hunk (forward hunk) */
1897                return error(_("unrecognized binary patch at line %d"), linenr-1);
1898        if (status)
1899                /* otherwise we already gave an error message */
1900                return status;
1901
1902        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1903        if (reverse)
1904                used += used_1;
1905        else if (status) {
1906                /*
1907                 * Not having reverse hunk is not an error, but having
1908                 * a corrupt reverse hunk is.
1909                 */
1910                free((void*) forward->patch);
1911                free(forward);
1912                return status;
1913        }
1914        forward->next = reverse;
1915        patch->fragments = forward;
1916        patch->is_binary = 1;
1917        return used;
1918}
1919
1920static void prefix_one(char **name)
1921{
1922        char *old_name = *name;
1923        if (!old_name)
1924                return;
1925        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
1926        free(old_name);
1927}
1928
1929static void prefix_patch(struct patch *p)
1930{
1931        if (!prefix || p->is_toplevel_relative)
1932                return;
1933        prefix_one(&p->new_name);
1934        prefix_one(&p->old_name);
1935}
1936
1937/*
1938 * include/exclude
1939 */
1940
1941static struct string_list limit_by_name;
1942static int has_include;
1943static void add_name_limit(const char *name, int exclude)
1944{
1945        struct string_list_item *it;
1946
1947        it = string_list_append(&limit_by_name, name);
1948        it->util = exclude ? NULL : (void *) 1;
1949}
1950
1951static int use_patch(struct patch *p)
1952{
1953        const char *pathname = p->new_name ? p->new_name : p->old_name;
1954        int i;
1955
1956        /* Paths outside are not touched regardless of "--include" */
1957        if (0 < prefix_length) {
1958                int pathlen = strlen(pathname);
1959                if (pathlen <= prefix_length ||
1960                    memcmp(prefix, pathname, prefix_length))
1961                        return 0;
1962        }
1963
1964        /* See if it matches any of exclude/include rule */
1965        for (i = 0; i < limit_by_name.nr; i++) {
1966                struct string_list_item *it = &limit_by_name.items[i];
1967                if (!wildmatch(it->string, pathname, 0, NULL))
1968                        return (it->util != NULL);
1969        }
1970
1971        /*
1972         * If we had any include, a path that does not match any rule is
1973         * not used.  Otherwise, we saw bunch of exclude rules (or none)
1974         * and such a path is used.
1975         */
1976        return !has_include;
1977}
1978
1979
1980/*
1981 * Read the patch text in "buffer" that extends for "size" bytes; stop
1982 * reading after seeing a single patch (i.e. changes to a single file).
1983 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1984 * Return the number of bytes consumed, so that the caller can call us
1985 * again for the next patch.
1986 */
1987static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1988{
1989        int hdrsize, patchsize;
1990        int offset = find_header(buffer, size, &hdrsize, patch);
1991
1992        if (offset < 0)
1993                return offset;
1994
1995        prefix_patch(patch);
1996
1997        if (!use_patch(patch))
1998                patch->ws_rule = 0;
1999        else
2000                patch->ws_rule = whitespace_rule(patch->new_name
2001                                                 ? patch->new_name
2002                                                 : patch->old_name);
2003
2004        patchsize = parse_single_patch(buffer + offset + hdrsize,
2005                                       size - offset - hdrsize, patch);
2006
2007        if (!patchsize) {
2008                static const char git_binary[] = "GIT binary patch\n";
2009                int hd = hdrsize + offset;
2010                unsigned long llen = linelen(buffer + hd, size - hd);
2011
2012                if (llen == sizeof(git_binary) - 1 &&
2013                    !memcmp(git_binary, buffer + hd, llen)) {
2014                        int used;
2015                        linenr++;
2016                        used = parse_binary(buffer + hd + llen,
2017                                            size - hd - llen, patch);
2018                        if (used)
2019                                patchsize = used + llen;
2020                        else
2021                                patchsize = 0;
2022                }
2023                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2024                        static const char *binhdr[] = {
2025                                "Binary files ",
2026                                "Files ",
2027                                NULL,
2028                        };
2029                        int i;
2030                        for (i = 0; binhdr[i]; i++) {
2031                                int len = strlen(binhdr[i]);
2032                                if (len < size - hd &&
2033                                    !memcmp(binhdr[i], buffer + hd, len)) {
2034                                        linenr++;
2035                                        patch->is_binary = 1;
2036                                        patchsize = llen;
2037                                        break;
2038                                }
2039                        }
2040                }
2041
2042                /* Empty patch cannot be applied if it is a text patch
2043                 * without metadata change.  A binary patch appears
2044                 * empty to us here.
2045                 */
2046                if ((apply || check) &&
2047                    (!patch->is_binary && !metadata_changes(patch)))
2048                        die(_("patch with only garbage at line %d"), linenr);
2049        }
2050
2051        return offset + hdrsize + patchsize;
2052}
2053
2054#define swap(a,b) myswap((a),(b),sizeof(a))
2055
2056#define myswap(a, b, size) do {         \
2057        unsigned char mytmp[size];      \
2058        memcpy(mytmp, &a, size);                \
2059        memcpy(&a, &b, size);           \
2060        memcpy(&b, mytmp, size);                \
2061} while (0)
2062
2063static void reverse_patches(struct patch *p)
2064{
2065        for (; p; p = p->next) {
2066                struct fragment *frag = p->fragments;
2067
2068                swap(p->new_name, p->old_name);
2069                swap(p->new_mode, p->old_mode);
2070                swap(p->is_new, p->is_delete);
2071                swap(p->lines_added, p->lines_deleted);
2072                swap(p->old_sha1_prefix, p->new_sha1_prefix);
2073
2074                for (; frag; frag = frag->next) {
2075                        swap(frag->newpos, frag->oldpos);
2076                        swap(frag->newlines, frag->oldlines);
2077                }
2078        }
2079}
2080
2081static const char pluses[] =
2082"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2083static const char minuses[]=
2084"----------------------------------------------------------------------";
2085
2086static void show_stats(struct patch *patch)
2087{
2088        struct strbuf qname = STRBUF_INIT;
2089        char *cp = patch->new_name ? patch->new_name : patch->old_name;
2090        int max, add, del;
2091
2092        quote_c_style(cp, &qname, NULL, 0);
2093
2094        /*
2095         * "scale" the filename
2096         */
2097        max = max_len;
2098        if (max > 50)
2099                max = 50;
2100
2101        if (qname.len > max) {
2102                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2103                if (!cp)
2104                        cp = qname.buf + qname.len + 3 - max;
2105                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2106        }
2107
2108        if (patch->is_binary) {
2109                printf(" %-*s |  Bin\n", max, qname.buf);
2110                strbuf_release(&qname);
2111                return;
2112        }
2113
2114        printf(" %-*s |", max, qname.buf);
2115        strbuf_release(&qname);
2116
2117        /*
2118         * scale the add/delete
2119         */
2120        max = max + max_change > 70 ? 70 - max : max_change;
2121        add = patch->lines_added;
2122        del = patch->lines_deleted;
2123
2124        if (max_change > 0) {
2125                int total = ((add + del) * max + max_change / 2) / max_change;
2126                add = (add * max + max_change / 2) / max_change;
2127                del = total - add;
2128        }
2129        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2130                add, pluses, del, minuses);
2131}
2132
2133static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2134{
2135        switch (st->st_mode & S_IFMT) {
2136        case S_IFLNK:
2137                if (strbuf_readlink(buf, path, st->st_size) < 0)
2138                        return error(_("unable to read symlink %s"), path);
2139                return 0;
2140        case S_IFREG:
2141                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2142                        return error(_("unable to open or read %s"), path);
2143                convert_to_git(path, buf->buf, buf->len, buf, 0);
2144                return 0;
2145        default:
2146                return -1;
2147        }
2148}
2149
2150/*
2151 * Update the preimage, and the common lines in postimage,
2152 * from buffer buf of length len. If postlen is 0 the postimage
2153 * is updated in place, otherwise it's updated on a new buffer
2154 * of length postlen
2155 */
2156
2157static void update_pre_post_images(struct image *preimage,
2158                                   struct image *postimage,
2159                                   char *buf,
2160                                   size_t len, size_t postlen)
2161{
2162        int i, ctx, reduced;
2163        char *new, *old, *fixed;
2164        struct image fixed_preimage;
2165
2166        /*
2167         * Update the preimage with whitespace fixes.  Note that we
2168         * are not losing preimage->buf -- apply_one_fragment() will
2169         * free "oldlines".
2170         */
2171        prepare_image(&fixed_preimage, buf, len, 1);
2172        assert(postlen
2173               ? fixed_preimage.nr == preimage->nr
2174               : fixed_preimage.nr <= preimage->nr);
2175        for (i = 0; i < fixed_preimage.nr; i++)
2176                fixed_preimage.line[i].flag = preimage->line[i].flag;
2177        free(preimage->line_allocated);
2178        *preimage = fixed_preimage;
2179
2180        /*
2181         * Adjust the common context lines in postimage. This can be
2182         * done in-place when we are shrinking it with whitespace
2183         * fixing, but needs a new buffer when ignoring whitespace or
2184         * expanding leading tabs to spaces.
2185         *
2186         * We trust the caller to tell us if the update can be done
2187         * in place (postlen==0) or not.
2188         */
2189        old = postimage->buf;
2190        if (postlen)
2191                new = postimage->buf = xmalloc(postlen);
2192        else
2193                new = old;
2194        fixed = preimage->buf;
2195
2196        for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2197                size_t len = postimage->line[i].len;
2198                if (!(postimage->line[i].flag & LINE_COMMON)) {
2199                        /* an added line -- no counterparts in preimage */
2200                        memmove(new, old, len);
2201                        old += len;
2202                        new += len;
2203                        continue;
2204                }
2205
2206                /* a common context -- skip it in the original postimage */
2207                old += len;
2208
2209                /* and find the corresponding one in the fixed preimage */
2210                while (ctx < preimage->nr &&
2211                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2212                        fixed += preimage->line[ctx].len;
2213                        ctx++;
2214                }
2215
2216                /*
2217                 * preimage is expected to run out, if the caller
2218                 * fixed addition of trailing blank lines.
2219                 */
2220                if (preimage->nr <= ctx) {
2221                        reduced++;
2222                        continue;
2223                }
2224
2225                /* and copy it in, while fixing the line length */
2226                len = preimage->line[ctx].len;
2227                memcpy(new, fixed, len);
2228                new += len;
2229                fixed += len;
2230                postimage->line[i].len = len;
2231                ctx++;
2232        }
2233
2234        if (postlen
2235            ? postlen < new - postimage->buf
2236            : postimage->len < new - postimage->buf)
2237                die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2238                    (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2239
2240        /* Fix the length of the whole thing */
2241        postimage->len = new - postimage->buf;
2242        postimage->nr -= reduced;
2243}
2244
2245static int match_fragment(struct image *img,
2246                          struct image *preimage,
2247                          struct image *postimage,
2248                          unsigned long try,
2249                          int try_lno,
2250                          unsigned ws_rule,
2251                          int match_beginning, int match_end)
2252{
2253        int i;
2254        char *fixed_buf, *buf, *orig, *target;
2255        struct strbuf fixed;
2256        size_t fixed_len, postlen;
2257        int preimage_limit;
2258
2259        if (preimage->nr + try_lno <= img->nr) {
2260                /*
2261                 * The hunk falls within the boundaries of img.
