builtin / apply.con commit Merge branch 'jk/maint-http-half-auth-push' into maint-1.7.11 (7d9483c)
   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 "cache-tree.h"
  11#include "quote.h"
  12#include "blob.h"
  13#include "delta.h"
  14#include "builtin.h"
  15#include "string-list.h"
  16#include "dir.h"
  17#include "diff.h"
  18#include "parse-options.h"
  19
  20/*
  21 *  --check turns on checking that the working tree matches the
  22 *    files that are being modified, but doesn't apply the patch
  23 *  --stat does just a diffstat, and doesn't actually apply
  24 *  --numstat does numeric diffstat, and doesn't actually apply
  25 *  --index-info shows the old and new index info for paths if available.
  26 *  --index updates the cache as well.
  27 *  --cached updates only the cache without ever touching the working tree.
  28 */
  29static const char *prefix;
  30static int prefix_length = -1;
  31static int newfd = -1;
  32
  33static int unidiff_zero;
  34static int p_value = 1;
  35static int p_value_known;
  36static int check_index;
  37static int update_index;
  38static int cached;
  39static int diffstat;
  40static int numstat;
  41static int summary;
  42static int check;
  43static int apply = 1;
  44static int apply_in_reverse;
  45static int apply_with_reject;
  46static int apply_verbosely;
  47static int allow_overlap;
  48static int no_add;
  49static const char *fake_ancestor;
  50static int line_termination = '\n';
  51static unsigned int p_context = UINT_MAX;
  52static const char * const apply_usage[] = {
  53        N_("git apply [options] [<patch>...]"),
  54        NULL
  55};
  56
  57static enum ws_error_action {
  58        nowarn_ws_error,
  59        warn_on_ws_error,
  60        die_on_ws_error,
  61        correct_ws_error
  62} ws_error_action = warn_on_ws_error;
  63static int whitespace_error;
  64static int squelch_whitespace_errors = 5;
  65static int applied_after_fixing_ws;
  66
  67static enum ws_ignore {
  68        ignore_ws_none,
  69        ignore_ws_change
  70} ws_ignore_action = ignore_ws_none;
  71
  72
  73static const char *patch_input_file;
  74static const char *root;
  75static int root_len;
  76static int read_stdin = 1;
  77static int options;
  78
  79static void parse_whitespace_option(const char *option)
  80{
  81        if (!option) {
  82                ws_error_action = warn_on_ws_error;
  83                return;
  84        }
  85        if (!strcmp(option, "warn")) {
  86                ws_error_action = warn_on_ws_error;
  87                return;
  88        }
  89        if (!strcmp(option, "nowarn")) {
  90                ws_error_action = nowarn_ws_error;
  91                return;
  92        }
  93        if (!strcmp(option, "error")) {
  94                ws_error_action = die_on_ws_error;
  95                return;
  96        }
  97        if (!strcmp(option, "error-all")) {
  98                ws_error_action = die_on_ws_error;
  99                squelch_whitespace_errors = 0;
 100                return;
 101        }
 102        if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
 103                ws_error_action = correct_ws_error;
 104                return;
 105        }
 106        die(_("unrecognized whitespace option '%s'"), option);
 107}
 108
 109static void parse_ignorewhitespace_option(const char *option)
 110{
 111        if (!option || !strcmp(option, "no") ||
 112            !strcmp(option, "false") || !strcmp(option, "never") ||
 113            !strcmp(option, "none")) {
 114                ws_ignore_action = ignore_ws_none;
 115                return;
 116        }
 117        if (!strcmp(option, "change")) {
 118                ws_ignore_action = ignore_ws_change;
 119                return;
 120        }
 121        die(_("unrecognized whitespace ignore option '%s'"), option);
 122}
 123
 124static void set_default_whitespace_mode(const char *whitespace_option)
 125{
 126        if (!whitespace_option && !apply_default_whitespace)
 127                ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
 128}
 129
 130/*
 131 * For "diff-stat" like behaviour, we keep track of the biggest change
 132 * we've seen, and the longest filename. That allows us to do simple
 133 * scaling.
 134 */
 135static int max_change, max_len;
 136
 137/*
 138 * Various "current state", notably line numbers and what
 139 * file (and how) we're patching right now.. The "is_xxxx"
 140 * things are flags, where -1 means "don't know yet".
 141 */
 142static int linenr = 1;
 143
 144/*
 145 * This represents one "hunk" from a patch, starting with
 146 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 147 * patch text is pointed at by patch, and its byte length
 148 * is stored in size.  leading and trailing are the number
 149 * of context lines.
 150 */
 151struct fragment {
 152        unsigned long leading, trailing;
 153        unsigned long oldpos, oldlines;
 154        unsigned long newpos, newlines;
 155        /*
 156         * 'patch' is usually borrowed from buf in apply_patch(),
 157         * but some codepaths store an allocated buffer.
 158         */
 159        const char *patch;
 160        unsigned free_patch:1,
 161                rejected:1;
 162        int size;
 163        int linenr;
 164        struct fragment *next;
 165};
 166
 167/*
 168 * When dealing with a binary patch, we reuse "leading" field
 169 * to store the type of the binary hunk, either deflated "delta"
 170 * or deflated "literal".
 171 */
 172#define binary_patch_method leading
 173#define BINARY_DELTA_DEFLATED   1
 174#define BINARY_LITERAL_DEFLATED 2
 175
 176/*
 177 * This represents a "patch" to a file, both metainfo changes
 178 * such as creation/deletion, filemode and content changes represented
 179 * as a series of fragments.
 180 */
 181struct patch {
 182        char *new_name, *old_name, *def_name;
 183        unsigned int old_mode, new_mode;
 184        int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
 185        int rejected;
 186        unsigned ws_rule;
 187        int lines_added, lines_deleted;
 188        int score;
 189        unsigned int is_toplevel_relative:1;
 190        unsigned int inaccurate_eof:1;
 191        unsigned int is_binary:1;
 192        unsigned int is_copy:1;
 193        unsigned int is_rename:1;
 194        unsigned int recount:1;
 195        struct fragment *fragments;
 196        char *result;
 197        size_t resultsize;
 198        char old_sha1_prefix[41];
 199        char new_sha1_prefix[41];
 200        struct patch *next;
 201};
 202
 203static void free_fragment_list(struct fragment *list)
 204{
 205        while (list) {
 206                struct fragment *next = list->next;
 207                if (list->free_patch)
 208                        free((char *)list->patch);
 209                free(list);
 210                list = next;
 211        }
 212}
 213
 214static void free_patch(struct patch *patch)
 215{
 216        free_fragment_list(patch->fragments);
 217        free(patch->def_name);
 218        free(patch->old_name);
 219        free(patch->new_name);
 220        free(patch->result);
 221        free(patch);
 222}
 223
 224static void free_patch_list(struct patch *list)
 225{
 226        while (list) {
 227                struct patch *next = list->next;
 228                free_patch(list);
 229                list = next;
 230        }
 231}
 232
 233/*
 234 * A line in a file, len-bytes long (includes the terminating LF,
 235 * except for an incomplete line at the end if the file ends with
 236 * one), and its contents hashes to 'hash'.
 237 */
 238struct line {
 239        size_t len;
 240        unsigned hash : 24;
 241        unsigned flag : 8;
 242#define LINE_COMMON     1
 243#define LINE_PATCHED    2
 244};
 245
 246/*
 247 * This represents a "file", which is an array of "lines".
 248 */
 249struct image {
 250        char *buf;
 251        size_t len;
 252        size_t nr;
 253        size_t alloc;
 254        struct line *line_allocated;
 255        struct line *line;
 256};
 257
 258/*
 259 * Records filenames that have been touched, in order to handle
 260 * the case where more than one patches touch the same file.
 261 */
 262
 263static struct string_list fn_table;
 264
 265static uint32_t hash_line(const char *cp, size_t len)
 266{
 267        size_t i;
 268        uint32_t h;
 269        for (i = 0, h = 0; i < len; i++) {
 270                if (!isspace(cp[i])) {
 271                        h = h * 3 + (cp[i] & 0xff);
 272                }
 273        }
 274        return h;
 275}
 276
 277/*
 278 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 279 * whitespace difference. Returns 1 if they match, 0 otherwise
 280 */
 281static int fuzzy_matchlines(const char *s1, size_t n1,
 282                            const char *s2, size_t n2)
 283{
 284        const char *last1 = s1 + n1 - 1;
 285        const char *last2 = s2 + n2 - 1;
 286        int result = 0;
 287
 288        /* ignore line endings */
 289        while ((*last1 == '\r') || (*last1 == '\n'))
 290                last1--;
 291        while ((*last2 == '\r') || (*last2 == '\n'))
 292                last2--;
 293
 294        /* skip leading whitespace */
 295        while (isspace(*s1) && (s1 <= last1))
 296                s1++;
 297        while (isspace(*s2) && (s2 <= last2))
 298                s2++;
 299        /* early return if both lines are empty */
 300        if ((s1 > last1) && (s2 > last2))
 301                return 1;
 302        while (!result) {
 303                result = *s1++ - *s2++;
 304                /*
 305                 * Skip whitespace inside. We check for whitespace on
 306                 * both buffers because we don't want "a b" to match
 307                 * "ab"
 308                 */
 309                if (isspace(*s1) && isspace(*s2)) {
 310                        while (isspace(*s1) && s1 <= last1)
 311                                s1++;
 312                        while (isspace(*s2) && s2 <= last2)
 313                                s2++;
 314                }
 315                /*
 316                 * If we reached the end on one side only,
 317                 * lines don't match
 318                 */
 319                if (
 320                    ((s2 > last2) && (s1 <= last1)) ||
 321                    ((s1 > last1) && (s2 <= last2)))
 322                        return 0;
 323                if ((s1 > last1) && (s2 > last2))
 324                        break;
 325        }
 326
 327        return !result;
 328}
 329
 330static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 331{
 332        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 333        img->line_allocated[img->nr].len = len;
 334        img->line_allocated[img->nr].hash = hash_line(bol, len);
 335        img->line_allocated[img->nr].flag = flag;
 336        img->nr++;
 337}
 338
 339/*
 340 * "buf" has the file contents to be patched (read from various sources).
 341 * attach it to "image" and add line-based index to it.
 342 * "image" now owns the "buf".
 343 */
 344static void prepare_image(struct image *image, char *buf, size_t len,
 345                          int prepare_linetable)
 346{
 347        const char *cp, *ep;
 348
 349        memset(image, 0, sizeof(*image));
 350        image->buf = buf;
 351        image->len = len;
 352
 353        if (!prepare_linetable)
 354                return;
 355
 356        ep = image->buf + image->len;
 357        cp = image->buf;
 358        while (cp < ep) {
 359                const char *next;
 360                for (next = cp; next < ep && *next != '\n'; next++)
 361                        ;
 362                if (next < ep)
 363                        next++;
 364                add_line_info(image, cp, next - cp, 0);
 365                cp = next;
 366        }
 367        image->line = image->line_allocated;
 368}
 369
 370static void clear_image(struct image *image)
 371{
 372        free(image->buf);
 373        image->buf = NULL;
 374        image->len = 0;
 375}
 376
 377/* fmt must contain _one_ %s and no other substitution */
 378static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 379{
 380        struct strbuf sb = STRBUF_INIT;
 381
 382        if (patch->old_name && patch->new_name &&
 383            strcmp(patch->old_name, patch->new_name)) {
 384                quote_c_style(patch->old_name, &sb, NULL, 0);
 385                strbuf_addstr(&sb, " => ");
 386                quote_c_style(patch->new_name, &sb, NULL, 0);
 387        } else {
 388                const char *n = patch->new_name;
 389                if (!n)
 390                        n = patch->old_name;
 391                quote_c_style(n, &sb, NULL, 0);
 392        }
 393        fprintf(output, fmt, sb.buf);
 394        fputc('\n', output);
 395        strbuf_release(&sb);
 396}
 397
 398#define SLOP (16)
 399
 400static void read_patch_file(struct strbuf *sb, int fd)
 401{
 402        if (strbuf_read(sb, fd, 0) < 0)
 403                die_errno("git apply: failed to read");
 404
 405        /*
 406         * Make sure that we have some slop in the buffer
 407         * so that we can do speculative "memcmp" etc, and
 408         * see to it that it is NUL-filled.