2262                 */
2263                preimage_limit = preimage->nr;
2264                if (match_end && (preimage->nr + try_lno != img->nr))
2265                        return 0;
2266        } else if (ws_error_action == correct_ws_error &&
2267                   (ws_rule & WS_BLANK_AT_EOF)) {
2268                /*
2269                 * This hunk extends beyond the end of img, and we are
2270                 * removing blank lines at the end of the file.  This
2271                 * many lines from the beginning of the preimage must
2272                 * match with img, and the remainder of the preimage
2273                 * must be blank.
2274                 */
2275                preimage_limit = img->nr - try_lno;
2276        } else {
2277                /*
2278                 * The hunk extends beyond the end of the img and
2279                 * we are not removing blanks at the end, so we
2280                 * should reject the hunk at this position.
2281                 */
2282                return 0;
2283        }
2284
2285        if (match_beginning && try_lno)
2286                return 0;
2287
2288        /* Quick hash check */
2289        for (i = 0; i < preimage_limit; i++)
2290                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2291                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2292                        return 0;
2293
2294        if (preimage_limit == preimage->nr) {
2295                /*
2296                 * Do we have an exact match?  If we were told to match
2297                 * at the end, size must be exactly at try+fragsize,
2298                 * otherwise try+fragsize must be still within the preimage,
2299                 * and either case, the old piece should match the preimage
2300                 * exactly.
2301                 */
2302                if ((match_end
2303                     ? (try + preimage->len == img->len)
2304                     : (try + preimage->len <= img->len)) &&
2305                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2306                        return 1;
2307        } else {
2308                /*
2309                 * The preimage extends beyond the end of img, so
2310                 * there cannot be an exact match.
2311                 *
2312                 * There must be one non-blank context line that match
2313                 * a line before the end of img.
2314                 */
2315                char *buf_end;
2316
2317                buf = preimage->buf;
2318                buf_end = buf;
2319                for (i = 0; i < preimage_limit; i++)
2320                        buf_end += preimage->line[i].len;
2321
2322                for ( ; buf < buf_end; buf++)
2323                        if (!isspace(*buf))
2324                                break;
2325                if (buf == buf_end)
2326                        return 0;
2327        }
2328
2329        /*
2330         * No exact match. If we are ignoring whitespace, run a line-by-line
2331         * fuzzy matching. We collect all the line length information because
2332         * we need it to adjust whitespace if we match.
2333         */
2334        if (ws_ignore_action == ignore_ws_change) {
2335                size_t imgoff = 0;
2336                size_t preoff = 0;
2337                size_t postlen = postimage->len;
2338                size_t extra_chars;
2339                char *preimage_eof;
2340                char *preimage_end;
2341                for (i = 0; i < preimage_limit; i++) {
2342                        size_t prelen = preimage->line[i].len;
2343                        size_t imglen = img->line[try_lno+i].len;
2344
2345                        if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2346                                              preimage->buf + preoff, prelen))
2347                                return 0;
2348                        if (preimage->line[i].flag & LINE_COMMON)
2349                                postlen += imglen - prelen;
2350                        imgoff += imglen;
2351                        preoff += prelen;
2352                }
2353
2354                /*
2355                 * Ok, the preimage matches with whitespace fuzz.
2356                 *
2357                 * imgoff now holds the true length of the target that
2358                 * matches the preimage before the end of the file.
2359                 *
2360                 * Count the number of characters in the preimage that fall
2361                 * beyond the end of the file and make sure that all of them
2362                 * are whitespace characters. (This can only happen if
2363                 * we are removing blank lines at the end of the file.)
2364                 */
2365                buf = preimage_eof = preimage->buf + preoff;
2366                for ( ; i < preimage->nr; i++)
2367                        preoff += preimage->line[i].len;
2368                preimage_end = preimage->buf + preoff;
2369                for ( ; buf < preimage_end; buf++)
2370                        if (!isspace(*buf))
2371                                return 0;
2372
2373                /*
2374                 * Update the preimage and the common postimage context
2375                 * lines to use the same whitespace as the target.
2376                 * If whitespace is missing in the target (i.e.
2377                 * if the preimage extends beyond the end of the file),
2378                 * use the whitespace from the preimage.
2379                 */
2380                extra_chars = preimage_end - preimage_eof;
2381                strbuf_init(&fixed, imgoff + extra_chars);
2382                strbuf_add(&fixed, img->buf + try, imgoff);
2383                strbuf_add(&fixed, preimage_eof, extra_chars);
2384                fixed_buf = strbuf_detach(&fixed, &fixed_len);
2385                update_pre_post_images(preimage, postimage,
2386                                fixed_buf, fixed_len, postlen);
2387                return 1;
2388        }
2389
2390        if (ws_error_action != correct_ws_error)
2391                return 0;
2392
2393        /*
2394         * The hunk does not apply byte-by-byte, but the hash says
2395         * it might with whitespace fuzz. We weren't asked to
2396         * ignore whitespace, we were asked to correct whitespace
2397         * errors, so let's try matching after whitespace correction.
2398         *
2399         * While checking the preimage against the target, whitespace
2400         * errors in both fixed, we count how large the corresponding
2401         * postimage needs to be.  The postimage prepared by
2402         * apply_one_fragment() has whitespace errors fixed on added
2403         * lines already, but the common lines were propagated as-is,
2404         * which may become longer when their whitespace errors are
2405         * fixed.
2406         */
2407
2408        /* First count added lines in postimage */
2409        postlen = 0;
2410        for (i = 0; i < postimage->nr; i++) {
2411                if (!(postimage->line[i].flag & LINE_COMMON))
2412                        postlen += postimage->line[i].len;
2413        }
2414
2415        /*
2416         * The preimage may extend beyond the end of the file,
2417         * but in this loop we will only handle the part of the
2418         * preimage that falls within the file.
2419         */
2420        strbuf_init(&fixed, preimage->len + 1);
2421        orig = preimage->buf;
2422        target = img->buf + try;
2423        for (i = 0; i < preimage_limit; i++) {
2424                size_t oldlen = preimage->line[i].len;
2425                size_t tgtlen = img->line[try_lno + i].len;
2426                size_t fixstart = fixed.len;
2427                struct strbuf tgtfix;
2428                int match;
2429
2430                /* Try fixing the line in the preimage */
2431                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2432
2433                /* Try fixing the line in the target */
2434                strbuf_init(&tgtfix, tgtlen);
2435                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2436
2437                /*
2438                 * If they match, either the preimage was based on
2439                 * a version before our tree fixed whitespace breakage,
2440                 * or we are lacking a whitespace-fix patch the tree
2441                 * the preimage was based on already had (i.e. target
2442                 * has whitespace breakage, the preimage doesn't).
2443                 * In either case, we are fixing the whitespace breakages
2444                 * so we might as well take the fix together with their
2445                 * real change.
2446                 */
2447                match = (tgtfix.len == fixed.len - fixstart &&
2448                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2449                                             fixed.len - fixstart));
2450
2451                /* Add the length if this is common with the postimage */
2452                if (preimage->line[i].flag & LINE_COMMON)
2453                        postlen += tgtfix.len;
2454
2455                strbuf_release(&tgtfix);
2456                if (!match)
2457                        goto unmatch_exit;
2458
2459                orig += oldlen;
2460                target += tgtlen;
2461        }
2462
2463
2464        /*
2465         * Now handle the lines in the preimage that falls beyond the
2466         * end of the file (if any). They will only match if they are
2467         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2468         * false).
2469         */
2470        for ( ; i < preimage->nr; i++) {
2471                size_t fixstart = fixed.len; /* start of the fixed preimage */
2472                size_t oldlen = preimage->line[i].len;
2473                int j;
2474
2475                /* Try fixing the line in the preimage */
2476                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2477
2478                for (j = fixstart; j < fixed.len; j++)
2479                        if (!isspace(fixed.buf[j]))
2480                                goto unmatch_exit;
2481
2482                orig += oldlen;
2483        }
2484
2485        /*
2486         * Yes, the preimage is based on an older version that still
2487         * has whitespace breakages unfixed, and fixing them makes the
2488         * hunk match.  Update the context lines in the postimage.
2489         */
2490        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2491        if (postlen < postimage->len)
2492                postlen = 0;
2493        update_pre_post_images(preimage, postimage,
2494                               fixed_buf, fixed_len, postlen);
2495        return 1;
2496
2497 unmatch_exit:
2498        strbuf_release(&fixed);
2499        return 0;
2500}
2501
2502static int find_pos(struct image *img,
2503                    struct image *preimage,
2504                    struct image *postimage,
2505                    int line,
2506                    unsigned ws_rule,
2507                    int match_beginning, int match_end)
2508{
2509        int i;
2510        unsigned long backwards, forwards, try;
2511        int backwards_lno, forwards_lno, try_lno;
2512
2513        /*
2514         * If match_beginning or match_end is specified, there is no
2515         * point starting from a wrong line that will never match and
2516         * wander around and wait for a match at the specified end.
2517         */
2518        if (match_beginning)
2519                line = 0;
2520        else if (match_end)
2521                line = img->nr - preimage->nr;
2522
2523        /*
2524         * Because the comparison is unsigned, the following test
2525         * will also take care of a negative line number that can
2526         * result when match_end and preimage is larger than the target.
2527         */
2528        if ((size_t) line > img->nr)
2529                line = img->nr;
2530
2531        try = 0;
2532        for (i = 0; i < line; i++)
2533                try += img->line[i].len;
2534
2535        /*
2536         * There's probably some smart way to do this, but I'll leave
2537         * that to the smart and beautiful people. I'm simple and stupid.
2538         */
2539        backwards = try;
2540        backwards_lno = line;
2541        forwards = try;
2542        forwards_lno = line;
2543        try_lno = line;
2544
2545        for (i = 0; ; i++) {
2546                if (match_fragment(img, preimage, postimage,
2547                                   try, try_lno, ws_rule,
2548                                   match_beginning, match_end))
2549                        return try_lno;
2550
2551        again:
2552                if (backwards_lno == 0 && forwards_lno == img->nr)
2553                        break;
2554
2555                if (i & 1) {
2556                        if (backwards_lno == 0) {
2557                                i++;
2558                                goto again;
2559                        }
2560                        backwards_lno--;
2561                        backwards -= img->line[backwards_lno].len;
2562                        try = backwards;
2563                        try_lno = backwards_lno;
2564                } else {
2565                        if (forwards_lno == img->nr) {
2566                                i++;
2567                                goto again;
2568                        }
2569                        forwards += img->line[forwards_lno].len;
2570                        forwards_lno++;
2571                        try = forwards;
2572                        try_lno = forwards_lno;
2573                }
2574
2575        }
2576        return -1;
2577}
2578
2579static void remove_first_line(struct image *img)
2580{
2581        img->buf += img->line[0].len;
2582        img->len -= img->line[0].len;
2583        img->line++;
2584        img->nr--;
2585}
2586
2587static void remove_last_line(struct image *img)
2588{
2589        img->len -= img->line[--img->nr].len;
2590}
2591
2592/*
2593 * The change from "preimage" and "postimage" has been found to
2594 * apply at applied_pos (counts in line numbers) in "img".
2595 * Update "img" to remove "preimage" and replace it with "postimage".
2596 */
2597static void update_image(struct image *img,
2598                         int applied_pos,
2599                         struct image *preimage,
2600                         struct image *postimage)
2601{
2602        /*
2603         * remove the copy of preimage at offset in img
2604         * and replace it with postimage
2605         */
2606        int i, nr;
2607        size_t remove_count, insert_count, applied_at = 0;
2608        char *result;
2609        int preimage_limit;
2610
2611        /*
2612         * If we are removing blank lines at the end of img,
2613         * the preimage may extend beyond the end.
2614         * If that is the case, we must be careful only to
2615         * remove the part of the preimage that falls within
2616         * the boundaries of img. Initialize preimage_limit
2617         * to the number of lines in the preimage that falls
2618         * within the boundaries.
2619         */
2620        preimage_limit = preimage->nr;
2621        if (preimage_limit > img->nr - applied_pos)
2622                preimage_limit = img->nr - applied_pos;
2623
2624        for (i = 0; i < applied_pos; i++)
2625                applied_at += img->line[i].len;
2626
2627        remove_count = 0;
2628        for (i = 0; i < preimage_limit; i++)
2629                remove_count += img->line[applied_pos + i].len;
2630        insert_count = postimage->len;
2631
2632        /* Adjust the contents */
2633        result = xmalloc(img->len + insert_count - remove_count + 1);
2634        memcpy(result, img->buf, applied_at);
2635        memcpy(result + applied_at, postimage->buf, postimage->len);
2636        memcpy(result + applied_at + postimage->len,
2637               img->buf + (applied_at + remove_count),
2638               img->len - (applied_at + remove_count));
2639        free(img->buf);
2640        img->buf = result;
2641        img->len += insert_count - remove_count;
2642        result[img->len] = '\0';
2643
2644        /* Adjust the line table */
2645        nr = img->nr + postimage->nr - preimage_limit;
2646        if (preimage_limit < postimage->nr) {
2647                /*
2648                 * NOTE: this knows that we never call remove_first_line()
2649                 * on anything other than pre/post image.
2650                 */
2651                REALLOC_ARRAY(img->line, nr);
2652                img->line_allocated = img->line;
2653        }
2654        if (preimage_limit != postimage->nr)
2655                memmove(img->line + applied_pos + postimage->nr,
2656                        img->line + applied_pos + preimage_limit,
2657                        (img->nr - (applied_pos + preimage_limit)) *
2658                        sizeof(*img->line));
2659        memcpy(img->line + applied_pos,
2660               postimage->line,
2661               postimage->nr * sizeof(*img->line));
2662        if (!allow_overlap)
2663                for (i = 0; i < postimage->nr; i++)
2664                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2665        img->nr = nr;
2666}
2667
2668/*
2669 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2670 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2671 * replace the part of "img" with "postimage" text.
2672 */
2673static int apply_one_fragment(struct image *img, struct fragment *frag,
2674                              int inaccurate_eof, unsigned ws_rule,
2675                              int nth_fragment)
2676{
2677        int match_beginning, match_end;
2678        const char *patch = frag->patch;
2679        int size = frag->size;
2680        char *old, *oldlines;
2681        struct strbuf newlines;
2682        int new_blank_lines_at_end = 0;
2683        int found_new_blank_lines_at_end = 0;
2684        int hunk_linenr = frag->linenr;
2685        unsigned long leading, trailing;
2686        int pos, applied_pos;
2687        struct image preimage;
2688        struct image postimage;
2689
2690        memset(&preimage, 0, sizeof(preimage));
2691        memset(&postimage, 0, sizeof(postimage));
2692        oldlines = xmalloc(size);
2693        strbuf_init(&newlines, size);
2694
2695        old = oldlines;
2696        while (size > 0) {
2697                char first;
2698                int len = linelen(patch, size);
2699                int plen;
2700                int added_blank_line = 0;
2701                int is_blank_context = 0;
2702                size_t start;
2703
2704                if (!len)
2705                        break;
2706
2707                /*
2708                 * "plen" is how much of the line we should use for
2709                 * the actual patch data. Normally we just remove the
2710                 * first character on the line, but if the line is
2711                 * followed by "\ No newline", then we also remove the
2712                 * last one (which is the newline, of course).
2713                 */
2714                plen = len - 1;
2715                if (len < size && patch[len] == '\\')
2716                        plen--;
2717                first = *patch;
2718                if (apply_in_reverse) {
2719                        if (first == '-')
2720                                first = '+';
2721                        else if (first == '+')
2722                                first = '-';
2723                }
2724
2725                switch (first) {
2726                case '\n':
2727                        /* Newer GNU diff, empty context line */
2728                        if (plen < 0)
2729                                /* ... followed by '\No newline'; nothing */
2730                                break;
2731                        *old++ = '\n';
2732                        strbuf_addch(&newlines, '\n');
2733                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2734                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2735                        is_blank_context = 1;
2736                        break;
2737                case ' ':
2738                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2739                            ws_blank_line(patch + 1, plen, ws_rule))
2740                                is_blank_context = 1;
2741                case '-':
2742                        memcpy(old, patch + 1, plen);
2743                        add_line_info(&preimage, old, plen,
2744                                      (first == ' ' ? LINE_COMMON : 0));
2745                        old += plen;
2746                        if (first == '-')
2747                                break;
2748                /* Fall-through for ' ' */
2749                case '+':
2750                        /* --no-add does not add new lines */
2751                        if (first == '+' && no_add)
2752                                break;
2753
2754                        start = newlines.len;
2755                        if (first != '+' ||
2756                            !whitespace_error ||
2757                            ws_error_action != correct_ws_error) {
2758                                strbuf_add(&newlines, patch + 1, plen);
2759                        }
2760                        else {
2761                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2762                        }
2763                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2764                                      (first == '+' ? 0 : LINE_COMMON));
2765                        if (first == '+' &&
2766                            (ws_rule & WS_BLANK_AT_EOF) &&
2767                            ws_blank_line(patch + 1, plen, ws_rule))
2768                                added_blank_line = 1;
2769                        break;
2770                case '@': case '\\':
2771                        /* Ignore it, we already handled it */
2772                        break;
2773                default:
2774                        if (apply_verbosely)
2775                                error(_("invalid start of line: '%c'"), first);
2776                        return -1;
2777                }
2778                if (added_blank_line) {
2779                        if (!new_blank_lines_at_end)
2780                                found_new_blank_lines_at_end = hunk_linenr;
2781                        new_blank_lines_at_end++;
2782                }
2783                else if (is_blank_context)
2784                        ;
2785                else
2786                        new_blank_lines_at_end = 0;
2787                patch += len;
2788                size -= len;
2789                hunk_linenr++;
2790        }
2791        if (inaccurate_eof &&
2792            old > oldlines && old[-1] == '\n' &&
2793            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2794                old--;
2795                strbuf_setlen(&newlines, newlines.len - 1);
2796        }
2797
2798        leading = frag->leading;
2799        trailing = frag->trailing;
2800
2801        /*
2802         * A hunk to change lines at the beginning would begin with
2803         * @@ -1,L +N,M @@
2804         * but we need to be careful.  -U0 that inserts before the second
2805         * line also has this pattern.
2806         *
2807         * And a hunk to add to an empty file would begin with
2808         * @@ -0,0 +N,M @@
2809         *
2810         * In other words, a hunk that is (frag->oldpos <= 1) with or
2811         * without leading context must match at the beginning.
2812         */
2813        match_beginning = (!frag->oldpos ||
2814                           (frag->oldpos == 1 && !unidiff_zero));
2815
2816        /*
2817         * A hunk without trailing lines must match at the end.
2818         * However, we simply cannot tell if a hunk must match end
2819         * from the lack of trailing lines if the patch was generated
2820         * with unidiff without any context.
2821         */
2822        match_end = !unidiff_zero && !trailing;
2823
2824        pos = frag->newpos ? (frag->newpos - 1) : 0;
2825        preimage.buf = oldlines;
2826        preimage.len = old - oldlines;
2827        postimage.buf = newlines.buf;
2828        postimage.len = newlines.len;
2829        preimage.line = preimage.line_allocated;
2830        postimage.line = postimage.line_allocated;
2831
2832        for (;;) {
2833
2834                applied_pos = find_pos(img, &preimage, &postimage, pos,
2835                                       ws_rule, match_beginning, match_end);
2836
2837                if (applied_pos >= 0)
2838                        break;
2839
2840                /* Am I at my context limits? */
2841                if ((leading <= p_context) && (trailing <= p_context))
2842                        break;
2843                if (match_beginning || match_end) {
2844                        match_beginning = match_end = 0;
2845                        continue;
2846                }
2847
2848                /*
2849                 * Reduce the number of context lines; reduce both
2850                 * leading and trailing if they are equal otherwise
2851                 * just reduce the larger context.
2852                 */
2853                if (leading >= trailing) {
2854                        remove_first_line(&preimage);
2855                        remove_first_line(&postimage);
2856                        pos--;
2857                        leading--;
2858                }
2859                if (trailing > leading) {
2860                        remove_last_line(&preimage);
2861                        remove_last_line(&postimage);
2862                        trailing--;
2863                }
2864        }
2865
2866        if (applied_pos >= 0) {
2867                if (new_blank_lines_at_end &&
2868                    preimage.nr + applied_pos >= img->nr &&
2869                    (ws_rule & WS_BLANK_AT_EOF) &&
2870                    ws_error_action != nowarn_ws_error) {
2871                        record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2872                                        found_new_blank_lines_at_end);
2873                        if (ws_error_action == correct_ws_error) {
2874                                while (new_blank_lines_at_end--)
2875                                        remove_last_line(&postimage);
2876                        }
2877                        /*
2878                         * We would want to prevent write_out_results()
2879                         * from taking place in apply_patch() that follows
2880                         * the callchain led us here, which is:
2881                         * apply_patch->check_patch_list->check_patch->
2882                         * apply_data->apply_fragments->apply_one_fragment
2883                         */
2884                        if (ws_error_action == die_on_ws_error)
2885                                apply = 0;
2886                }
2887
2888                if (apply_verbosely && applied_pos != pos) {
2889                        int offset = applied_pos - pos;
2890                        if (apply_in_reverse)
2891                                offset = 0 - offset;
2892                        fprintf_ln(stderr,
2893                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
2894                                      "Hunk #%d succeeded at %d (offset %d lines).",
2895                                      offset),
2896                                   nth_fragment, applied_pos + 1, offset);
2897                }
2898
2899                /*
2900                 * Warn if it was necessary to reduce the number
2901                 * of context lines.
2902                 */
2903                if ((leading != frag->leading) ||
2904                    (trailing != frag->trailing))
2905                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2906                                             " to apply fragment at %d"),
2907                                   leading, trailing, applied_pos+1);
2908                update_image(img, applied_pos, &preimage, &postimage);
2909        } else {
2910                if (apply_verbosely)
2911                        error(_("while searching for:\n%.*s"),
2912                              (int)(old - oldlines), oldlines);
2913        }
2914
2915        free(oldlines);
2916        strbuf_release(&newlines);
2917        free(preimage.line_allocated);
2918        free(postimage.line_allocated);
2919
2920        return (applied_pos < 0);
2921}
2922
2923static int apply_binary_fragment(struct image *img, struct patch *patch)
2924{
2925        struct fragment *fragment = patch->fragments;
2926        unsigned long len;
2927        void *dst;
2928
2929        if (!fragment)
2930                return error(_("missing binary patch data for '%s'"),
2931                             patch->new_name ?