 409         */
 410        strbuf_grow(sb, SLOP);
 411        memset(sb->buf + sb->len, 0, SLOP);
 412}
 413
 414static unsigned long linelen(const char *buffer, unsigned long size)
 415{
 416        unsigned long len = 0;
 417        while (size--) {
 418                len++;
 419                if (*buffer++ == '\n')
 420                        break;
 421        }
 422        return len;
 423}
 424
 425static int is_dev_null(const char *str)
 426{
 427        return !memcmp("/dev/null", str, 9) && isspace(str[9]);
 428}
 429
 430#define TERM_SPACE      1
 431#define TERM_TAB        2
 432
 433static int name_terminate(const char *name, int namelen, int c, int terminate)
 434{
 435        if (c == ' ' && !(terminate & TERM_SPACE))
 436                return 0;
 437        if (c == '\t' && !(terminate & TERM_TAB))
 438                return 0;
 439
 440        return 1;
 441}
 442
 443/* remove double slashes to make --index work with such filenames */
 444static char *squash_slash(char *name)
 445{
 446        int i = 0, j = 0;
 447
 448        if (!name)
 449                return NULL;
 450
 451        while (name[i]) {
 452                if ((name[j++] = name[i++]) == '/')
 453                        while (name[i] == '/')
 454                                i++;
 455        }
 456        name[j] = '\0';
 457        return name;
 458}
 459
 460static char *find_name_gnu(const char *line, const char *def, int p_value)
 461{
 462        struct strbuf name = STRBUF_INIT;
 463        char *cp;
 464
 465        /*
 466         * Proposed "new-style" GNU patch/diff format; see
 467         * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
 468         */
 469        if (unquote_c_style(&name, line, NULL)) {
 470                strbuf_release(&name);
 471                return NULL;
 472        }
 473
 474        for (cp = name.buf; p_value; p_value--) {
 475                cp = strchr(cp, '/');
 476                if (!cp) {
 477                        strbuf_release(&name);
 478                        return NULL;
 479                }
 480                cp++;
 481        }
 482
 483        strbuf_remove(&name, 0, cp - name.buf);
 484        if (root)
 485                strbuf_insert(&name, 0, root, root_len);
 486        return squash_slash(strbuf_detach(&name, NULL));
 487}
 488
 489static size_t sane_tz_len(const char *line, size_t len)
 490{
 491        const char *tz, *p;
 492
 493        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 494                return 0;
 495        tz = line + len - strlen(" +0500");
 496
 497        if (tz[1] != '+' && tz[1] != '-')
 498                return 0;
 499
 500        for (p = tz + 2; p != line + len; p++)
 501                if (!isdigit(*p))
 502                        return 0;
 503
 504        return line + len - tz;
 505}
 506
 507static size_t tz_with_colon_len(const char *line, size_t len)
 508{
 509        const char *tz, *p;
 510
 511        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 512                return 0;
 513        tz = line + len - strlen(" +08:00");
 514
 515        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 516                return 0;
 517        p = tz + 2;
 518        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 519            !isdigit(*p++) || !isdigit(*p++))
 520                return 0;
 521
 522        return line + len - tz;
 523}
 524
 525static size_t date_len(const char *line, size_t len)
 526{
 527        const char *date, *p;
 528
 529        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 530                return 0;
 531        p = date = line + len - strlen("72-02-05");
 532
 533        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 534            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 535            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 536                return 0;
 537
 538        if (date - line >= strlen("19") &&
 539            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 540                date -= strlen("19");
 541
 542        return line + len - date;
 543}
 544
 545static size_t short_time_len(const char *line, size_t len)
 546{
 547        const char *time, *p;
 548
 549        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 550                return 0;
 551        p = time = line + len - strlen(" 07:01:32");
 552
 553        /* Permit 1-digit hours? */
 554        if (*p++ != ' ' ||
 555            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 556            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 557            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 558                return 0;
 559
 560        return line + len - time;
 561}
 562
 563static size_t fractional_time_len(const char *line, size_t len)
 564{
 565        const char *p;
 566        size_t n;
 567
 568        /* Expected format: 19:41:17.620000023 */
 569        if (!len || !isdigit(line[len - 1]))
 570                return 0;
 571        p = line + len - 1;
 572
 573        /* Fractional seconds. */
 574        while (p > line && isdigit(*p))
 575                p--;
 576        if (*p != '.')
 577                return 0;
 578
 579        /* Hours, minutes, and whole seconds. */
 580        n = short_time_len(line, p - line);
 581        if (!n)
 582                return 0;
 583
 584        return line + len - p + n;
 585}
 586
 587static size_t trailing_spaces_len(const char *line, size_t len)
 588{
 589        const char *p;
 590
 591        /* Expected format: ' ' x (1 or more)  */
 592        if (!len || line[len - 1] != ' ')
 593                return 0;
 594
 595        p = line + len;
 596        while (p != line) {
 597                p--;
 598                if (*p != ' ')
 599                        return line + len - (p + 1);
 600        }
 601
 602        /* All spaces! */
 603        return len;
 604}
 605
 606static size_t diff_timestamp_len(const char *line, size_t len)
 607{
 608        const char *end = line + len;
 609        size_t n;
 610
 611        /*
 612         * Posix: 2010-07-05 19:41:17
 613         * GNU: 2010-07-05 19:41:17.620000023 -0500
 614         */
 615
 616        if (!isdigit(end[-1]))
 617                return 0;
 618
 619        n = sane_tz_len(line, end - line);
 620        if (!n)
 621                n = tz_with_colon_len(line, end - line);
 622        end -= n;
 623
 624        n = short_time_len(line, end - line);
 625        if (!n)
 626                n = fractional_time_len(line, end - line);
 627        end -= n;
 628
 629        n = date_len(line, end - line);
 630        if (!n) /* No date.  Too bad. */
 631                return 0;
 632        end -= n;
 633
 634        if (end == line)        /* No space before date. */
 635                return 0;
 636        if (end[-1] == '\t') {  /* Success! */
 637                end--;
 638                return line + len - end;
 639        }
 640        if (end[-1] != ' ')     /* No space before date. */
 641                return 0;
 642
 643        /* Whitespace damage. */
 644        end -= trailing_spaces_len(line, end - line);
 645        return line + len - end;
 646}
 647
 648static char *null_strdup(const char *s)
 649{
 650        return s ? xstrdup(s) : NULL;
 651}
 652
 653static char *find_name_common(const char *line, const char *def,
 654                              int p_value, const char *end, int terminate)
 655{
 656        int len;
 657        const char *start = NULL;
 658
 659        if (p_value == 0)
 660                start = line;
 661        while (line != end) {
 662                char c = *line;
 663
 664                if (!end && isspace(c)) {
 665                        if (c == '\n')
 666                                break;
 667                        if (name_terminate(start, line-start, c, terminate))
 668                                break;
 669                }
 670                line++;
 671                if (c == '/' && !--p_value)
 672                        start = line;
 673        }
 674        if (!start)
 675                return squash_slash(null_strdup(def));
 676        len = line - start;
 677        if (!len)
 678                return squash_slash(null_strdup(def));
 679
 680        /*
 681         * Generally we prefer the shorter name, especially
 682         * if the other one is just a variation of that with
 683         * something else tacked on to the end (ie "file.orig"
 684         * or "file~").
 685         */
 686        if (def) {
 687                int deflen = strlen(def);
 688                if (deflen < len && !strncmp(start, def, deflen))
 689                        return squash_slash(xstrdup(def));
 690        }
 691
 692        if (root) {
 693                char *ret = xmalloc(root_len + len + 1);
 694                strcpy(ret, root);
 695                memcpy(ret + root_len, start, len);
 696                ret[root_len + len] = '\0';
 697                return squash_slash(ret);
 698        }
 699
 700        return squash_slash(xmemdupz(start, len));
 701}
 702
 703static char *find_name(const char *line, char *def, int p_value, int terminate)
 704{
 705        if (*line == '"') {
 706                char *name = find_name_gnu(line, def, p_value);
 707                if (name)
 708                        return name;
 709        }
 710
 711        return find_name_common(line, def, p_value, NULL, terminate);
 712}
 713
 714static char *find_name_traditional(const char *line, char *def, int p_value)
 715{
 716        size_t len = strlen(line);
 717        size_t date_len;
 718
 719        if (*line == '"') {
 720                char *name = find_name_gnu(line, def, p_value);
 721                if (name)
 722                        return name;
 723        }
 724
 725        len = strchrnul(line, '\n') - line;
 726        date_len = diff_timestamp_len(line, len);
 727        if (!date_len)
 728                return find_name_common(line, def, p_value, NULL, TERM_TAB);
 729        len -= date_len;
 730
 731        return find_name_common(line, def, p_value, line + len, 0);
 732}
 733
 734static int count_slashes(const char *cp)
 735{
 736        int cnt = 0;
 737        char ch;
 738
 739        while ((ch = *cp++))
 740                if (ch == '/')
 741                        cnt++;
 742        return cnt;
 743}
 744
 745/*
 746 * Given the string after "--- " or "+++ ", guess the appropriate
 747 * p_value for the given patch.
 748 */
 749static int guess_p_value(const char *nameline)
 750{
 751        char *name, *cp;
 752        int val = -1;
 753
 754        if (is_dev_null(nameline))
 755                return -1;
 756        name = find_name_traditional(nameline, NULL, 0);
 757        if (!name)
 758                return -1;
 759        cp = strchr(name, '/');
 760        if (!cp)
 761                val = 0;
 762        else if (prefix) {
 763                /*
 764                 * Does it begin with "a/$our-prefix" and such?  Then this is
 765                 * very likely to apply to our directory.
 766                 */
 767                if (!strncmp(name, prefix, prefix_length))
 768                        val = count_slashes(prefix);
 769                else {
 770                        cp++;
 771                        if (!strncmp(cp, prefix, prefix_length))
 772                                val = count_slashes(prefix) + 1;
 773                }
 774        }
 775        free(name);
 776        return val;
 777}
 778
 779/*
 780 * Does the ---/+++ line has the POSIX timestamp after the last HT?
 781 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 782 * this such a timestamp?
 783 */
 784static int has_epoch_timestamp(const char *nameline)
 785{
 786        /*
 787         * We are only interested in epoch timestamp; any non-zero
 788         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 789         * For the same reason, the date must be either 1969-12-31 or
 790         * 1970-01-01, and the seconds part must be "00".
 791         */
 792        const char stamp_regexp[] =
 793                "^(1969-12-31|1970-01-01)"
 794                " "
 795                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 796                " "
 797                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 798        const char *timestamp = NULL, *cp, *colon;
 799        static regex_t *stamp;
 800        regmatch_t m[10];
 801        int zoneoffset;
 802        int hourminute;
 803        int status;
 804
 805        for (cp = nameline; *cp != '\n'; cp++) {
 806                if (*cp == '\t')
 807                        timestamp = cp + 1;
 808        }
 809        if (!timestamp)
 810                return 0;
 811        if (!stamp) {
 812                stamp = xmalloc(sizeof(*stamp));
 813                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 814                        warning(_("Cannot prepare timestamp regexp %s"),
 815                                stamp_regexp);
 816                        return 0;
 817                }
 818        }
 819
 820        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 821        if (status) {
 822                if (status != REG_NOMATCH)
 823                        warning(_("regexec returned %d for input: %s"),
 824                                status, timestamp);
 825                return 0;
 826        }
 827
 828        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 829        if (*colon == ':')
 830                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 831        else
 832                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 833        if (timestamp[m[3].rm_so] == '-')
 834                zoneoffset = -zoneoffset;
 835
 836        /*
 837         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 838         * (west of GMT) or 1970-01-01 (east of GMT)
 839         */
 840        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 841            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 842                return 0;
 843
 844        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 845                      strtol(timestamp + 14, NULL, 10) -
 846                      zoneoffset);
 847
 848        return ((zoneoffset < 0 && hourminute == 1440) ||
 849                (0 <= zoneoffset && !hourminute));
 850}
 851
 852/*
 853 * Get the name etc info from the ---/+++ lines of a traditional patch header
 854 *
 855 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 856 * files, we can happily check the index for a match, but for creating a
 857 * new file we should try to match whatever "patch" does. I have no idea.
 858 */
 859static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 860{
 861        char *name;
 862
 863        first += 4;     /* skip "--- " */
 864        second += 4;    /* skip "+++ " */
 865        if (!p_value_known) {
 866                int p, q;
 867                p = guess_p_value(first);
 868                q = guess_p_value(second);
 869                if (p < 0) p = q;
 870                if (0 <= p && p == q) {
 871                        p_value = p;
 872                        p_value_known = 1;
 873                }
 874        }
 875        if (is_dev_null(first)) {
 876                patch->is_new = 1;
 877                patch->is_delete = 0;
 878                name = find_name_traditional(second, NULL, p_value);
 879                patch->new_name = name;
 880        } else if (is_dev_null(second)) {
 881                patch->is_new = 0;
 882                patch->is_delete = 1;
 883                name = find_name_traditional(first, NULL, p_value);
 884                patch->old_name = name;
 885        } else {
 886                char *first_name;
 887                first_name = find_name_traditional(first, NULL, p_value);
 888                name = find_name_traditional(second, first_name, p_value);
 889                free(first_name);
 890                if (has_epoch_timestamp(first)) {
 891                        patch->is_new = 1;
 892                        patch->is_delete = 0;
 893                        patch->new_name = name;
 894                } else if (has_epoch_timestamp(second)) {
 895                        patch->is_new = 0;
 896                        patch->is_delete = 1;
 897                        patch->old_name = name;
 898                } else {
 899                        patch->old_name = name;
 900                        patch->new_name = xstrdup(name);
 901                }
 902        }
 903        if (!name)
 904                die(_("unable to find filename in patch at line %d"), linenr);
 905}
 906
 907static int gitdiff_hdrend(const char *line, struct patch *patch)
 908{
 909        return -1;
 910}
 911
 912/*
 913 * We're anal about diff header consistency, to make
 914 * sure that we don't end up having strange ambiguous
 915 * patches floating around.
 916 *
 917 * As a result, gitdiff_{old|new}name() will check
 918 * their names against any previous information, just
 919 * to make sure..
 920 */
 921#define DIFF_OLD_NAME 0
 922#define DIFF_NEW_NAME 1
 923
 924static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
 925{
 926        if (!orig_name && !isnull)
 927                return find_name(line, NULL, p_value, TERM_TAB);
 928
 929        if (orig_name) {
 930                int len;
 931                const char *name;
 932                char *another;
 933                name = orig_name;
 934                len = strlen(name);
 935                if (isnull)
 936                        die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
 937                another = find_name(line, NULL, p_value, TERM_TAB);
 938                if (!another || memcmp(another, name, len + 1))
 939                        die((side == DIFF_NEW_NAME) ?
 940                            _("git apply: bad git-diff - inconsistent new filename on line %d") :
 941                            _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
 942                free(another);
 943                return orig_name;
 944        }
 945        else {
 946                /* expect "/dev/null" */
 947                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 948                        die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
 949                return NULL;
 950        }
 951}
 952
 953static int gitdiff_oldname(const char *line, struct patch *patch)
 954{
 955        char *orig = patch->old_name;
 956        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
 957                                              DIFF_OLD_NAME);
 958        if (orig != patch->old_name)
 959                free(orig);
 960        return 0;
 961}
 962
 963static int gitdiff_newname(const char *line, struct patch *patch)
 964{
 965        char *orig = patch->new_name;
 966        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
 967                                              DIFF_NEW_NAME);
 968        if (orig != patch->new_name)
 969                free(orig);
 970        return 0;
 971}
 972
 973static int gitdiff_oldmode(const char *line, struct patch *patch)
 974{
 975        patch->old_mode = strtoul(line, NULL, 8);
 976        return 0;
 977}
 978
 979static int gitdiff_newmode(const char *line, struct patch *patch)
 980{
 981        patch->new_mode = strtoul(line, NULL, 8);
 982        return 0;
 983}
 984
 985static int gitdiff_delete(const char *line, struct patch *patch)
 986{
 987        patch->is_delete = 1;
 988        free(patch->old_name);
 989        patch->old_name = null_strdup(patch->def_name);
 990        return gitdiff_oldmode(line, patch);
 991}
 992
 993static int gitdiff_newfile(const char *line, struct patch *patch)
 994{
 995        patch->is_new = 1;
 996        free(patch->new_name);
 997        patch->new_name = null_strdup(patch->def_name);
 998        return gitdiff_newmode(line, patch);
 999}
1000
1001static int gitdiff_copysrc(const char *line, struct patch *patch)
1002{
1003        patch->is_copy = 1;
1004        free(patch->old_name);
1005        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1006        return 0;
1007}
1008
1009static int gitdiff_copydst(const char *line, struct patch *patch)
1010{
1011        patch->is_copy = 1;
1012        free(patch->new_name);
1013        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1014        return 0;
1015}
1016
1017static int gitdiff_renamesrc(const char *line, struct patch *patch)
1018{
1019        patch->is_rename = 1;
1020        free(patch->old_name);
1021        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1022        return 0;
1023}
1024
1025static int gitdiff_renamedst(const char *line, struct patch *patch)
1026{
1027        patch->is_rename = 1;
1028        free(patch->new_name);
1029        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1030        return 0;
1031}
1032
1033static int gitdiff_similarity(const char *line, struct patch *patch)
1034{
1035        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1036                patch->score = 0;
1037        return 0;
1038}
1039
1040static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1041{
1042        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1043                patch->score = 0;
1044        return 0;
1045}
1046
1047static int gitdiff_index(const char *line, struct patch *patch)
1048{
1049        /*
1050         * index line is N hexadecimal, "..", N hexadecimal,
1051         * and optional space with octal mode.
1052         */
1053        const char *ptr, *eol;
1054        int len;
1055
1056        ptr = strchr(line, '.');
1057        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1058                return 0;
1059        len = ptr - line;
1060        memcpy(patch->old_sha1_prefix, line, len);
1061        patch->old_sha1_prefix[len] = 0;
1062
1063        line = ptr + 2;
1064        ptr = strchr(line, ' ');
1065        eol = strchr(line, '\n');
1066
1067        if (!ptr || eol < ptr)
1068                ptr = eol;
1069        len = ptr - line;
1070
1071        if (40 < len)
1072                return 0;
1073        memcpy(patch->new_sha1_prefix, line, len);
1074        patch->new_sha1_prefix[len] = 0;
1075        if (*ptr == ' ')
1076                patch->old_mode = strtoul(ptr+1, NULL, 8);
1077        return 0;
1078}
1079
1080/*
1081 * This is normal for a diff that doesn't change anything: we'll fall through
1082 * into the next diff. Tell the parser to break out.
1083 */
1084static int gitdiff_unrecognized(const char *line, struct patch *patch)
1085{
1086        return -1;
1087}
1088
1089static const char *stop_at_slash(const char *line, int llen)
1090{
1091        int nslash = p_value;
1092        int i;
1093
1094        for (i = 0; i < llen; i++) {
1095                int ch = line[i];
1096                if (ch == '/' && --nslash <= 0)
1097                        return &line[i];
1098        }
1099        return NULL;
1100}
1101
1102/*
1103 * This is to extract the same name that appears on "diff --git"
1104 * line.  We do not find and return anything if it is a rename
1105 * patch, and it is OK because we will find the name elsewhere.
1106 * We need to reliably find name only when it is mode-change only,
1107 * creation or deletion of an empty file.  In any of these cases,
1108 * both sides are the same name under a/ and b/ respectively.
1109 */
1110static char *git_header_name(const char *line, int llen)
1111{
1112        const char *name;
1113        const char *second = NULL;
1114        size_t len, line_len;
1115
1116        line += strlen("diff --git ");
1117        llen -= strlen("diff --git ");
1118
1119        if (*line == '"') {
1120                const char *cp;
1121                struct strbuf first = STRBUF_INIT;
1122                struct strbuf sp = STRBUF_INIT;
1123
1124                if (unquote_c_style(&first, line, &second))
1125                        goto free_and_fail1;
1126
1127                /* advance to the first slash */
1128                cp = stop_at_slash(first.buf, first.len);
1129                /* we do not accept absolute paths */
1130                if (!cp || cp == first.buf)
1131                        goto free_and_fail1;
1132                strbuf_remove(&first, 0, cp + 1 - first.buf);
1133
1134                /*
1135                 * second points at one past closing dq of name.
1136                 * find the second name.
1137                 */
1138                while ((second < line + llen) && isspace(*second))
1139                        second++;
1140
1141                if (line + llen <= second)
1142                        goto free_and_fail1;
1143                if (*second == '"') {
1144                        if (unquote_c_style(&sp, second, NULL))
1145                                goto free_and_fail1;
1146                        cp = stop_at_slash(sp.buf, sp.len);
1147                        if (!cp || cp == sp.buf)
1148                                goto free_and_fail1;
1149                        /* They must match, otherwise ignore */
1150                        if (strcmp(cp + 1, first.buf))
1151                                goto free_and_fail1;
1152                        strbuf_release(&sp);
1153                        return strbuf_detach(&first, NULL);
1154                }
1155
1156                /* unquoted second */
1157                cp = stop_at_slash(second, line + llen - second);
1158                if (!cp || cp == second)
1159                        goto free_and_fail1;
1160                cp++;
1161                if (line + llen - cp != first.len + 1 ||
1162                    memcmp(first.buf, cp, first.len))
1163                        goto free_and_fail1;
1164                return strbuf_detach(&first, NULL);
1165
1166        free_and_fail1:
1167                strbuf_release(&first);
1168                strbuf_release(&sp);
1169                return NULL;
1170        }
1171
1172        /* unquoted first name */
1173        name = stop_at_slash(line, llen);
1174        if (!name || name == line)
1175                return NULL;
1176        name++;
1177
1178        /*
1179         * since the first name is unquoted, a dq if exists must be
1180         * the beginning of the second name.
1181         */
1182        for (second = name; second < line + llen; second++) {
1183                if (*second == '"') {
1184                        struct strbuf sp = STRBUF_INIT;
1185                        const char *np;
1186
1187                        if (unquote_c_style(&sp, second, NULL))
1188                                goto free_and_fail2;
1189
1190                        np = stop_at_slash(sp.buf, sp.len);
1191                        if (!np || np == sp.buf)
1192                                goto free_and_fail2;
1193                        np++;
1194
1195                        len = sp.buf + sp.len - np;
1196                        if (len < second - name &&
1197                            !strncmp(np, name, len) &&
1198                            isspace(name[len])) {
1199                                /* Good */
1200                                strbuf_remove(&sp, 0, np - sp.buf);
1201                                return strbuf_detach(&sp, NULL);
1202                        }
1203
1204                free_and_fail2:
1205                        strbuf_release(&sp);
1206                        return NULL;
1207                }
1208        }
1209
1210        /*
1211         * Accept a name only if it shows up twice, exactly the same
1212         * form.