2932                             patch->new_name :
2933                             patch->old_name);
2934
2935        /* Binary patch is irreversible without the optional second hunk */
2936        if (apply_in_reverse) {
2937                if (!fragment->next)
2938                        return error("cannot reverse-apply a binary patch "
2939                                     "without the reverse hunk to '%s'",
2940                                     patch->new_name
2941                                     ? patch->new_name : patch->old_name);
2942                fragment = fragment->next;
2943        }
2944        switch (fragment->binary_patch_method) {
2945        case BINARY_DELTA_DEFLATED:
2946                dst = patch_delta(img->buf, img->len, fragment->patch,
2947                                  fragment->size, &len);
2948                if (!dst)
2949                        return -1;
2950                clear_image(img);
2951                img->buf = dst;
2952                img->len = len;
2953                return 0;
2954        case BINARY_LITERAL_DEFLATED:
2955                clear_image(img);
2956                img->len = fragment->size;
2957                img->buf = xmemdupz(fragment->patch, img->len);
2958                return 0;
2959        }
2960        return -1;
2961}
2962
2963/*
2964 * Replace "img" with the result of applying the binary patch.
2965 * The binary patch data itself in patch->fragment is still kept
2966 * but the preimage prepared by the caller in "img" is freed here
2967 * or in the helper function apply_binary_fragment() this calls.
2968 */
2969static int apply_binary(struct image *img, struct patch *patch)
2970{
2971        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2972        unsigned char sha1[20];
2973
2974        /*
2975         * For safety, we require patch index line to contain
2976         * full 40-byte textual SHA1 for old and new, at least for now.
2977         */
2978        if (strlen(patch->old_sha1_prefix) != 40 ||
2979            strlen(patch->new_sha1_prefix) != 40 ||
2980            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2981            get_sha1_hex(patch->new_sha1_prefix, sha1))
2982                return error("cannot apply binary patch to '%s' "
2983                             "without full index line", name);
2984
2985        if (patch->old_name) {
2986                /*
2987                 * See if the old one matches what the patch
2988                 * applies to.
2989                 */
2990                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2991                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2992                        return error("the patch applies to '%s' (%s), "
2993                                     "which does not match the "
2994                                     "current contents.",
2995                                     name, sha1_to_hex(sha1));
2996        }
2997        else {
2998                /* Otherwise, the old one must be empty. */
2999                if (img->len)
3000                        return error("the patch applies to an empty "
3001                                     "'%s' but it is not empty", name);
3002        }
3003
3004        get_sha1_hex(patch->new_sha1_prefix, sha1);
3005        if (is_null_sha1(sha1)) {
3006                clear_image(img);
3007                return 0; /* deletion patch */
3008        }
3009
3010        if (has_sha1_file(sha1)) {
3011                /* We already have the postimage */
3012                enum object_type type;
3013                unsigned long size;
3014                char *result;
3015
3016                result = read_sha1_file(sha1, &type, &size);
3017                if (!result)
3018                        return error("the necessary postimage %s for "
3019                                     "'%s' cannot be read",
3020                                     patch->new_sha1_prefix, name);
3021                clear_image(img);
3022                img->buf = result;
3023                img->len = size;
3024        } else {
3025                /*
3026                 * We have verified buf matches the preimage;
3027                 * apply the patch data to it, which is stored
3028                 * in the patch->fragments->{patch,size}.
3029                 */
3030                if (apply_binary_fragment(img, patch))
3031                        return error(_("binary patch does not apply to '%s'"),
3032                                     name);
3033
3034                /* verify that the result matches */
3035                hash_sha1_file(img->buf, img->len, blob_type, sha1);
3036                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
3037                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3038                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
3039        }
3040
3041        return 0;
3042}
3043
3044static int apply_fragments(struct image *img, struct patch *patch)
3045{
3046        struct fragment *frag = patch->fragments;
3047        const char *name = patch->old_name ? patch->old_name : patch->new_name;
3048        unsigned ws_rule = patch->ws_rule;
3049        unsigned inaccurate_eof = patch->inaccurate_eof;
3050        int nth = 0;
3051
3052        if (patch->is_binary)
3053                return apply_binary(img, patch);
3054
3055        while (frag) {
3056                nth++;
3057                if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
3058                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
3059                        if (!apply_with_reject)
3060                                return -1;
3061                        frag->rejected = 1;
3062                }
3063                frag = frag->next;
3064        }
3065        return 0;
3066}
3067
3068static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3069{
3070        if (S_ISGITLINK(mode)) {
3071                strbuf_grow(buf, 100);
3072                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3073        } else {
3074                enum object_type type;
3075                unsigned long sz;
3076                char *result;
3077
3078                result = read_sha1_file(sha1, &type, &sz);
3079                if (!result)
3080                        return -1;
3081                /* XXX read_sha1_file NUL-terminates */
3082                strbuf_attach(buf, result, sz, sz + 1);
3083        }
3084        return 0;
3085}
3086
3087static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3088{
3089        if (!ce)
3090                return 0;
3091        return read_blob_object(buf, ce->sha1, ce->ce_mode);
3092}
3093
3094static struct patch *in_fn_table(const char *name)
3095{
3096        struct string_list_item *item;
3097
3098        if (name == NULL)
3099                return NULL;
3100
3101        item = string_list_lookup(&fn_table, name);
3102        if (item != NULL)
3103                return (struct patch *)item->util;
3104
3105        return NULL;
3106}
3107
3108/*
3109 * item->util in the filename table records the status of the path.
3110 * Usually it points at a patch (whose result records the contents
3111 * of it after applying it), but it could be PATH_WAS_DELETED for a
3112 * path that a previously applied patch has already removed, or
3113 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3114 *
3115 * The latter is needed to deal with a case where two paths A and B
3116 * are swapped by first renaming A to B and then renaming B to A;
3117 * moving A to B should not be prevented due to presence of B as we
3118 * will remove it in a later patch.
3119 */
3120#define PATH_TO_BE_DELETED ((struct patch *) -2)
3121#define PATH_WAS_DELETED ((struct patch *) -1)
3122
3123static int to_be_deleted(struct patch *patch)
3124{
3125        return patch == PATH_TO_BE_DELETED;
3126}
3127
3128static int was_deleted(struct patch *patch)
3129{
3130        return patch == PATH_WAS_DELETED;
3131}
3132
3133static void add_to_fn_table(struct patch *patch)
3134{
3135        struct string_list_item *item;
3136
3137        /*
3138         * Always add new_name unless patch is a deletion
3139         * This should cover the cases for normal diffs,
3140         * file creations and copies
3141         */
3142        if (patch->new_name != NULL) {
3143                item = string_list_insert(&fn_table, patch->new_name);
3144                item->util = patch;
3145        }
3146
3147        /*
3148         * store a failure on rename/deletion cases because
3149         * later chunks shouldn't patch old names
3150         */
3151        if ((patch->new_name == NULL) || (patch->is_rename)) {
3152                item = string_list_insert(&fn_table, patch->old_name);
3153                item->util = PATH_WAS_DELETED;
3154        }
3155}
3156
3157static void prepare_fn_table(struct patch *patch)
3158{
3159        /*
3160         * store information about incoming file deletion
3161         */
3162        while (patch) {
3163                if ((patch->new_name == NULL) || (patch->is_rename)) {
3164                        struct string_list_item *item;
3165                        item = string_list_insert(&fn_table, patch->old_name);
3166                        item->util = PATH_TO_BE_DELETED;
3167                }
3168                patch = patch->next;
3169        }
3170}
3171
3172static int checkout_target(struct index_state *istate,
3173                           struct cache_entry *ce, struct stat *st)
3174{
3175        struct checkout costate;
3176
3177        memset(&costate, 0, sizeof(costate));
3178        costate.base_dir = "";
3179        costate.refresh_cache = 1;
3180        costate.istate = istate;
3181        if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3182                return error(_("cannot checkout %s"), ce->name);
3183        return 0;
3184}
3185
3186static struct patch *previous_patch(struct patch *patch, int *gone)
3187{
3188        struct patch *previous;
3189
3190        *gone = 0;
3191        if (patch->is_copy || patch->is_rename)
3192                return NULL; /* "git" patches do not depend on the order */
3193
3194        previous = in_fn_table(patch->old_name);
3195        if (!previous)
3196                return NULL;
3197
3198        if (to_be_deleted(previous))
3199                return NULL; /* the deletion hasn't happened yet */
3200
3201        if (was_deleted(previous))
3202                *gone = 1;
3203
3204        return previous;
3205}
3206
3207static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3208{
3209        if (S_ISGITLINK(ce->ce_mode)) {
3210                if (!S_ISDIR(st->st_mode))
3211                        return -1;
3212                return 0;
3213        }
3214        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3215}
3216
3217#define SUBMODULE_PATCH_WITHOUT_INDEX 1
3218
3219static int load_patch_target(struct strbuf *buf,
3220                             const struct cache_entry *ce,
3221                             struct stat *st,
3222                             const char *name,
3223                             unsigned expected_mode)
3224{
3225        if (cached || check_index) {
3226                if (read_file_or_gitlink(ce, buf))
3227                        return error(_("read of %s failed"), name);
3228        } else if (name) {
3229                if (S_ISGITLINK(expected_mode)) {
3230                        if (ce)
3231                                return read_file_or_gitlink(ce, buf);
3232                        else
3233                                return SUBMODULE_PATCH_WITHOUT_INDEX;
3234                } else if (has_symlink_leading_path(name, strlen(name))) {
3235                        return error(_("reading from '%s' beyond a symbolic link"), name);
3236                } else {
3237                        if (read_old_data(st, name, buf))
3238                                return error(_("read of %s failed"), name);
3239                }
3240        }
3241        return 0;
3242}
3243
3244/*
3245 * We are about to apply "patch"; populate the "image" with the
3246 * current version we have, from the working tree or from the index,
3247 * depending on the situation e.g. --cached/--index.  If we are
3248 * applying a non-git patch that incrementally updates the tree,
3249 * we read from the result of a previous diff.
3250 */
3251static int load_preimage(struct image *image,
3252                         struct patch *patch, struct stat *st,
3253                         const struct cache_entry *ce)
3254{
3255        struct strbuf buf = STRBUF_INIT;
3256        size_t len;
3257        char *img;
3258        struct patch *previous;
3259        int status;
3260
3261        previous = previous_patch(patch, &status);
3262        if (status)
3263                return error(_("path %s has been renamed/deleted"),
3264                             patch->old_name);
3265        if (previous) {
3266                /* We have a patched copy in memory; use that. */
3267                strbuf_add(&buf, previous->result, previous->resultsize);
3268        } else {
3269                status = load_patch_target(&buf, ce, st,
3270                                           patch->old_name, patch->old_mode);
3271                if (status < 0)
3272                        return status;
3273                else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3274                        /*
3275                         * There is no way to apply subproject
3276                         * patch without looking at the index.
3277                         * NEEDSWORK: shouldn't this be flagged
3278                         * as an error???