1213         */
1214        second = strchr(name, '\n');
1215        if (!second)
1216                return NULL;
1217        line_len = second - name;
1218        for (len = 0 ; ; len++) {
1219                switch (name[len]) {
1220                default:
1221                        continue;
1222                case '\n':
1223                        return NULL;
1224                case '\t': case ' ':
1225                        second = stop_at_slash(name + len, line_len - len);
1226                        if (!second)
1227                                return NULL;
1228                        second++;
1229                        if (second[len] == '\n' && !strncmp(name, second, len)) {
1230                                return xmemdupz(name, len);
1231                        }
1232                }
1233        }
1234}
1235
1236/* Verify that we recognize the lines following a git header */
1237static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1238{
1239        unsigned long offset;
1240
1241        /* A git diff has explicit new/delete information, so we don't guess */
1242        patch->is_new = 0;
1243        patch->is_delete = 0;
1244
1245        /*
1246         * Some things may not have the old name in the
1247         * rest of the headers anywhere (pure mode changes,
1248         * or removing or adding empty files), so we get
1249         * the default name from the header.
1250         */
1251        patch->def_name = git_header_name(line, len);
1252        if (patch->def_name && root) {
1253                char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1254                strcpy(s, root);
1255                strcpy(s + root_len, patch->def_name);
1256                free(patch->def_name);
1257                patch->def_name = s;
1258        }
1259
1260        line += len;
1261        size -= len;
1262        linenr++;
1263        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1264                static const struct opentry {
1265                        const char *str;
1266                        int (*fn)(const char *, struct patch *);
1267                } optable[] = {
1268                        { "@@ -", gitdiff_hdrend },
1269                        { "--- ", gitdiff_oldname },
1270                        { "+++ ", gitdiff_newname },
1271                        { "old mode ", gitdiff_oldmode },
1272                        { "new mode ", gitdiff_newmode },
1273                        { "deleted file mode ", gitdiff_delete },
1274                        { "new file mode ", gitdiff_newfile },
1275                        { "copy from ", gitdiff_copysrc },
1276                        { "copy to ", gitdiff_copydst },
1277                        { "rename old ", gitdiff_renamesrc },
1278                        { "rename new ", gitdiff_renamedst },
1279                        { "rename from ", gitdiff_renamesrc },
1280                        { "rename to ", gitdiff_renamedst },
1281                        { "similarity index ", gitdiff_similarity },
1282                        { "dissimilarity index ", gitdiff_dissimilarity },
1283                        { "index ", gitdiff_index },
1284                        { "", gitdiff_unrecognized },
1285                };
1286                int i;
1287
1288                len = linelen(line, size);
1289                if (!len || line[len-1] != '\n')
1290                        break;
1291                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1292                        const struct opentry *p = optable + i;
1293                        int oplen = strlen(p->str);
1294                        if (len < oplen || memcmp(p->str, line, oplen))
1295                                continue;
1296                        if (p->fn(line + oplen, patch) < 0)
1297                                return offset;
1298                        break;
1299                }
1300        }
1301
1302        return offset;
1303}
1304
1305static int parse_num(const char *line, unsigned long *p)
1306{
1307        char *ptr;
1308
1309        if (!isdigit(*line))
1310                return 0;
1311        *p = strtoul(line, &ptr, 10);
1312        return ptr - line;
1313}
1314
1315static int parse_range(const char *line, int len, int offset, const char *expect,
1316                       unsigned long *p1, unsigned long *p2)
1317{
1318        int digits, ex;
1319
1320        if (offset < 0 || offset >= len)
1321                return -1;
1322        line += offset;
1323        len -= offset;
1324
1325        digits = parse_num(line, p1);
1326        if (!digits)
1327                return -1;
1328
1329        offset += digits;
1330        line += digits;
1331        len -= digits;
1332
1333        *p2 = 1;
1334        if (*line == ',') {
1335                digits = parse_num(line+1, p2);
1336                if (!digits)
1337                        return -1;
1338
1339                offset += digits+1;
1340                line += digits+1;
1341                len -= digits+1;
1342        }
1343
1344        ex = strlen(expect);
1345        if (ex > len)
1346                return -1;
1347        if (memcmp(line, expect, ex))
1348                return -1;
1349
1350        return offset + ex;
1351}
1352
1353static void recount_diff(const char *line, int size, struct fragment *fragment)
1354{
1355        int oldlines = 0, newlines = 0, ret = 0;
1356
1357        if (size < 1) {
1358                warning("recount: ignore empty hunk");
1359                return;
1360        }
1361
1362        for (;;) {
1363                int len = linelen(line, size);
1364                size -= len;
1365                line += len;
1366
1367                if (size < 1)
1368                        break;
1369
1370                switch (*line) {
1371                case ' ': case '\n':
1372                        newlines++;
1373                        /* fall through */
1374                case '-':
1375                        oldlines++;
1376                        continue;
1377                case '+':
1378                        newlines++;
1379                        continue;
1380                case '\\':
1381                        continue;
1382                case '@':
1383                        ret = size < 3 || prefixcmp(line, "@@ ");
1384                        break;
1385                case 'd':
1386                        ret = size < 5 || prefixcmp(line, "diff ");
1387                        break;
1388                default:
1389                        ret = -1;
1390                        break;
1391                }
1392                if (ret) {
1393                        warning(_("recount: unexpected line: %.*s"),
1394                                (int)linelen(line, size), line);
1395                        return;
1396                }
1397                break;
1398        }
1399        fragment->oldlines = oldlines;
1400        fragment->newlines = newlines;
1401}
1402
1403/*
1404 * Parse a unified diff fragment header of the
1405 * form "@@ -a,b +c,d @@"
1406 */
1407static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1408{
1409        int offset;
1410
1411        if (!len || line[len-1] != '\n')
1412                return -1;
1413
1414        /* Figure out the number of lines in a fragment */
1415        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1416        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1417
1418        return offset;
1419}
1420
1421static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1422{
1423        unsigned long offset, len;
1424
1425        patch->is_toplevel_relative = 0;
1426        patch->is_rename = patch->is_copy = 0;
1427        patch->is_new = patch->is_delete = -1;
1428        patch->old_mode = patch->new_mode = 0;
1429        patch->old_name = patch->new_name = NULL;
1430        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1431                unsigned long nextlen;
1432
1433                len = linelen(line, size);
1434                if (!len)
1435                        break;
1436
1437                /* Testing this early allows us to take a few shortcuts.. */
1438                if (len < 6)
1439                        continue;
1440
1441                /*
1442                 * Make sure we don't find any unconnected patch fragments.
1443                 * That's a sign that we didn't find a header, and that a
1444                 * patch has become corrupted/broken up.
1445                 */
1446                if (!memcmp("@@ -", line, 4)) {
1447                        struct fragment dummy;
1448                        if (parse_fragment_header(line, len, &dummy) < 0)
1449                                continue;
1450                        die(_("patch fragment without header at line %d: %.*s"),
1451                            linenr, (int)len-1, line);
1452                }
1453
1454                if (size < len + 6)
1455                        break;
1456
1457                /*
1458                 * Git patch? It might not have a real patch, just a rename
1459                 * or mode change, so we handle that specially
1460                 */
1461                if (!memcmp("diff --git ", line, 11)) {
1462                        int git_hdr_len = parse_git_header(line, len, size, patch);
1463                        if (git_hdr_len <= len)
1464                                continue;
1465                        if (!patch->old_name && !patch->new_name) {
1466                                if (!patch->def_name)
1467                                        die(Q_("git diff header lacks filename information when removing "
1468                                               "%d leading pathname component (line %d)",
1469                                               "git diff header lacks filename information when removing "
1470                                               "%d leading pathname components (line %d)",
1471                                               p_value),
1472                                            p_value, linenr);
1473                                patch->old_name = xstrdup(patch->def_name);
1474                                patch->new_name = xstrdup(patch->def_name);
1475                        }
1476                        if (!patch->is_delete && !patch->new_name)
1477                                die("git diff header lacks filename information "
1478                                    "(line %d)", linenr);
1479                        patch->is_toplevel_relative = 1;
1480                        *hdrsize = git_hdr_len;
1481                        return offset;
1482                }
1483
1484                /* --- followed by +++ ? */
1485                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1486                        continue;
1487
1488                /*
1489                 * We only accept unified patches, so we want it to
1490                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1491                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1492                 */
1493                nextlen = linelen(line + len, size - len);
1494                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1495                        continue;
1496
1497                /* Ok, we'll consider it a patch */
1498                parse_traditional_patch(line, line+len, patch);
1499                *hdrsize = len + nextlen;
1500                linenr += 2;
1501                return offset;
1502        }
1503        return -1;
1504}
1505
1506static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1507{
1508        char *err;
1509
1510        if (!result)
1511                return;
1512
1513        whitespace_error++;
1514        if (squelch_whitespace_errors &&
1515            squelch_whitespace_errors < whitespace_error)
1516                return;
1517
1518        err = whitespace_error_string(result);
1519        fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1520                patch_input_file, linenr, err, len, line);
1521        free(err);
1522}
1523
1524static void check_whitespace(const char *line, int len, unsigned ws_rule)
1525{
1526        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1527
1528        record_ws_error(result, line + 1, len - 2, linenr);
1529}
1530
1531/*
1532 * Parse a unified diff. Note that this really needs to parse each
1533 * fragment separately, since the only way to know the difference
1534 * between a "---" that is part of a patch, and a "---" that starts
1535 * the next patch is to look at the line counts..
1536 */
1537static int parse_fragment(const char *line, unsigned long size,
1538                          struct patch *patch, struct fragment *fragment)
1539{
1540        int added, deleted;
1541        int len = linelen(line, size), offset;
1542        unsigned long oldlines, newlines;
1543        unsigned long leading, trailing;
1544
1545        offset = parse_fragment_header(line, len, fragment);
1546        if (offset < 0)
1547                return -1;
1548        if (offset > 0 && patch->recount)
1549                recount_diff(line + offset, size - offset, fragment);
1550        oldlines = fragment->oldlines;
1551        newlines = fragment->newlines;
1552        leading = 0;
1553        trailing = 0;
1554
1555        /* Parse the thing.. */
1556        line += len;
1557        size -= len;
1558        linenr++;
1559        added = deleted = 0;
1560        for (offset = len;
1561             0 < size;
1562             offset += len, size -= len, line += len, linenr++) {
1563                if (!oldlines && !newlines)
1564                        break;
1565                len = linelen(line, size);
1566                if (!len || line[len-1] != '\n')
1567                        return -1;
1568                switch (*line) {
1569                default:
1570                        return -1;
1571                case '\n': /* newer GNU diff, an empty context line */
1572                case ' ':
1573                        oldlines--;
1574                        newlines--;
1575                        if (!deleted && !added)
1576                                leading++;
1577                        trailing++;
1578                        break;
1579                case '-':
1580                        if (apply_in_reverse &&
1581                            ws_error_action != nowarn_ws_error)
1582                                check_whitespace(line, len, patch->ws_rule);
1583                        deleted++;
1584                        oldlines--;
1585                        trailing = 0;
1586                        break;
1587                case '+':
1588                        if (!apply_in_reverse &&
1589                            ws_error_action != nowarn_ws_error)
1590                                check_whitespace(line, len, patch->ws_rule);
1591                        added++;
1592                        newlines--;
1593                        trailing = 0;
1594                        break;
1595
1596                /*
1597                 * We allow "\ No newline at end of file". Depending
1598                 * on locale settings when the patch was produced we
1599                 * don't know what this line looks like. The only
1600                 * thing we do know is that it begins with "\ ".
1601                 * Checking for 12 is just for sanity check -- any
1602                 * l10n of "\ No newline..." is at least that long.
1603                 */
1604                case '\\':
1605                        if (len < 12 || memcmp(line, "\\ ", 2))
1606                                return -1;
1607                        break;
1608                }
1609        }
1610        if (oldlines || newlines)
1611                return -1;
1612        fragment->leading = leading;
1613        fragment->trailing = trailing;
1614
1615        /*
1616         * If a fragment ends with an incomplete line, we failed to include
1617         * it in the above loop because we hit oldlines == newlines == 0
1618         * before seeing it.
1619         */
1620        if (12 < size && !memcmp(line, "\\ ", 2))
1621                offset += linelen(line, size);
1622
1623        patch->lines_added += added;
1624        patch->lines_deleted += deleted;
1625
1626        if (0 < patch->is_new && oldlines)
1627                return error(_("new file depends on old contents"));
1628        if (0 < patch->is_delete && newlines)
1629                return error(_("deleted file still has contents"));
1630        return offset;
1631}
1632
1633/*
1634 * We have seen "diff --git a/... b/..." header (or a traditional patch
1635 * header).  Read hunks that belong to this patch into fragments and hang
1636 * them to the given patch structure.
1637 *
1638 * The (fragment->patch, fragment->size) pair points into the memory given
1639 * by the caller, not a copy, when we return.
1640 */
1641static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1642{
1643        unsigned long offset = 0;
1644        unsigned long oldlines = 0, newlines = 0, context = 0;
1645        struct fragment **fragp = &patch->fragments;
1646
1647        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1648                struct fragment *fragment;
1649                int len;
1650
1651                fragment = xcalloc(1, sizeof(*fragment));
1652                fragment->linenr = linenr;
1653                len = parse_fragment(line, size, patch, fragment);
1654                if (len <= 0)
1655                        die(_("corrupt patch at line %d"), linenr);
1656                fragment->patch = line;
1657                fragment->size = len;
1658                oldlines += fragment->oldlines;
1659                newlines += fragment->newlines;
1660                context += fragment->leading + fragment->trailing;
1661
1662                *fragp = fragment;
1663                fragp = &fragment->next;
1664
1665                offset += len;
1666                line += len;
1667                size -= len;
1668        }
1669
1670        /*
1671         * If something was removed (i.e. we have old-lines) it cannot
1672         * be creation, and if something was added it cannot be
1673         * deletion.  However, the reverse is not true; --unified=0
1674         * patches that only add are not necessarily creation even
1675         * though they do not have any old lines, and ones that only
1676         * delete are not necessarily deletion.
1677         *
1678         * Unfortunately, a real creation/deletion patch do _not_ have
1679         * any context line by definition, so we cannot safely tell it
1680         * apart with --unified=0 insanity.  At least if the patch has
1681         * more than one hunk it is not creation or deletion.
1682         */
1683        if (patch->is_new < 0 &&
1684            (oldlines || (patch->fragments && patch->fragments->next)))
1685                patch->is_new = 0;
1686        if (patch->is_delete < 0 &&
1687            (newlines || (patch->fragments && patch->fragments->next)))
1688                patch->is_delete = 0;
1689
1690        if (0 < patch->is_new && oldlines)
1691                die(_("new file %s depends on old contents"), patch->new_name);
1692        if (0 < patch->is_delete && newlines)
1693                die(_("deleted file %s still has contents"), patch->old_name);
1694        if (!patch->is_delete && !newlines && context)
1695                fprintf_ln(stderr,
1696                           _("** warning: "
1697                             "file %s becomes empty but is not deleted"),
1698                           patch->new_name);
1699
1700        return offset;
1701}
1702
1703static inline int metadata_changes(struct patch *patch)
1704{
1705        return  patch->is_rename > 0 ||
1706                patch->is_copy > 0 ||
1707                patch->is_new > 0 ||
1708                patch->is_delete ||
1709                (patch->old_mode && patch->new_mode &&
1710                 patch->old_mode != patch->new_mode);
1711}
1712
1713static char *inflate_it(const void *data, unsigned long size,
1714                        unsigned long inflated_size)
1715{
1716        git_zstream stream;
1717        void *out;
1718        int st;
1719
1720        memset(&stream, 0, sizeof(stream));
1721
1722        stream.next_in = (unsigned char *)data;
1723        stream.avail_in = size;
1724        stream.next_out = out = xmalloc(inflated_size);
1725        stream.avail_out = inflated_size;
1726        git_inflate_init(&stream);
1727        st = git_inflate(&stream, Z_FINISH);
1728        git_inflate_end(&stream);
1729        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1730                free(out);
1731                return NULL;
1732        }
1733        return out;
1734}
1735
1736/*
1737 * Read a binary hunk and return a new fragment; fragment->patch
1738 * points at an allocated memory that the caller must free, so
1739 * it is marked as "->free_patch = 1".
1740 */
1741static struct fragment *parse_binary_hunk(char **buf_p,
1742                                          unsigned long *sz_p,
1743                                          int *status_p,
1744                                          int *used_p)
1745{
1746        /*
1747         * Expect a line that begins with binary patch method ("literal"
1748         * or "delta"), followed by the length of data before deflating.
1749         * a sequence of 'length-byte' followed by base-85 encoded data
1750         * should follow, terminated by a newline.
1751         *
1752         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1753         * and we would limit the patch line to 66 characters,
1754         * so one line can fit up to 13 groups that would decode
1755         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1756         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1757         */
1758        int llen, used;
1759        unsigned long size = *sz_p;
1760        char *buffer = *buf_p;
1761        int patch_method;
1762        unsigned long origlen;
1763        char *data = NULL;
1764        int hunk_size = 0;
1765        struct fragment *frag;
1766
1767        llen = linelen(buffer, size);
1768        used = llen;
1769
1770        *status_p = 0;
1771
1772        if (!prefixcmp(buffer, "delta ")) {
1773                patch_method = BINARY_DELTA_DEFLATED;
1774                origlen = strtoul(buffer + 6, NULL, 10);
1775        }
1776        else if (!prefixcmp(buffer, "literal ")) {
1777                patch_method = BINARY_LITERAL_DEFLATED;
1778                origlen = strtoul(buffer + 8, NULL, 10);
1779        }
1780        else
1781                return NULL;
1782
1783        linenr++;
1784        buffer += llen;
1785        while (1) {
1786                int byte_length, max_byte_length, newsize;
1787                llen = linelen(buffer, size);
1788                used += llen;
1789                linenr++;
1790                if (llen == 1) {
1791                        /* consume the blank line */
1792                        buffer++;
1793                        size--;
1794                        break;
1795                }
1796                /*
1797                 * Minimum line is "A00000\n" which is 7-byte long,
1798                 * and the line length must be multiple of 5 plus 2.
1799                 */
1800                if ((llen < 7) || (llen-2) % 5)
1801                        goto corrupt;
1802                max_byte_length = (llen - 2) / 5 * 4;
1803                byte_length = *buffer;
1804                if ('A' <= byte_length && byte_length <= 'Z')
1805                        byte_length = byte_length - 'A' + 1;
1806                else if ('a' <= byte_length && byte_length <= 'z')
1807                        byte_length = byte_length - 'a' + 27;
1808                else
1809                        goto corrupt;
1810                /* if the input length was not multiple of 4, we would
1811                 * have filler at the end but the filler should never
1812                 * exceed 3 bytes
1813                 */
1814                if (max_byte_length < byte_length ||
1815                    byte_length <= max_byte_length - 4)
1816                        goto corrupt;
1817                newsize = hunk_size + byte_length;
1818                data = xrealloc(data, newsize);
1819                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1820                        goto corrupt;
1821                hunk_size = newsize;
1822                buffer += llen;
1823                size -= llen;
1824        }
1825
1826        frag = xcalloc(1, sizeof(*frag));
1827        frag->patch = inflate_it(data, hunk_size, origlen);
1828        frag->free_patch = 1;
1829        if (!frag->patch)
1830                goto corrupt;
1831        free(data);
1832        frag->size = origlen;
1833        *buf_p = buffer;
1834        *sz_p = size;
1835        *used_p = used;
1836        frag->binary_patch_method = patch_method;
1837        return frag;
1838
1839 corrupt:
1840        free(data);
1841        *status_p = -1;
1842        error(_("corrupt binary patch at line %d: %.*s"),
1843              linenr-1, llen-1, buffer);
1844        return NULL;
1845}
1846
1847static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1848{
1849        /*
1850         * We have read "GIT binary patch\n"; what follows is a line
1851         * that says the patch method (currently, either "literal" or
1852         * "delta") and the length of data before deflating; a
1853         * sequence of 'length-byte' followed by base-85 encoded data
1854         * follows.
1855         *
1856         * When a binary patch is reversible, there is another binary
1857         * hunk in the same format, starting with patch method (either
1858         * "literal" or "delta") with the length of data, and a sequence
1859         * of length-byte + base-85 encoded data, terminated with another
1860         * empty line.  This data, when applied to the postimage, produces
1861         * the preimage.
1862         */
1863        struct fragment *forward;
1864        struct fragment *reverse;
1865        int status;
1866        int used, used_1;
1867
1868        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1869        if (!forward && !status)
1870                /* there has to be one hunk (forward hunk) */
1871                return error(_("unrecognized binary patch at line %d"), linenr-1);
1872        if (status)
1873                /* otherwise we already gave an error message */
1874                return status;
1875
1876        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1877        if (reverse)
1878                used += used_1;
1879        else if (status) {
1880                /*
1881                 * Not having reverse hunk is not an error, but having
1882                 * a corrupt reverse hunk is.
1883                 */
1884                free((void*) forward->patch);
1885                free(forward);
1886                return status;
1887        }
1888        forward->next = reverse;
1889        patch->fragments = forward;
1890        patch->is_binary = 1;
1891        return used;
1892}
1893
1894/*
1895 * Read the patch text in "buffer" taht extends for "size" bytes; stop
1896 * reading after seeing a single patch (i.e. changes to a single file).
1897 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1898 * Return the number of bytes consumed, so that the caller can call us
1899 * again for the next patch.
1900 */
1901static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1902{
1903        int hdrsize, patchsize;
1904        int offset = find_header(buffer, size, &hdrsize, patch);
1905
1906        if (offset < 0)
1907                return offset;
1908
1909        patch->ws_rule = whitespace_rule(patch->new_name
1910                                         ? patch->new_name
1911                                         : patch->old_name);
1912
1913        patchsize = parse_single_patch(buffer + offset + hdrsize,
1914                                       size - offset - hdrsize, patch);
1915
1916        if (!patchsize) {
1917                static const char *binhdr[] = {
1918                        "Binary files ",
1919                        "Files ",
1920                        NULL,
1921                };
1922                static const char git_binary[] = "GIT binary patch\n";
1923                int i;
1924                int hd = hdrsize + offset;
1925                unsigned long llen = linelen(buffer + hd, size - hd);
1926
1927                if (llen == sizeof(git_binary) - 1 &&
1928                    !memcmp(git_binary, buffer + hd, llen)) {
1929                        int used;
1930                        linenr++;
1931                        used = parse_binary(buffer + hd + llen,
1932                                            size - hd - llen, patch);
1933                        if (used)
1934                                patchsize = used + llen;
1935                        else
1936                                patchsize = 0;
1937                }
1938                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1939                        for (i = 0; binhdr[i]; i++) {
1940                                int len = strlen(binhdr[i]);
1941                                if (len < size - hd &&
1942                                    !memcmp(binhdr[i], buffer + hd, len)) {
1943                                        linenr++;
1944                                        patch->is_binary = 1;
1945                                        patchsize = llen;
1946                                        break;
1947                                }
1948                        }
1949                }
1950
1951                /* Empty patch cannot be applied if it is a text patch
1952                 * without metadata change.  A binary patch appears
1953                 * empty to us here.