3279                         */
3280                        free_fragment_list(patch->fragments);
3281                        patch->fragments = NULL;
3282                } else if (status) {
3283                        return error(_("read of %s failed"), patch->old_name);
3284                }
3285        }
3286
3287        img = strbuf_detach(&buf, &len);
3288        prepare_image(image, img, len, !patch->is_binary);
3289        return 0;
3290}
3291
3292static int three_way_merge(struct image *image,
3293                           char *path,
3294                           const unsigned char *base,
3295                           const unsigned char *ours,
3296                           const unsigned char *theirs)
3297{
3298        mmfile_t base_file, our_file, their_file;
3299        mmbuffer_t result = { NULL };
3300        int status;
3301
3302        read_mmblob(&base_file, base);
3303        read_mmblob(&our_file, ours);
3304        read_mmblob(&their_file, theirs);
3305        status = ll_merge(&result, path,
3306                          &base_file, "base",
3307                          &our_file, "ours",
3308                          &their_file, "theirs", NULL);
3309        free(base_file.ptr);
3310        free(our_file.ptr);
3311        free(their_file.ptr);
3312        if (status < 0 || !result.ptr) {
3313                free(result.ptr);
3314                return -1;
3315        }
3316        clear_image(image);
3317        image->buf = result.ptr;
3318        image->len = result.size;
3319
3320        return status;
3321}
3322
3323/*
3324 * When directly falling back to add/add three-way merge, we read from
3325 * the current contents of the new_name.  In no cases other than that
3326 * this function will be called.
3327 */
3328static int load_current(struct image *image, struct patch *patch)
3329{
3330        struct strbuf buf = STRBUF_INIT;
3331        int status, pos;
3332        size_t len;
3333        char *img;
3334        struct stat st;
3335        struct cache_entry *ce;
3336        char *name = patch->new_name;
3337        unsigned mode = patch->new_mode;
3338
3339        if (!patch->is_new)
3340                die("BUG: patch to %s is not a creation", patch->old_name);
3341
3342        pos = cache_name_pos(name, strlen(name));
3343        if (pos < 0)
3344                return error(_("%s: does not exist in index"), name);
3345        ce = active_cache[pos];
3346        if (lstat(name, &st)) {
3347                if (errno != ENOENT)
3348                        return error(_("%s: %s"), name, strerror(errno));
3349                if (checkout_target(&the_index, ce, &st))
3350                        return -1;
3351        }
3352        if (verify_index_match(ce, &st))
3353                return error(_("%s: does not match index"), name);
3354
3355        status = load_patch_target(&buf, ce, &st, name, mode);
3356        if (status < 0)
3357                return status;
3358        else if (status)
3359                return -1;
3360        img = strbuf_detach(&buf, &len);
3361        prepare_image(image, img, len, !patch->is_binary);
3362        return 0;
3363}
3364
3365static int try_threeway(struct image *image, struct patch *patch,
3366                        struct stat *st, const struct cache_entry *ce)
3367{
3368        unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3369        struct strbuf buf = STRBUF_INIT;
3370        size_t len;
3371        int status;
3372        char *img;
3373        struct image tmp_image;
3374
3375        /* No point falling back to 3-way merge in these cases */
3376        if (patch->is_delete ||
3377            S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3378                return -1;
3379
3380        /* Preimage the patch was prepared for */
3381        if (patch->is_new)
3382                write_sha1_file("", 0, blob_type, pre_sha1);
3383        else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3384                 read_blob_object(&buf, pre_sha1, patch->old_mode))
3385                return error("repository lacks the necessary blob to fall back on 3-way merge.");
3386
3387        fprintf(stderr, "Falling back to three-way merge...\n");
3388
3389        img = strbuf_detach(&buf, &len);
3390        prepare_image(&tmp_image, img, len, 1);
3391        /* Apply the patch to get the post image */
3392        if (apply_fragments(&tmp_image, patch) < 0) {
3393                clear_image(&tmp_image);
3394                return -1;
3395        }
3396        /* post_sha1[] is theirs */
3397        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3398        clear_image(&tmp_image);
3399
3400        /* our_sha1[] is ours */
3401        if (patch->is_new) {
3402                if (load_current(&tmp_image, patch))
3403                        return error("cannot read the current contents of '%s'",
3404                                     patch->new_name);
3405        } else {
3406                if (load_preimage(&tmp_image, patch, st, ce))
3407                        return error("cannot read the current contents of '%s'",
3408                                     patch->old_name);
3409        }
3410        write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3411        clear_image(&tmp_image);
3412
3413        /* in-core three-way merge between post and our using pre as base */
3414        status = three_way_merge(image, patch->new_name,
3415                                 pre_sha1, our_sha1, post_sha1);
3416        if (status < 0) {
3417                fprintf(stderr, "Failed to fall back on three-way merge...\n");
3418                return status;
3419        }
3420
3421        if (status) {
3422                patch->conflicted_threeway = 1;
3423                if (patch->is_new)
3424                        hashclr(patch->threeway_stage[0]);
3425                else
3426                        hashcpy(patch->threeway_stage[0], pre_sha1);
3427                hashcpy(patch->threeway_stage[1], our_sha1);
3428                hashcpy(patch->threeway_stage[2], post_sha1);
3429                fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3430        } else {
3431                fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3432        }
3433        return 0;
3434}
3435
3436static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3437{
3438        struct image image;
3439
3440        if (load_preimage(&image, patch, st, ce) < 0)
3441                return -1;
3442
3443        if (patch->direct_to_threeway ||
3444            apply_fragments(&image, patch) < 0) {
3445                /* Note: with --reject, apply_fragments() returns 0 */
3446                if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3447                        return -1;
3448        }
3449        patch->result = image.buf;
3450        patch->resultsize = image.len;
3451        add_to_fn_table(patch);
3452        free(image.line_allocated);
3453
3454        if (0 < patch->is_delete && patch->resultsize)
3455                return error(_("removal patch leaves file contents"));
3456
3457        return 0;
3458}
3459
3460/*
3461 * If "patch" that we are looking at modifies or deletes what we have,
3462 * we would want it not to lose any local modification we have, either
3463 * in the working tree or in the index.
3464 *
3465 * This also decides if a non-git patch is a creation patch or a
3466 * modification to an existing empty file.  We do not check the state
3467 * of the current tree for a creation patch in this function; the caller
3468 * check_patch() separately makes sure (and errors out otherwise) that
3469 * the path the patch creates does not exist in the current tree.
3470 */
3471static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3472{
3473        const char *old_name = patch->old_name;
3474        struct patch *previous = NULL;
3475        int stat_ret = 0, status;
3476        unsigned st_mode = 0;
3477
3478        if (!old_name)
3479                return 0;
3480
3481        assert(patch->is_new <= 0);
3482        previous = previous_patch(patch, &status);
3483
3484        if (status)
3485                return error(_("path %s has been renamed/deleted"), old_name);
3486        if (previous) {
3487                st_mode = previous->new_mode;
3488        } else if (!cached) {
3489                stat_ret = lstat(old_name, st);
3490                if (stat_ret && errno != ENOENT)
3491                        return error(_("%s: %s"), old_name, strerror(errno));
3492        }
3493
3494        if (check_index && !previous) {
3495                int pos = cache_name_pos(old_name, strlen(old_name));
3496                if (pos < 0) {
3497                        if (patch->is_new < 0)
3498                                goto is_new;
3499                        return error(_("%s: does not exist in index"), old_name);
3500                }
3501                *ce = active_cache[pos];
3502                if (stat_ret < 0) {
3503                        if (checkout_target(&the_index, *ce, st))
3504                                return -1;
3505                }
3506                if (!cached && verify_index_match(*ce, st))
3507                        return error(_("%s: does not match index"), old_name);
3508                if (cached)
3509                        st_mode = (*ce)->ce_mode;
3510        } else if (stat_ret < 0) {
3511                if (patch->is_new < 0)
3512                        goto is_new;
3513                return error(_("%s: %s"), old_name, strerror(errno));
3514        }
3515
3516        if (!cached && !previous)
3517                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3518
3519        if (patch->is_new < 0)
3520                patch->is_new = 0;
3521        if (!patch->old_mode)
3522                patch->old_mode = st_mode;
3523        if ((st_mode ^ patch->old_mode) & S_IFMT)
3524                return error(_("%s: wrong type"), old_name);
3525        if (st_mode != patch->old_mode)
3526                warning(_("%s has type %o, expected %o"),
3527                        old_name, st_mode, patch->old_mode);
3528        if (!patch->new_mode && !patch->is_delete)
3529                patch->new_mode = st_mode;
3530        return 0;
3531
3532 is_new:
3533        patch->is_new = 1;
3534        patch->is_delete = 0;
3535        free(patch->old_name);
3536        patch->old_name = NULL;
3537        return 0;
3538}
3539
3540
3541#define EXISTS_IN_INDEX 1
3542#define EXISTS_IN_WORKTREE 2
3543
3544static int check_to_create(const char *new_name, int ok_if_exists)
3545{
3546        struct stat nst;
3547
3548        if (check_index &&
3549            cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3550            !ok_if_exists)
3551                return EXISTS_IN_INDEX;
3552        if (cached)
3553                return 0;
3554
3555        if (!lstat(new_name, &nst)) {
3556                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3557                        return 0;
3558                /*
3559                 * A leading component of new_name might be a symlink
3560                 * that is going to be removed with this patch, but
3561                 * still pointing at somewhere that has the path.
3562                 * In such a case, path "new_name" does not exist as
3563                 * far as git is concerned.
3564                 */
3565                if (has_symlink_leading_path(new_name, strlen(new_name)))
3566                        return 0;
3567
3568                return EXISTS_IN_WORKTREE;
3569        } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3570                return error("%s: %s", new_name, strerror(errno));
3571        }
3572        return 0;
3573}
3574
3575/*
3576 * We need to keep track of how symlinks in the preimage are
3577 * manipulated by the patches.  A patch to add a/b/c where a/b
3578 * is a symlink should not be allowed to affect the directory
3579 * the symlink points at, but if the same patch removes a/b,
3580 * it is perfectly fine, as the patch removes a/b to make room
3581 * to create a directory a/b so that a/b/c can be created.
3582 */
3583static struct string_list symlink_changes;
3584#define SYMLINK_GOES_AWAY 01
3585#define SYMLINK_IN_RESULT 02
3586
3587static uintptr_t register_symlink_changes(const char *path, uintptr_t what)
3588{
3589        struct string_list_item *ent;
3590
3591        ent = string_list_lookup(&symlink_changes, path);
3592        if (!ent) {
3593                ent = string_list_insert(&symlink_changes, path);
3594                ent->util = (void *)0;
3595        }
3596        ent->util = (void *)(what | ((uintptr_t)ent->util));
3597        return (uintptr_t)ent->util;
3598}
3599
3600static uintptr_t check_symlink_changes(const char *path)
3601{
3602        struct string_list_item *ent;
3603
3604        ent = string_list_lookup(&symlink_changes, path);
3605        if (!ent)
3606                return 0;
3607        return (uintptr_t)ent->util;
3608}
3609
3610static void prepare_symlink_changes(struct patch *patch)
3611{
3612        for ( ; patch; patch = patch->next) {
3613                if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3614                    (patch->is_rename || patch->is_delete))
3615                        /* the symlink at patch->old_name is removed */
3616                        register_symlink_changes(patch->old_name, SYMLINK_GOES_AWAY);
3617
3618                if (patch->new_name && S_ISLNK(patch->new_mode))
3619                        /* the symlink at patch->new_name is created or remains */
3620                        register_symlink_changes(patch->new_name, SYMLINK_IN_RESULT);
3621        }
3622}
3623
3624static int path_is_beyond_symlink_1(struct strbuf *name)
3625{
3626        do {
3627                unsigned int change;
3628
3629                while (--name->len && name->buf[name->len] != '/')
3630                        ; /* scan backwards */
3631                if (!name->len)
3632                        break;
3633                name->buf[name->len] = '\0';
3634                change = check_symlink_changes(name->buf);
3635                if (change & SYMLINK_IN_RESULT)
3636                        return 1;
3637                if (change & SYMLINK_GOES_AWAY)
3638                        /*
3639                         * This cannot be "return 0", because we may
3640                         * see a new one created at a higher level.