1954                 */
1955                if ((apply || check) &&
1956                    (!patch->is_binary && !metadata_changes(patch)))
1957                        die(_("patch with only garbage at line %d"), linenr);
1958        }
1959
1960        return offset + hdrsize + patchsize;
1961}
1962
1963#define swap(a,b) myswap((a),(b),sizeof(a))
1964
1965#define myswap(a, b, size) do {         \
1966        unsigned char mytmp[size];      \
1967        memcpy(mytmp, &a, size);                \
1968        memcpy(&a, &b, size);           \
1969        memcpy(&b, mytmp, size);                \
1970} while (0)
1971
1972static void reverse_patches(struct patch *p)
1973{
1974        for (; p; p = p->next) {
1975                struct fragment *frag = p->fragments;
1976
1977                swap(p->new_name, p->old_name);
1978                swap(p->new_mode, p->old_mode);
1979                swap(p->is_new, p->is_delete);
1980                swap(p->lines_added, p->lines_deleted);
1981                swap(p->old_sha1_prefix, p->new_sha1_prefix);
1982
1983                for (; frag; frag = frag->next) {
1984                        swap(frag->newpos, frag->oldpos);
1985                        swap(frag->newlines, frag->oldlines);
1986                }
1987        }
1988}
1989
1990static const char pluses[] =
1991"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1992static const char minuses[]=
1993"----------------------------------------------------------------------";
1994
1995static void show_stats(struct patch *patch)
1996{
1997        struct strbuf qname = STRBUF_INIT;
1998        char *cp = patch->new_name ? patch->new_name : patch->old_name;
1999        int max, add, del;
2000
2001        quote_c_style(cp, &qname, NULL, 0);
2002
2003        /*
2004         * "scale" the filename
2005         */
2006        max = max_len;
2007        if (max > 50)
2008                max = 50;
2009
2010        if (qname.len > max) {
2011                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2012                if (!cp)
2013                        cp = qname.buf + qname.len + 3 - max;
2014                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2015        }
2016
2017        if (patch->is_binary) {
2018                printf(" %-*s |  Bin\n", max, qname.buf);
2019                strbuf_release(&qname);
2020                return;
2021        }
2022
2023        printf(" %-*s |", max, qname.buf);
2024        strbuf_release(&qname);
2025
2026        /*
2027         * scale the add/delete
2028         */
2029        max = max + max_change > 70 ? 70 - max : max_change;
2030        add = patch->lines_added;
2031        del = patch->lines_deleted;
2032
2033        if (max_change > 0) {
2034                int total = ((add + del) * max + max_change / 2) / max_change;
2035                add = (add * max + max_change / 2) / max_change;
2036                del = total - add;
2037        }
2038        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2039                add, pluses, del, minuses);
2040}
2041
2042static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2043{
2044        switch (st->st_mode & S_IFMT) {
2045        case S_IFLNK:
2046                if (strbuf_readlink(buf, path, st->st_size) < 0)
2047                        return error(_("unable to read symlink %s"), path);
2048                return 0;
2049        case S_IFREG:
2050                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2051                        return error(_("unable to open or read %s"), path);
2052                convert_to_git(path, buf->buf, buf->len, buf, 0);
2053                return 0;
2054        default:
2055                return -1;
2056        }
2057}
2058
2059/*
2060 * Update the preimage, and the common lines in postimage,
2061 * from buffer buf of length len. If postlen is 0 the postimage
2062 * is updated in place, otherwise it's updated on a new buffer
2063 * of length postlen
2064 */
2065
2066static void update_pre_post_images(struct image *preimage,
2067                                   struct image *postimage,
2068                                   char *buf,
2069                                   size_t len, size_t postlen)
2070{
2071        int i, ctx;
2072        char *new, *old, *fixed;
2073        struct image fixed_preimage;
2074
2075        /*
2076         * Update the preimage with whitespace fixes.  Note that we
2077         * are not losing preimage->buf -- apply_one_fragment() will
2078         * free "oldlines".
2079         */
2080        prepare_image(&fixed_preimage, buf, len, 1);
2081        assert(fixed_preimage.nr == preimage->nr);
2082        for (i = 0; i < preimage->nr; i++)
2083                fixed_preimage.line[i].flag = preimage->line[i].flag;
2084        free(preimage->line_allocated);
2085        *preimage = fixed_preimage;
2086
2087        /*
2088         * Adjust the common context lines in postimage. This can be
2089         * done in-place when we are just doing whitespace fixing,
2090         * which does not make the string grow, but needs a new buffer
2091         * when ignoring whitespace causes the update, since in this case
2092         * we could have e.g. tabs converted to multiple spaces.
2093         * We trust the caller to tell us if the update can be done
2094         * in place (postlen==0) or not.
2095         */
2096        old = postimage->buf;
2097        if (postlen)
2098                new = postimage->buf = xmalloc(postlen);
2099        else
2100                new = old;
2101        fixed = preimage->buf;
2102        for (i = ctx = 0; i < postimage->nr; i++) {
2103                size_t len = postimage->line[i].len;
2104                if (!(postimage->line[i].flag & LINE_COMMON)) {
2105                        /* an added line -- no counterparts in preimage */
2106                        memmove(new, old, len);
2107                        old += len;
2108                        new += len;
2109                        continue;
2110                }
2111
2112                /* a common context -- skip it in the original postimage */
2113                old += len;
2114
2115                /* and find the corresponding one in the fixed preimage */
2116                while (ctx < preimage->nr &&
2117                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2118                        fixed += preimage->line[ctx].len;
2119                        ctx++;
2120                }
2121                if (preimage->nr <= ctx)
2122                        die(_("oops"));
2123
2124                /* and copy it in, while fixing the line length */
2125                len = preimage->line[ctx].len;
2126                memcpy(new, fixed, len);
2127                new += len;
2128                fixed += len;
2129                postimage->line[i].len = len;
2130                ctx++;
2131        }
2132
2133        /* Fix the length of the whole thing */
2134        postimage->len = new - postimage->buf;
2135}
2136
2137static int match_fragment(struct image *img,
2138                          struct image *preimage,
2139                          struct image *postimage,
2140                          unsigned long try,
2141                          int try_lno,
2142                          unsigned ws_rule,
2143                          int match_beginning, int match_end)
2144{
2145        int i;
2146        char *fixed_buf, *buf, *orig, *target;
2147        struct strbuf fixed;
2148        size_t fixed_len;
2149        int preimage_limit;
2150
2151        if (preimage->nr + try_lno <= img->nr) {
2152                /*
2153                 * The hunk falls within the boundaries of img.
2154                 */
2155                preimage_limit = preimage->nr;
2156                if (match_end && (preimage->nr + try_lno != img->nr))
2157                        return 0;
2158        } else if (ws_error_action == correct_ws_error &&
2159                   (ws_rule & WS_BLANK_AT_EOF)) {
2160                /*
2161                 * This hunk extends beyond the end of img, and we are
2162                 * removing blank lines at the end of the file.  This
2163                 * many lines from the beginning of the preimage must
2164                 * match with img, and the remainder of the preimage
2165                 * must be blank.
2166                 */
2167                preimage_limit = img->nr - try_lno;
2168        } else {
2169                /*
2170                 * The hunk extends beyond the end of the img and
2171                 * we are not removing blanks at the end, so we
2172                 * should reject the hunk at this position.
2173                 */
2174                return 0;
2175        }
2176
2177        if (match_beginning && try_lno)
2178                return 0;
2179
2180        /* Quick hash check */
2181        for (i = 0; i < preimage_limit; i++)
2182                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2183                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2184                        return 0;
2185
2186        if (preimage_limit == preimage->nr) {
2187                /*
2188                 * Do we have an exact match?  If we were told to match
2189                 * at the end, size must be exactly at try+fragsize,
2190                 * otherwise try+fragsize must be still within the preimage,
2191                 * and either case, the old piece should match the preimage
2192                 * exactly.
2193                 */
2194                if ((match_end
2195                     ? (try + preimage->len == img->len)
2196                     : (try + preimage->len <= img->len)) &&
2197                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2198                        return 1;
2199        } else {
2200                /*
2201                 * The preimage extends beyond the end of img, so
2202                 * there cannot be an exact match.
2203                 *
2204                 * There must be one non-blank context line that match
2205                 * a line before the end of img.
2206                 */
2207                char *buf_end;
2208
2209                buf = preimage->buf;
2210                buf_end = buf;
2211                for (i = 0; i < preimage_limit; i++)
2212                        buf_end += preimage->line[i].len;
2213
2214                for ( ; buf < buf_end; buf++)
2215                        if (!isspace(*buf))
2216                                break;
2217                if (buf == buf_end)
2218                        return 0;
2219        }
2220
2221        /*
2222         * No exact match. If we are ignoring whitespace, run a line-by-line
2223         * fuzzy matching. We collect all the line length information because
2224         * we need it to adjust whitespace if we match.
2225         */
2226        if (ws_ignore_action == ignore_ws_change) {
2227                size_t imgoff = 0;
2228                size_t preoff = 0;
2229                size_t postlen = postimage->len;
2230                size_t extra_chars;
2231                char *preimage_eof;
2232                char *preimage_end;
2233                for (i = 0; i < preimage_limit; i++) {
2234                        size_t prelen = preimage->line[i].len;
2235                        size_t imglen = img->line[try_lno+i].len;
2236
2237                        if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2238                                              preimage->buf + preoff, prelen))
2239                                return 0;
2240                        if (preimage->line[i].flag & LINE_COMMON)
2241                                postlen += imglen - prelen;
2242                        imgoff += imglen;
2243                        preoff += prelen;
2244                }
2245
2246                /*
2247                 * Ok, the preimage matches with whitespace fuzz.
2248                 *
2249                 * imgoff now holds the true length of the target that
2250                 * matches the preimage before the end of the file.
2251                 *
2252                 * Count the number of characters in the preimage that fall
2253                 * beyond the end of the file and make sure that all of them
2254                 * are whitespace characters. (This can only happen if
2255                 * we are removing blank lines at the end of the file.)
2256                 */
2257                buf = preimage_eof = preimage->buf + preoff;
2258                for ( ; i < preimage->nr; i++)
2259                        preoff += preimage->line[i].len;
2260                preimage_end = preimage->buf + preoff;
2261                for ( ; buf < preimage_end; buf++)
2262                        if (!isspace(*buf))
2263                                return 0;
2264
2265                /*
2266                 * Update the preimage and the common postimage context
2267                 * lines to use the same whitespace as the target.
2268                 * If whitespace is missing in the target (i.e.
2269                 * if the preimage extends beyond the end of the file),
2270                 * use the whitespace from the preimage.
2271                 */
2272                extra_chars = preimage_end - preimage_eof;
2273                strbuf_init(&fixed, imgoff + extra_chars);
2274                strbuf_add(&fixed, img->buf + try, imgoff);
2275                strbuf_add(&fixed, preimage_eof, extra_chars);
2276                fixed_buf = strbuf_detach(&fixed, &fixed_len);
2277                update_pre_post_images(preimage, postimage,
2278                                fixed_buf, fixed_len, postlen);
2279                return 1;
2280        }
2281
2282        if (ws_error_action != correct_ws_error)
2283                return 0;
2284
2285        /*
2286         * The hunk does not apply byte-by-byte, but the hash says
2287         * it might with whitespace fuzz. We haven't been asked to
2288         * ignore whitespace, we were asked to correct whitespace
2289         * errors, so let's try matching after whitespace correction.