3641                         */
3642                        continue;
3643
3644                /* otherwise, check the preimage */
3645                if (check_index) {
3646                        struct cache_entry *ce;
3647
3648                        ce = cache_file_exists(name->buf, name->len, ignore_case);
3649                        if (ce && S_ISLNK(ce->ce_mode))
3650                                return 1;
3651                } else {
3652                        struct stat st;
3653                        if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3654                                return 1;
3655                }
3656        } while (1);
3657        return 0;
3658}
3659
3660static int path_is_beyond_symlink(const char *name_)
3661{
3662        int ret;
3663        struct strbuf name = STRBUF_INIT;
3664
3665        assert(*name_ != '\0');
3666        strbuf_addstr(&name, name_);
3667        ret = path_is_beyond_symlink_1(&name);
3668        strbuf_release(&name);
3669
3670        return ret;
3671}
3672
3673static void die_on_unsafe_path(struct patch *patch)
3674{
3675        const char *old_name = NULL;
3676        const char *new_name = NULL;
3677        if (patch->is_delete)
3678                old_name = patch->old_name;
3679        else if (!patch->is_new && !patch->is_copy)
3680                old_name = patch->old_name;
3681        if (!patch->is_delete)
3682                new_name = patch->new_name;
3683
3684        if (old_name && !verify_path(old_name))
3685                die(_("invalid path '%s'"), old_name);
3686        if (new_name && !verify_path(new_name))
3687                die(_("invalid path '%s'"), new_name);
3688}
3689
3690/*
3691 * Check and apply the patch in-core; leave the result in patch->result
3692 * for the caller to write it out to the final destination.
3693 */
3694static int check_patch(struct patch *patch)
3695{
3696        struct stat st;
3697        const char *old_name = patch->old_name;
3698        const char *new_name = patch->new_name;
3699        const char *name = old_name ? old_name : new_name;
3700        struct cache_entry *ce = NULL;
3701        struct patch *tpatch;
3702        int ok_if_exists;
3703        int status;
3704
3705        patch->rejected = 1; /* we will drop this after we succeed */
3706
3707        status = check_preimage(patch, &ce, &st);
3708        if (status)
3709                return status;
3710        old_name = patch->old_name;
3711
3712        /*
3713         * A type-change diff is always split into a patch to delete
3714         * old, immediately followed by a patch to create new (see
3715         * diff.c::run_diff()); in such a case it is Ok that the entry
3716         * to be deleted by the previous patch is still in the working
3717         * tree and in the index.
3718         *
3719         * A patch to swap-rename between A and B would first rename A
3720         * to B and then rename B to A.  While applying the first one,
3721         * the presence of B should not stop A from getting renamed to
3722         * B; ask to_be_deleted() about the later rename.  Removal of
3723         * B and rename from A to B is handled the same way by asking
3724         * was_deleted().
3725         */
3726        if ((tpatch = in_fn_table(new_name)) &&
3727            (was_deleted(tpatch) || to_be_deleted(tpatch)))
3728                ok_if_exists = 1;
3729        else
3730                ok_if_exists = 0;
3731
3732        if (new_name &&
3733            ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3734                int err = check_to_create(new_name, ok_if_exists);
3735
3736                if (err && threeway) {
3737                        patch->direct_to_threeway = 1;
3738                } else switch (err) {
3739                case 0:
3740                        break; /* happy */
3741                case EXISTS_IN_INDEX:
3742                        return error(_("%s: already exists in index"), new_name);
3743                        break;
3744                case EXISTS_IN_WORKTREE:
3745                        return error(_("%s: already exists in working directory"),
3746                                     new_name);
3747                default:
3748                        return err;
3749                }
3750
3751                if (!patch->new_mode) {
3752                        if (0 < patch->is_new)
3753                                patch->new_mode = S_IFREG | 0644;
3754                        else
3755                                patch->new_mode = patch->old_mode;
3756                }
3757        }
3758
3759        if (new_name && old_name) {
3760                int same = !strcmp(old_name, new_name);
3761                if (!patch->new_mode)
3762                        patch->new_mode = patch->old_mode;
3763                if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3764                        if (same)
3765                                return error(_("new mode (%o) of %s does not "
3766                                               "match old mode (%o)"),
3767                                        patch->new_mode, new_name,
3768                                        patch->old_mode);
3769                        else
3770                                return error(_("new mode (%o) of %s does not "
3771                                               "match old mode (%o) of %s"),
3772                                        patch->new_mode, new_name,
3773                                        patch->old_mode, old_name);
3774                }
3775        }
3776
3777        if (!unsafe_paths)
3778                die_on_unsafe_path(patch);
3779
3780        /*
3781         * An attempt to read from or delete a path that is beyond a
3782         * symbolic link will be prevented by load_patch_target() that
3783         * is called at the beginning of apply_data() so we do not
3784         * have to worry about a patch marked with "is_delete" bit
3785         * here.  We however need to make sure that the patch result
3786         * is not deposited to a path that is beyond a symbolic link
3787         * here.
3788         */
3789        if (!patch->is_delete && path_is_beyond_symlink(patch->new_name))
3790                return error(_("affected file '%s' is beyond a symbolic link"),
3791                             patch->new_name);
3792
3793        if (apply_data(patch, &st, ce) < 0)
3794                return error(_("%s: patch does not apply"), name);
3795        patch->rejected = 0;
3796        return 0;
3797}
3798
3799static int check_patch_list(struct patch *patch)
3800{
3801        int err = 0;
3802
3803        prepare_symlink_changes(patch);
3804        prepare_fn_table(patch);
3805        while (patch) {
3806                if (apply_verbosely)
3807                        say_patch_name(stderr,
3808                                       _("Checking patch %s..."), patch);
3809                err |= check_patch(patch);
3810                patch = patch->next;
3811        }
3812        return err;
3813}
3814
3815/* This function tries to read the sha1 from the current index */
3816static int get_current_sha1(const char *path, unsigned char *sha1)
3817{
3818        int pos;
3819
3820        if (read_cache() < 0)
3821                return -1;
3822        pos = cache_name_pos(path, strlen(path));
3823        if (pos < 0)
3824                return -1;
3825        hashcpy(sha1, active_cache[pos]->sha1);
3826        return 0;
3827}
3828
3829static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3830{
3831        /*
3832         * A usable gitlink patch has only one fragment (hunk) that looks like:
3833         * @@ -1 +1 @@
3834         * -Subproject commit <old sha1>
3835         * +Subproject commit <new sha1>
3836         * or
3837         * @@ -1 +0,0 @@
3838         * -Subproject commit <old sha1>
3839         * for a removal patch.
3840         */
3841        struct fragment *hunk = p->fragments;
3842        static const char heading[] = "-Subproject commit ";
3843        char *preimage;
3844
3845        if (/* does the patch have only one hunk? */
3846            hunk && !hunk->next &&
3847            /* is its preimage one line? */
3848            hunk->oldpos == 1 && hunk->oldlines == 1 &&
3849            /* does preimage begin with the heading? */
3850            (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3851            starts_with(++preimage, heading) &&
3852            /* does it record full SHA-1? */
3853            !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3854            preimage[sizeof(heading) + 40 - 1] == '\n' &&
3855            /* does the abbreviated name on the index line agree with it? */
3856            starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3857                return 0; /* it all looks fine */
3858
3859        /* we may have full object name on the index line */
3860        return get_sha1_hex(p->old_sha1_prefix, sha1);
3861}
3862
3863/* Build an index that contains the just the files needed for a 3way merge */
3864static void build_fake_ancestor(struct patch *list, const char *filename)
3865{
3866        struct patch *patch;
3867        struct index_state result = { NULL };
3868        static struct lock_file lock;
3869
3870        /* Once we start supporting the reverse patch, it may be
3871         * worth showing the new sha1 prefix, but until then...