2290         *
2291         * The preimage may extend beyond the end of the file,
2292         * but in this loop we will only handle the part of the
2293         * preimage that falls within the file.
2294         */
2295        strbuf_init(&fixed, preimage->len + 1);
2296        orig = preimage->buf;
2297        target = img->buf + try;
2298        for (i = 0; i < preimage_limit; i++) {
2299                size_t oldlen = preimage->line[i].len;
2300                size_t tgtlen = img->line[try_lno + i].len;
2301                size_t fixstart = fixed.len;
2302                struct strbuf tgtfix;
2303                int match;
2304
2305                /* Try fixing the line in the preimage */
2306                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2307
2308                /* Try fixing the line in the target */
2309                strbuf_init(&tgtfix, tgtlen);
2310                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2311
2312                /*
2313                 * If they match, either the preimage was based on
2314                 * a version before our tree fixed whitespace breakage,
2315                 * or we are lacking a whitespace-fix patch the tree
2316                 * the preimage was based on already had (i.e. target
2317                 * has whitespace breakage, the preimage doesn't).
2318                 * In either case, we are fixing the whitespace breakages
2319                 * so we might as well take the fix together with their
2320                 * real change.
2321                 */
2322                match = (tgtfix.len == fixed.len - fixstart &&
2323                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2324                                             fixed.len - fixstart));
2325
2326                strbuf_release(&tgtfix);
2327                if (!match)
2328                        goto unmatch_exit;
2329
2330                orig += oldlen;
2331                target += tgtlen;
2332        }
2333
2334
2335        /*
2336         * Now handle the lines in the preimage that falls beyond the
2337         * end of the file (if any). They will only match if they are
2338         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2339         * false).
2340         */
2341        for ( ; i < preimage->nr; i++) {
2342                size_t fixstart = fixed.len; /* start of the fixed preimage */
2343                size_t oldlen = preimage->line[i].len;
2344                int j;
2345
2346                /* Try fixing the line in the preimage */
2347                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2348
2349                for (j = fixstart; j < fixed.len; j++)
2350                        if (!isspace(fixed.buf[j]))
2351                                goto unmatch_exit;
2352
2353                orig += oldlen;
2354        }
2355
2356        /*
2357         * Yes, the preimage is based on an older version that still
2358         * has whitespace breakages unfixed, and fixing them makes the
2359         * hunk match.  Update the context lines in the postimage.
2360         */
2361        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2362        update_pre_post_images(preimage, postimage,
2363                               fixed_buf, fixed_len, 0);
2364        return 1;
2365
2366 unmatch_exit:
2367        strbuf_release(&fixed);
2368        return 0;
2369}
2370
2371static int find_pos(struct image *img,
2372                    struct image *preimage,
2373                    struct image *postimage,
2374                    int line,
2375                    unsigned ws_rule,
2376                    int match_beginning, int match_end)
2377{
2378        int i;
2379        unsigned long backwards, forwards, try;
2380        int backwards_lno, forwards_lno, try_lno;
2381
2382        /*
2383         * If match_beginning or match_end is specified, there is no
2384         * point starting from a wrong line that will never match and
2385         * wander around and wait for a match at the specified end.
2386         */
2387        if (match_beginning)
2388                line = 0;
2389        else if (match_end)
2390                line = img->nr - preimage->nr;
2391
2392        /*
2393         * Because the comparison is unsigned, the following test
2394         * will also take care of a negative line number that can
2395         * result when match_end and preimage is larger than the target.
2396         */
2397        if ((size_t) line > img->nr)
2398                line = img->nr;
2399
2400        try = 0;
2401        for (i = 0; i < line; i++)
2402                try += img->line[i].len;
2403
2404        /*
2405         * There's probably some smart way to do this, but I'll leave
2406         * that to the smart and beautiful people. I'm simple and stupid.
2407         */
2408        backwards = try;
2409        backwards_lno = line;
2410        forwards = try;
2411        forwards_lno = line;
2412        try_lno = line;
2413
2414        for (i = 0; ; i++) {
2415                if (match_fragment(img, preimage, postimage,
2416                                   try, try_lno, ws_rule,
2417                                   match_beginning, match_end))
2418                        return try_lno;
2419
2420        again:
2421                if (backwards_lno == 0 && forwards_lno == img->nr)
2422                        break;
2423
2424                if (i & 1) {
2425                        if (backwards_lno == 0) {
2426                                i++;
2427                                goto again;
2428                        }
2429                        backwards_lno--;
2430                        backwards -= img->line[backwards_lno].len;
2431                        try = backwards;
2432                        try_lno = backwards_lno;
2433                } else {
2434                        if (forwards_lno == img->nr) {
2435                                i++;
2436                                goto again;
2437                        }
2438                        forwards += img->line[forwards_lno].len;
2439                        forwards_lno++;
2440                        try = forwards;
2441                        try_lno = forwards_lno;
2442                }
2443
2444        }
2445        return -1;
2446}
2447
2448static void remove_first_line(struct image *img)
2449{
2450        img->buf += img->line[0].len;
2451        img->len -= img->line[0].len;
2452        img->line++;
2453        img->nr--;
2454}
2455
2456static void remove_last_line(struct image *img)
2457{
2458        img->len -= img->line[--img->nr].len;
2459}
2460
2461/*
2462 * The change from "preimage" and "postimage" has been found to
2463 * apply at applied_pos (counts in line numbers) in "img".
2464 * Update "img" to remove "preimage" and replace it with "postimage".
2465 */
2466static void update_image(struct image *img,
2467                         int applied_pos,
2468                         struct image *preimage,
2469                         struct image *postimage)
2470{
2471        /*
2472         * remove the copy of preimage at offset in img
2473         * and replace it with postimage
2474         */
2475        int i, nr;
2476        size_t remove_count, insert_count, applied_at = 0;
2477        char *result;
2478        int preimage_limit;
2479
2480        /*
2481         * If we are removing blank lines at the end of img,
2482         * the preimage may extend beyond the end.
2483         * If that is the case, we must be careful only to
2484         * remove the part of the preimage that falls within
2485         * the boundaries of img. Initialize preimage_limit
2486         * to the number of lines in the preimage that falls
2487         * within the boundaries.
2488         */
2489        preimage_limit = preimage->nr;
2490        if (preimage_limit > img->nr - applied_pos)
2491                preimage_limit = img->nr - applied_pos;
2492
2493        for (i = 0; i < applied_pos; i++)
2494                applied_at += img->line[i].len;
2495
2496        remove_count = 0;
2497        for (i = 0; i < preimage_limit; i++)
2498                remove_count += img->line[applied_pos + i].len;
2499        insert_count = postimage->len;
2500
2501        /* Adjust the contents */
2502        result = xmalloc(img->len + insert_count - remove_count + 1);
2503        memcpy(result, img->buf, applied_at);
2504        memcpy(result + applied_at, postimage->buf, postimage->len);
2505        memcpy(result + applied_at + postimage->len,
2506               img->buf + (applied_at + remove_count),
2507               img->len - (applied_at + remove_count));
2508        free(img->buf);
2509        img->buf = result;
2510        img->len += insert_count - remove_count;
2511        result[img->len] = '\0';
2512
2513        /* Adjust the line table */
2514        nr = img->nr + postimage->nr - preimage_limit;
2515        if (preimage_limit < postimage->nr) {
2516                /*
2517                 * NOTE: this knows that we never call remove_first_line()
2518                 * on anything other than pre/post image.
2519                 */
2520                img->line = xrealloc(img->line, nr * sizeof(*img->line));
2521                img->line_allocated = img->line;
2522        }
2523        if (preimage_limit != postimage->nr)
2524                memmove(img->line + applied_pos + postimage->nr,
2525                        img->line + applied_pos + preimage_limit,
2526                        (img->nr - (applied_pos + preimage_limit)) *
2527                        sizeof(*img->line));
2528        memcpy(img->line + applied_pos,
2529               postimage->line,
2530               postimage->nr * sizeof(*img->line));
2531        if (!allow_overlap)
2532                for (i = 0; i < postimage->nr; i++)
2533                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2534        img->nr = nr;
2535}
2536
2537/*
2538 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2539 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2540 * replace the part of "img" with "postimage" text.
2541 */
2542static int apply_one_fragment(struct image *img, struct fragment *frag,
2543                              int inaccurate_eof, unsigned ws_rule,
2544                              int nth_fragment)
2545{
2546        int match_beginning, match_end;
2547        const char *patch = frag->patch;
2548        int size = frag->size;
2549        char *old, *oldlines;
2550        struct strbuf newlines;
2551        int new_blank_lines_at_end = 0;
2552        int found_new_blank_lines_at_end = 0;
2553        int hunk_linenr = frag->linenr;
2554        unsigned long leading, trailing;
2555        int pos, applied_pos;
2556        struct image preimage;
2557        struct image postimage;
2558
2559        memset(&preimage, 0, sizeof(preimage));
2560        memset(&postimage, 0, sizeof(postimage));
2561        oldlines = xmalloc(size);
2562        strbuf_init(&newlines, size);
2563
2564        old = oldlines;
2565        while (size > 0) {
2566                char first;
2567                int len = linelen(patch, size);
2568                int plen;
2569                int added_blank_line = 0;
2570                int is_blank_context = 0;
2571                size_t start;
2572
2573                if (!len)
2574                        break;
2575
2576                /*
2577                 * "plen" is how much of the line we should use for
2578                 * the actual patch data. Normally we just remove the
2579                 * first character on the line, but if the line is
2580                 * followed by "\ No newline", then we also remove the
2581                 * last one (which is the newline, of course).
2582                 */
2583                plen = len - 1;
2584                if (len < size && patch[len] == '\\')
2585                        plen--;
2586                first = *patch;
2587                if (apply_in_reverse) {
2588                        if (first == '-')
2589                                first = '+';
2590                        else if (first == '+')
2591                                first = '-';
2592                }
2593
2594                switch (first) {
2595                case '\n':
2596                        /* Newer GNU diff, empty context line */
2597                        if (plen < 0)
2598                                /* ... followed by '\No newline'; nothing */
2599                                break;
2600                        *old++ = '\n';
2601                        strbuf_addch(&newlines, '\n');
2602                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2603                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2604                        is_blank_context = 1;
2605                        break;
2606                case ' ':
2607                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2608                            ws_blank_line(patch + 1, plen, ws_rule))
2609                                is_blank_context = 1;
2610                case '-':
2611                        memcpy(old, patch + 1, plen);
2612                        add_line_info(&preimage, old, plen,
2613                                      (first == ' ' ? LINE_COMMON : 0));
2614                        old += plen;
2615                        if (first == '-')
2616                                break;
2617                /* Fall-through for ' ' */
2618                case '+':
2619                        /* --no-add does not add new lines */
2620                        if (first == '+' && no_add)
2621                                break;
2622
2623                        start = newlines.len;
2624                        if (first != '+' ||
2625                            !whitespace_error ||
2626                            ws_error_action != correct_ws_error) {
2627                                strbuf_add(&newlines, patch + 1, plen);
2628                        }
2629                        else {
2630                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2631                        }
2632                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2633                                      (first == '+' ? 0 : LINE_COMMON));
2634                        if (first == '+' &&
2635                            (ws_rule & WS_BLANK_AT_EOF) &&
2636                            ws_blank_line(patch + 1, plen, ws_rule))
2637                                added_blank_line = 1;
2638                        break;
2639                case '@': case '\\':
2640                        /* Ignore it, we already handled it */
2641                        break;
2642                default:
2643                        if (apply_verbosely)
2644                                error(_("invalid start of line: '%c'"), first);
2645                        return -1;
2646                }
2647                if (added_blank_line) {
2648                        if (!new_blank_lines_at_end)
2649                                found_new_blank_lines_at_end = hunk_linenr;
2650                        new_blank_lines_at_end++;
2651                }
2652                else if (is_blank_context)
2653                        ;
2654                else
2655                        new_blank_lines_at_end = 0;
2656                patch += len;
2657                size -= len;
2658                hunk_linenr++;
2659        }
2660        if (inaccurate_eof &&
2661            old > oldlines && old[-1] == '\n' &&
2662            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2663                old--;
2664                strbuf_setlen(&newlines, newlines.len - 1);
2665        }
2666
2667        leading = frag->leading;
2668        trailing = frag->trailing;
2669
2670        /*
2671         * A hunk to change lines at the beginning would begin with
2672         * @@ -1,L +N,M @@
2673         * but we need to be careful.  -U0 that inserts before the second
2674         * line also has this pattern.