3872         */
3873        for (patch = list; patch; patch = patch->next) {
3874                unsigned char sha1[20];
3875                struct cache_entry *ce;
3876                const char *name;
3877
3878                name = patch->old_name ? patch->old_name : patch->new_name;
3879                if (0 < patch->is_new)
3880                        continue;
3881
3882                if (S_ISGITLINK(patch->old_mode)) {
3883                        if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3884                                ; /* ok, the textual part looks sane */
3885                        else
3886                                die("sha1 information is lacking or useless for submodule %s",
3887                                    name);
3888                } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3889                        ; /* ok */
3890                } else if (!patch->lines_added && !patch->lines_deleted) {
3891                        /* mode-only change: update the current */
3892                        if (get_current_sha1(patch->old_name, sha1))
3893                                die("mode change for %s, which is not "
3894                                    "in current HEAD", name);
3895                } else
3896                        die("sha1 information is lacking or useless "
3897                            "(%s).", name);
3898
3899                ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3900                if (!ce)
3901                        die(_("make_cache_entry failed for path '%s'"), name);
3902                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3903                        die ("Could not add %s to temporary index", name);
3904        }
3905
3906        hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);
3907        if (write_locked_index(&result, &lock, COMMIT_LOCK))
3908                die ("Could not write temporary index to %s", filename);
3909
3910        discard_index(&result);
3911}
3912
3913static void stat_patch_list(struct patch *patch)
3914{
3915        int files, adds, dels;
3916
3917        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3918                files++;
3919                adds += patch->lines_added;
3920                dels += patch->lines_deleted;
3921                show_stats(patch);
3922        }
3923
3924        print_stat_summary(stdout, files, adds, dels);
3925}
3926
3927static void numstat_patch_list(struct patch *patch)
3928{
3929        for ( ; patch; patch = patch->next) {
3930                const char *name;
3931                name = patch->new_name ? patch->new_name : patch->old_name;
3932                if (patch->is_binary)
3933                        printf("-\t-\t");
3934                else
3935                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3936                write_name_quoted(name, stdout, line_termination);
3937        }
3938}
3939
3940static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3941{
3942        if (mode)
3943                printf(" %s mode %06o %s\n", newdelete, mode, name);
3944        else
3945                printf(" %s %s\n", newdelete, name);
3946}
3947
3948static void show_mode_change(struct patch *p, int show_name)
3949{
3950        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3951                if (show_name)
3952                        printf(" mode change %06o => %06o %s\n",
3953                               p->old_mode, p->new_mode, p->new_name);
3954                else
3955                        printf(" mode change %06o => %06o\n",
3956                               p->old_mode, p->new_mode);
3957        }
3958}
3959
3960static void show_rename_copy(struct patch *p)
3961{
3962        const char *renamecopy = p->is_rename ? "rename" : "copy";
3963        const char *old, *new;
3964
3965        /* Find common prefix */
3966        old = p->old_name;
3967        new = p->new_name;
3968        while (1) {
3969                const char *slash_old, *slash_new;
3970                slash_old = strchr(old, '/');
3971                slash_new = strchr(new, '/');
3972                if (!slash_old ||
3973                    !slash_new ||
3974                    slash_old - old != slash_new - new ||
3975                    memcmp(old, new, slash_new - new))
3976                        break;
3977                old = slash_old + 1;
3978                new = slash_new + 1;
3979        }
3980        /* p->old_name thru old is the common prefix, and old and new
3981         * through the end of names are renames
3982         */
3983        if (old != p->old_name)
3984                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3985                       (int)(old - p->old_name), p->old_name,
3986                       old, new, p->score);
3987        else
3988                printf(" %s %s => %s (%d%%)\n", renamecopy,
3989                       p->old_name, p->new_name, p->score);
3990        show_mode_change(p, 0);
3991}
3992
3993static void summary_patch_list(struct patch *patch)
3994{
3995        struct patch *p;
3996
3997        for (p = patch; p; p = p->next) {
3998                if (p->is_new)
3999                        show_file_mode_name("create", p->new_mode, p->new_name);
4000                else if (p->is_delete)
4001                        show_file_mode_name("delete", p->old_mode, p->old_name);
4002                else {
4003                        if (p->is_rename || p->is_copy)
4004                                show_rename_copy(p);
4005                        else {
4006                                if (p->score) {
4007                                        printf(" rewrite %s (%d%%)\n",
4008                                               p->new_name, p->score);
4009                                        show_mode_change(p, 0);
4010                                }
4011                                else
4012                                        show_mode_change(p, 1);
4013                        }
4014                }
4015        }
4016}
4017
4018static void patch_stats(struct patch *patch)
4019{
4020        int lines = patch->lines_added + patch->lines_deleted;
4021
4022        if (lines > max_change)
4023                max_change = lines;
4024        if (patch->old_name) {
4025                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4026                if (!len)
4027                        len = strlen(patch->old_name);
4028                if (len > max_len)
4029                        max_len = len;
4030        }
4031        if (patch->new_name) {
4032                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4033                if (!len)
4034                        len = strlen(patch->new_name);
4035                if (len > max_len)
4036                        max_len = len;
4037        }
4038}
4039
4040static void remove_file(struct patch *patch, int rmdir_empty)
4041{
4042        if (update_index) {
4043                if (remove_file_from_cache(patch->old_name) < 0)
4044                        die(_("unable to remove %s from index"), patch->old_name);
4045        }
4046        if (!cached) {
4047                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4048                        remove_path(patch->old_name);
4049                }
4050        }
4051}
4052
4053static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
4054{
4055        struct stat st;
4056        struct cache_entry *ce;
4057        int namelen = strlen(path);
4058        unsigned ce_size = cache_entry_size(namelen);
4059
4060        if (!update_index)
4061                return;
4062
4063        ce = xcalloc(1, ce_size);
4064        memcpy(ce->name, path, namelen);
4065        ce->ce_mode = create_ce_mode(mode);
4066        ce->ce_flags = create_ce_flags(0);
4067        ce->ce_namelen = namelen;
4068        if (S_ISGITLINK(mode)) {
4069                const char *s;
4070
4071                if (!skip_prefix(buf, "Subproject commit ", &s) ||
4072                    get_sha1_hex(s, ce->sha1))
4073                        die(_("corrupt patch for submodule %s"), path);
4074        } else {
4075                if (!cached) {
4076                        if (lstat(path, &st) < 0)
4077                                die_errno(_("unable to stat newly created file '%s'"),
4078                                          path);
4079                        fill_stat_cache_info(ce, &st);
4080                }
4081                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
4082                        die(_("unable to create backing store for newly created file %s"), path);
4083        }
4084        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4085                die(_("unable to add cache entry for %s"), path);
4086}
4087
4088static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4089{
4090        int fd;
4091        struct strbuf nbuf = STRBUF_INIT;
4092
4093        if (S_ISGITLINK(mode)) {
4094                struct stat st;
4095                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4096                        return 0;
4097                return mkdir(path, 0777);
4098        }
4099
4100        if (has_symlinks && S_ISLNK(mode))
4101                /* Although buf:size is counted string, it also is NUL
4102                 * terminated.
4103                 */
4104                return symlink(buf, path);
4105
4106        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4107        if (fd < 0)
4108                return -1;
4109
4110        if (convert_to_working_tree(path, buf, size, &nbuf)) {
4111                size = nbuf.len;
4112                buf  = nbuf.buf;
4113        }
4114        write_or_die(fd, buf, size);
4115        strbuf_release(&nbuf);
4116
4117        if (close(fd) < 0)
4118                die_errno(_("closing file '%s'"), path);
4119        return 0;
4120}
4121
4122/*
4123 * We optimistically assume that the directories exist,
4124 * which is true 99% of the time anyway. If they don't,
4125 * we create them and try again.
4126 */
4127static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
4128{
4129        if (cached)
4130                return;
4131        if (!try_create_file(path, mode, buf, size))
4132                return;
4133
4134        if (errno == ENOENT) {
4135                if (safe_create_leading_directories(path))
4136                        return;
4137                if (!try_create_file(path, mode, buf, size))
4138                        return;
4139        }
4140
4141        if (errno == EEXIST || errno == EACCES) {
4142                /* We may be trying to create a file where a directory
4143                 * used to be.
4144                 */
4145                struct stat st;
4146                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4147                        errno = EEXIST;
4148        }
4149
4150        if (errno == EEXIST) {
4151                unsigned int nr = getpid();
4152
4153                for (;;) {
4154                        char newpath[PATH_MAX];
4155                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4156                        if (!try_create_file(newpath, mode, buf, size)) {
4157                                if (!rename(newpath, path))
4158                                        return;
4159                                unlink_or_warn(newpath);
4160                                break;
4161                        }
4162                        if (errno != EEXIST)
4163                                break;
4164                        ++nr;
4165                }
4166        }
4167        die_errno(_("unable to write file '%s' mode %o"), path, mode);
4168}
4169
4170static void add_conflicted_stages_file(struct patch *patch)
4171{
4172        int stage, namelen;
4173        unsigned ce_size, mode;
4174        struct cache_entry *ce;
4175
4176        if (!update_index)
4177                return;
4178        namelen = strlen(patch->new_name);
4179        ce_size = cache_entry_size(namelen);
4180        mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4181
4182        remove_file_from_cache(patch->new_name);
4183        for (stage = 1; stage < 4; stage++) {
4184                if (is_null_sha1(patch->threeway_stage[stage - 1]))
4185                        continue;
4186                ce = xcalloc(1, ce_size);
4187                memcpy(ce->name, patch->new_name, namelen);
4188                ce->ce_mode = create_ce_mode(mode);
4189                ce->ce_flags = create_ce_flags(stage);
4190                ce->ce_namelen = namelen;
4191                hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
4192                if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4193                        die(_("unable to add cache entry for %s"), patch->new_name);
4194        }
4195}
4196
4197static void create_file(struct patch *patch)
4198{
4199        char *path = patch->new_name;
4200        unsigned mode = patch->new_mode;
4201        unsigned long size = patch->resultsize;
4202        char *buf = patch->result;
4203
4204        if (!mode)
4205                mode = S_IFREG | 0644;
4206        create_one_file(path, mode, buf, size);
4207
4208        if (patch->conflicted_threeway)
4209                add_conflicted_stages_file(patch);
4210        else
4211                add_index_file(path, mode, buf, size);
4212}
4213
4214/* phase zero is to remove, phase one is to create */
4215static void write_out_one_result(struct patch *patch, int phase)
4216{
4217        if (patch->is_delete > 0) {
4218                if (phase == 0)
4219                        remove_file(patch, 1);
4220                return;
4221        }
4222        if (patch->is_new > 0 || patch->is_copy) {
4223                if (phase == 1)
4224                        create_file(patch);
4225                return;
4226        }
4227        /*
4228         * Rename or modification boils down to the same
4229         * thing: remove the old, write the new
4230         */
4231        if (phase == 0)
4232                remove_file(patch, patch->is_rename);
4233        if (phase == 1)
4234                create_file(patch);
4235}
4236
4237static int write_out_one_reject(struct patch *patch)
4238{
4239        FILE *rej;
4240        char namebuf[PATH_MAX];
4241        struct fragment *frag;
4242        int cnt = 0;
4243        struct strbuf sb = STRBUF_INIT;
4244
4245        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4246                if (!frag->rejected)
4247                        continue;
4248                cnt++;
4249        }
4250
4251        if (!cnt) {
4252                if (apply_verbosely)
4253                        say_patch_name(stderr,
4254                                       _("Applied patch %s cleanly."), patch);
4255                return 0;
4256        }
4257
4258        /* This should not happen, because a removal patch that leaves
4259         * contents are marked "rejected" at the patch level.
4260         */
4261        if (!patch->new_name)
4262                die(_("internal error"));
4263
4264        /* Say this even without --verbose */
4265        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4266                            "Applying patch %%s with %d rejects...",
4267                            cnt),
4268                    cnt);
4269        say_patch_name(stderr, sb.buf, patch);
4270        strbuf_release(&sb);
4271
4272        cnt = strlen(patch->new_name);
4273        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4274                cnt = ARRAY_SIZE(namebuf) - 5;
4275                warning(_("truncating .rej filename to %.*s.rej"),
4276                        cnt - 1, patch->new_name);
4277        }
4278        memcpy(namebuf, patch->new_name, cnt);
4279        memcpy(namebuf + cnt, ".rej", 5);
4280
4281        rej = fopen(namebuf, "w");
4282        if (!rej)
4283                return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4284
4285        /* Normal git tools never deal with .rej, so do not pretend
4286         * this is a git patch by saying --git or giving extended
4287         * headers.  While at it, maybe please "kompare" that wants
4288         * the trailing TAB and some garbage at the end of line ;-).
4289         */
4290        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4291                patch->new_name, patch->new_name);
4292        for (cnt = 1, frag = patch->fragments;
4293             frag;
4294             cnt++, frag = frag->next) {
4295                if (!frag->rejected) {
4296                        fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4297                        continue;
4298                }
4299                fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4300                fprintf(rej, "%.*s", frag->size, frag->patch);
4301                if (frag->patch[frag->size-1] != '\n')
4302                        fputc('\n', rej);
4303        }
4304        fclose(rej);
4305        return -1;
4306}
4307
4308static int write_out_results(struct patch *list)
4309{
4310        int phase;
4311        int errs = 0;
4312        struct patch *l;
4313        struct string_list cpath = STRING_LIST_INIT_DUP;
4314
4315        for (phase = 0; phase < 2; phase++) {
4316                l = list;
4317                while (l) {
4318                        if (l->rejected)
4319                                errs = 1;
4320                        else {
4321                                write_out_one_result(l, phase);
4322                                if (phase == 1) {
4323                                        if (write_out_one_reject(l))
4324                                                errs = 1;
4325                                        if (l->conflicted_threeway) {
4326                                                string_list_append(&cpath, l->new_name);
4327                                                errs = 1;
4328                                        }
4329                                }
4330                        }
4331                        l = l->next;
4332                }
4333        }
4334
4335        if (cpath.nr) {
4336                struct string_list_item *item;
4337
4338                string_list_sort(&cpath);
4339                for_each_string_list_item(item, &cpath)
4340                        fprintf(stderr, "U %s\n", item->string);
4341                string_list_clear(&cpath, 0);
4342
4343                rerere(0);
4344        }
4345
4346        return errs;
4347}
4348
4349static struct lock_file lock_file;
4350
4351#define INACCURATE_EOF  (1<<0)
4352#define RECOUNT         (1<<1)
4353
4354static int apply_patch(int fd, const char *filename, int options)
4355{
4356        size_t offset;
4357        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4358        struct patch *list = NULL, **listp = &list;
4359        int skipped_patch = 0;
4360
4361        patch_input_file = filename;
4362        read_patch_file(&buf, fd);
4363        offset = 0;
4364        while (offset < buf.len) {
4365                struct patch *patch;
4366                int nr;
4367
4368                patch = xcalloc(1, sizeof(*patch));
4369                patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4370                patch->recount =  !!(options & RECOUNT);
4371                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4372                if (nr < 0)
4373                        break;
4374                if (apply_in_reverse)
4375                        reverse_patches(patch);
4376                if (use_patch(patch)) {
4377                        patch_stats(patch);
4378                        *listp = patch;
4379                        listp = &patch->next;
4380                }
4381                else {
4382                        free_patch(patch);
4383                        skipped_patch++;
4384                }
4385                offset += nr;
4386        }
4387
4388        if (!list && !skipped_patch)
4389                die(_("unrecognized input"));
4390
4391        if (whitespace_error && (ws_error_action == die_on_ws_error))
4392                apply = 0;
4393
4394        update_index = check_index && apply;
4395        if (update_index && newfd < 0)
4396                newfd = hold_locked_index(&lock_file, 1);
4397
4398        if (check_index) {
4399                if (read_cache() < 0)
4400                        die(_("unable to read index file"));
4401        }
4402
4403        if ((check || apply) &&
4404            check_patch_list(list) < 0 &&
4405            !apply_with_reject)
4406                exit(1);
4407
4408        if (apply && write_out_results(list)) {
4409                if (apply_with_reject)
4410                        exit(1);
4411                /* with --3way, we still need to write the index out */
4412                return 1;
4413        }
4414
4415        if (fake_ancestor)
4416                build_fake_ancestor(list, fake_ancestor);
4417
4418        if (diffstat)
4419                stat_patch_list(list);
4420
4421        if (numstat)
4422                numstat_patch_list(list);
4423
4424        if (summary)
4425                summary_patch_list(list);
4426
4427        free_patch_list(list);
4428        strbuf_release(&buf);
4429        string_list_clear(&fn_table, 0);
4430        return 0;
4431}
4432
4433static void git_apply_config(void)
4434{
4435        git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
4436        git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
4437        git_config(git_default_config, NULL);
4438}
4439
4440static int option_parse_exclude(const struct option *opt,
4441                                const char *arg, int unset)
4442{
4443        add_name_limit(arg, 1);
4444        return 0;
4445}
4446
4447static int option_parse_include(const struct option *opt,
4448                                const char *arg, int unset)
4449{
4450        add_name_limit(arg, 0);
4451        has_include = 1;
4452        return 0;
4453}
4454
4455static int option_parse_p(const struct option *opt,
4456                          const char *arg, int unset)
4457{
4458        p_value = atoi(arg);
4459        p_value_known = 1;
4460        return 0;
4461}
4462
4463static int option_parse_z(const struct option *opt,
4464                          const char *arg, int unset)
4465{
4466        if (unset)
4467                line_termination = '\n';
4468        else
4469                line_termination = 0;
4470        return 0;
4471}
4472
4473static int option_parse_space_change(const struct option *opt,
4474                          const char *arg, int unset)
4475{
4476        if (unset)
4477                ws_ignore_action = ignore_ws_none;
4478        else
4479                ws_ignore_action = ignore_ws_change;
4480        return 0;
4481}
4482
4483static int option_parse_whitespace(const struct option *opt,
4484                                   const char *arg, int unset)
4485{
4486        const char **whitespace_option = opt->value;
4487
4488        *whitespace_option = arg;
4489        parse_whitespace_option(arg);
4490        return 0;
4491}
4492
4493static int option_parse_directory(const struct option *opt,
4494                                  const char *arg, int unset)
4495{
4496        root_len = strlen(arg);
4497        if (root_len && arg[root_len - 1] != '/') {
4498                char *new_root;
4499                root = new_root = xmalloc(root_len + 2);
4500                strcpy(new_root, arg);
4501                strcpy(new_root + root_len++, "/");
4502        } else
4503                root = arg;
4504        return 0;
4505}
4506
4507int cmd_apply(int argc, const char **argv, const char *prefix_)
4508{
4509        int i;
4510        int errs = 0;
4511        int is_not_gitdir = !startup_info->have_repository;
4512        int force_apply = 0;
4513
4514        const char *whitespace_option = NULL;
4515
4516        struct option builtin_apply_options[] = {
4517                { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4518                        N_("don't apply changes matching the given path"),
4519                        0, option_parse_exclude },
4520                { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4521                        N_("apply changes matching the given path"),
4522                        0, option_parse_include },
4523                { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4524                        N_("remove <num> leading slashes from traditional diff paths"),
4525                        0, option_parse_p },
4526                OPT_BOOL(0, "no-add", &no_add,
4527                        N_("ignore additions made by the patch")),
4528                OPT_BOOL(0, "stat", &diffstat,
4529                        N_("instead of applying the patch, output diffstat for the input")),
4530                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4531                OPT_NOOP_NOARG(0, "binary"),
4532                OPT_BOOL(0, "numstat", &numstat,
4533                        N_("show number of added and deleted lines in decimal notation")),
4534                OPT_BOOL(0, "summary", &summary,
4535                        N_("instead of applying the patch, output a summary for the input")),
4536                OPT_BOOL(0, "check", &check,
4537                        N_("instead of applying the patch, see if the patch is applicable")),
4538                OPT_BOOL(0, "index", &check_index,
4539                        N_("make sure the patch is applicable to the current index")),
4540                OPT_BOOL(0, "cached", &cached,
4541                        N_("apply a patch without touching the working tree")),
4542                OPT_BOOL(0, "unsafe-paths", &unsafe_paths,
4543                        N_("accept a patch that touches outside the working area")),
4544                OPT_BOOL(0, "apply", &force_apply,
4545                        N_("also apply the patch (use with --stat/--summary/--check)")),
4546                OPT_BOOL('3', "3way", &threeway,
4547                         N_( "attempt three-way merge if a patch does not apply")),
4548                OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4549                        N_("build a temporary index based on embedded index information")),
4550                { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4551                        N_("paths are separated with NUL character"),
4552                        PARSE_OPT_NOARG, option_parse_z },
4553                OPT_INTEGER('C', NULL, &p_context,
4554                                N_("ensure at least <n> lines of context match")),
4555                { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4556                        N_("detect new or modified lines that have whitespace errors"),
4557                        0, option_parse_whitespace },
4558                { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4559                        N_("ignore changes in whitespace when finding context"),
4560                        PARSE_OPT_NOARG, option_parse_space_change },
4561                { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4562                        N_("ignore changes in whitespace when finding context"),
4563                        PARSE_OPT_NOARG, option_parse_space_change },
4564                OPT_BOOL('R', "reverse", &apply_in_reverse,
4565                        N_("apply the patch in reverse")),
4566                OPT_BOOL(0, "unidiff-zero", &unidiff_zero,
4567                        N_("don't expect at least one line of context")),
4568                OPT_BOOL(0, "reject", &apply_with_reject,
4569                        N_("leave the rejected hunks in corresponding *.rej files")),
4570                OPT_BOOL(0, "allow-overlap", &allow_overlap,
4571                        N_("allow overlapping hunks")),
4572                OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4573                OPT_BIT(0, "inaccurate-eof", &options,
4574                        N_("tolerate incorrectly detected missing new-line at the end of file"),
4575                        INACCURATE_EOF),
4576                OPT_BIT(0, "recount", &options,
4577                        N_("do not trust the line counts in the hunk headers"),
4578                        RECOUNT),
4579                { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4580                        N_("prepend <root> to all filenames"),
4581                        0, option_parse_directory },
4582                OPT_END()
4583        };
4584
4585        prefix = prefix_;
4586        prefix_length = prefix ? strlen(prefix) : 0;
4587        git_apply_config();
4588        if (apply_default_whitespace)
4589                parse_whitespace_option(apply_default_whitespace);
4590        if (apply_default_ignorewhitespace)
4591                parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4592
4593        argc = parse_options(argc, argv, prefix, builtin_apply_options,
4594                        apply_usage, 0);
4595
4596        if (apply_with_reject && threeway)
4597                die("--reject and --3way cannot be used together.");
4598        if (cached && threeway)
4599                die("--cached and --3way cannot be used together.");
4600        if (threeway) {
4601                if (is_not_gitdir)
4602                        die(_("--3way outside a repository"));
4603                check_index = 1;
4604        }
4605        if (apply_with_reject)
4606                apply = apply_verbosely = 1;
4607        if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4608                apply = 0;
4609        if (check_index && is_not_gitdir)
4610                die(_("--index outside a repository"));
4611        if (cached) {
4612                if (is_not_gitdir)
4613                        die(_("--cached outside a repository"));
4614                check_index = 1;
4615        }
4616        if (check_index)
4617                unsafe_paths = 0;
4618
4619        for (i = 0; i < argc; i++) {
4620                const char *arg = argv[i];
4621                int fd;
4622
4623                if (!strcmp(arg, "-")) {
4624                        errs |= apply_patch(0, "<stdin>", options);
4625                        read_stdin = 0;
4626                        continue;
4627                } else if (0 < prefix_length)
4628                        arg = prefix_filename(prefix, prefix_length, arg);
4629
4630                fd = open(arg, O_RDONLY);
4631                if (fd < 0)
4632                        die_errno(_("can't open patch '%s'"), arg);
4633                read_stdin = 0;
4634                set_default_whitespace_mode(whitespace_option);
4635                errs |= apply_patch(fd, arg, options);
4636                close(fd);
4637        }
4638        set_default_whitespace_mode(whitespace_option);
4639        if (read_stdin)
4640                errs |= apply_patch(0, "<stdin>", options);
4641        if (whitespace_error) {
4642                if (squelch_whitespace_errors &&
4643                    squelch_whitespace_errors < whitespace_error) {
4644                        int squelched =
4645                                whitespace_error - squelch_whitespace_errors;
4646                        warning(Q_("squelched %d whitespace error",
4647                                   "squelched %d whitespace errors",
4648                                   squelched),
4649                                squelched);
4650                }
4651                if (ws_error_action == die_on_ws_error)
4652                        die(Q_("%d line adds whitespace errors.",
4653                               "%d lines add whitespace errors.",
4654                               whitespace_error),
4655                            whitespace_error);
4656                if (applied_after_fixing_ws && apply)
4657                        warning("%d line%s applied after"
4658                                " fixing whitespace errors.",
4659                                applied_after_fixing_ws,
4660                                applied_after_fixing_ws == 1 ? "" : "s");
4661                else if (whitespace_error)
4662                        warning(Q_("%d line adds whitespace errors.",
4663                                   "%d lines add whitespace errors.",
4664                                   whitespace_error),
4665                                whitespace_error);
4666        }
4667
4668        if (update_index) {
4669                if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))
4670                        die(_("Unable to write new index file"));
4671        }
4672
4673        return !!errs;
4674}