2675         *
2676         * And a hunk to add to an empty file would begin with
2677         * @@ -0,0 +N,M @@
2678         *
2679         * In other words, a hunk that is (frag->oldpos <= 1) with or
2680         * without leading context must match at the beginning.
2681         */
2682        match_beginning = (!frag->oldpos ||
2683                           (frag->oldpos == 1 && !unidiff_zero));
2684
2685        /*
2686         * A hunk without trailing lines must match at the end.
2687         * However, we simply cannot tell if a hunk must match end
2688         * from the lack of trailing lines if the patch was generated
2689         * with unidiff without any context.
2690         */
2691        match_end = !unidiff_zero && !trailing;
2692
2693        pos = frag->newpos ? (frag->newpos - 1) : 0;
2694        preimage.buf = oldlines;
2695        preimage.len = old - oldlines;
2696        postimage.buf = newlines.buf;
2697        postimage.len = newlines.len;
2698        preimage.line = preimage.line_allocated;
2699        postimage.line = postimage.line_allocated;
2700
2701        for (;;) {
2702
2703                applied_pos = find_pos(img, &preimage, &postimage, pos,
2704                                       ws_rule, match_beginning, match_end);
2705
2706                if (applied_pos >= 0)
2707                        break;
2708
2709                /* Am I at my context limits? */
2710                if ((leading <= p_context) && (trailing <= p_context))
2711                        break;
2712                if (match_beginning || match_end) {
2713                        match_beginning = match_end = 0;
2714                        continue;
2715                }
2716
2717                /*
2718                 * Reduce the number of context lines; reduce both
2719                 * leading and trailing if they are equal otherwise
2720                 * just reduce the larger context.
2721                 */
2722                if (leading >= trailing) {
2723                        remove_first_line(&preimage);
2724                        remove_first_line(&postimage);
2725                        pos--;
2726                        leading--;
2727                }
2728                if (trailing > leading) {
2729                        remove_last_line(&preimage);
2730                        remove_last_line(&postimage);
2731                        trailing--;
2732                }
2733        }
2734
2735        if (applied_pos >= 0) {
2736                if (new_blank_lines_at_end &&
2737                    preimage.nr + applied_pos >= img->nr &&
2738                    (ws_rule & WS_BLANK_AT_EOF) &&
2739                    ws_error_action != nowarn_ws_error) {
2740                        record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2741                                        found_new_blank_lines_at_end);
2742                        if (ws_error_action == correct_ws_error) {
2743                                while (new_blank_lines_at_end--)
2744                                        remove_last_line(&postimage);
2745                        }
2746                        /*
2747                         * We would want to prevent write_out_results()
2748                         * from taking place in apply_patch() that follows
2749                         * the callchain led us here, which is:
2750                         * apply_patch->check_patch_list->check_patch->
2751                         * apply_data->apply_fragments->apply_one_fragment
2752                         */
2753                        if (ws_error_action == die_on_ws_error)
2754                                apply = 0;
2755                }
2756
2757                if (apply_verbosely && applied_pos != pos) {
2758                        int offset = applied_pos - pos;
2759                        if (apply_in_reverse)
2760                                offset = 0 - offset;
2761                        fprintf_ln(stderr,
2762                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
2763                                      "Hunk #%d succeeded at %d (offset %d lines).",
2764                                      offset),
2765                                   nth_fragment, applied_pos + 1, offset);
2766                }
2767
2768                /*
2769                 * Warn if it was necessary to reduce the number
2770                 * of context lines.
2771                 */
2772                if ((leading != frag->leading) ||
2773                    (trailing != frag->trailing))
2774                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2775                                             " to apply fragment at %d"),
2776                                   leading, trailing, applied_pos+1);
2777                update_image(img, applied_pos, &preimage, &postimage);
2778        } else {
2779                if (apply_verbosely)
2780                        error(_("while searching for:\n%.*s"),
2781                              (int)(old - oldlines), oldlines);
2782        }
2783
2784        free(oldlines);
2785        strbuf_release(&newlines);
2786        free(preimage.line_allocated);
2787        free(postimage.line_allocated);
2788
2789        return (applied_pos < 0);
2790}
2791
2792static int apply_binary_fragment(struct image *img, struct patch *patch)
2793{
2794        struct fragment *fragment = patch->fragments;
2795        unsigned long len;
2796        void *dst;
2797
2798        if (!fragment)
2799                return error(_("missing binary patch data for '%s'"),
2800                             patch->new_name ?
2801                             patch->new_name :
2802                             patch->old_name);
2803
2804        /* Binary patch is irreversible without the optional second hunk */
2805        if (apply_in_reverse) {
2806                if (!fragment->next)
2807                        return error("cannot reverse-apply a binary patch "
2808                                     "without the reverse hunk to '%s'",
2809                                     patch->new_name
2810                                     ? patch->new_name : patch->old_name);
2811                fragment = fragment->next;
2812        }
2813        switch (fragment->binary_patch_method) {
2814        case BINARY_DELTA_DEFLATED:
2815                dst = patch_delta(img->buf, img->len, fragment->patch,
2816                                  fragment->size, &len);
2817                if (!dst)
2818                        return -1;
2819                clear_image(img);
2820                img->buf = dst;
2821                img->len = len;
2822                return 0;
2823        case BINARY_LITERAL_DEFLATED:
2824                clear_image(img);
2825                img->len = fragment->size;
2826                img->buf = xmalloc(img->len+1);
2827                memcpy(img->buf, fragment->patch, img->len);
2828                img->buf[img->len] = '\0';
2829                return 0;
2830        }
2831        return -1;
2832}
2833
2834/*
2835 * Replace "img" with the result of applying the binary patch.
2836 * The binary patch data itself in patch->fragment is still kept
2837 * but the preimage prepared by the caller in "img" is freed here
2838 * or in the helper function apply_binary_fragment() this calls.
2839 */
2840static int apply_binary(struct image *img, struct patch *patch)
2841{
2842        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2843        unsigned char sha1[20];
2844
2845        /*
2846         * For safety, we require patch index line to contain
2847         * full 40-byte textual SHA1 for old and new, at least for now.
2848         */
2849        if (strlen(patch->old_sha1_prefix) != 40 ||
2850            strlen(patch->new_sha1_prefix) != 40 ||
2851            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2852            get_sha1_hex(patch->new_sha1_prefix, sha1))
2853                return error("cannot apply binary patch to '%s' "
2854                             "without full index line", name);
2855
2856        if (patch->old_name) {
2857                /*
2858                 * See if the old one matches what the patch
2859                 * applies to.
2860                 */
2861                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2862                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2863                        return error("the patch applies to '%s' (%s), "
2864                                     "which does not match the "
2865                                     "current contents.",
2866                                     name, sha1_to_hex(sha1));
2867        }
2868        else {
2869                /* Otherwise, the old one must be empty. */
2870                if (img->len)
2871                        return error("the patch applies to an empty "
2872                                     "'%s' but it is not empty", name);
2873        }
2874
2875        get_sha1_hex(patch->new_sha1_prefix, sha1);
2876        if (is_null_sha1(sha1)) {
2877                clear_image(img);
2878                return 0; /* deletion patch */
2879        }
2880
2881        if (has_sha1_file(sha1)) {
2882                /* We already have the postimage */
2883                enum object_type type;
2884                unsigned long size;
2885                char *result;
2886
2887                result = read_sha1_file(sha1, &type, &size);
2888                if (!result)
2889                        return error("the necessary postimage %s for "
2890                                     "'%s' cannot be read",
2891                                     patch->new_sha1_prefix, name);
2892                clear_image(img);
2893                img->buf = result;
2894                img->len = size;
2895        } else {
2896                /*
2897                 * We have verified buf matches the preimage;
2898                 * apply the patch data to it, which is stored
2899                 * in the patch->fragments->{patch,size}.
2900                 */
2901                if (apply_binary_fragment(img, patch))
2902                        return error(_("binary patch does not apply to '%s'"),
2903                                     name);
2904
2905                /* verify that the result matches */
2906                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2907                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2908                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2909                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2910        }
2911
2912        return 0;
2913}
2914
2915static int apply_fragments(struct image *img, struct patch *patch)
2916{
2917        struct fragment *frag = patch->fragments;
2918        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2919        unsigned ws_rule = patch->ws_rule;
2920        unsigned inaccurate_eof = patch->inaccurate_eof;
2921        int nth = 0;
2922
2923        if (patch->is_binary)
2924                return apply_binary(img, patch);
2925
2926        while (frag) {
2927                nth++;
2928                if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2929                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
2930                        if (!apply_with_reject)
2931                                return -1;
2932                        frag->rejected = 1;
2933                }
2934                frag = frag->next;
2935        }
2936        return 0;
2937}
2938
2939static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2940{
2941        if (!ce)
2942                return 0;
2943
2944        if (S_ISGITLINK(ce->ce_mode)) {
2945                strbuf_grow(buf, 100);
2946                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2947        } else {
2948                enum object_type type;
2949                unsigned long sz;
2950                char *result;
2951
2952                result = read_sha1_file(ce->sha1, &type, &sz);
2953                if (!result)
2954                        return -1;
2955                /* XXX read_sha1_file NUL-terminates */
2956                strbuf_attach(buf, result, sz, sz + 1);
2957        }
2958        return 0;
2959}
2960
2961static struct patch *in_fn_table(const char *name)
2962{
2963        struct string_list_item *item;
2964
2965        if (name == NULL)
2966                return NULL;
2967
2968        item = string_list_lookup(&fn_table, name);
2969        if (item != NULL)
2970                return (struct patch *)item->util;
2971
2972        return NULL;
2973}
2974
2975/*
2976 * item->util in the filename table records the status of the path.
2977 * Usually it points at a patch (whose result records the contents
2978 * of it after applying it), but it could be PATH_WAS_DELETED for a
2979 * path that a previously applied patch has already removed.
2980 */
2981 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2982#define PATH_WAS_DELETED ((struct patch *) -1)
2983
2984static int to_be_deleted(struct patch *patch)
2985{
2986        return patch == PATH_TO_BE_DELETED;
2987}
2988
2989static int was_deleted(struct patch *patch)
2990{
2991        return patch == PATH_WAS_DELETED;
2992}
2993
2994static void add_to_fn_table(struct patch *patch)
2995{
2996        struct string_list_item *item;
2997
2998        /*
2999         * Always add new_name unless patch is a deletion
3000         * This should cover the cases for normal diffs,
3001         * file creations and copies
3002         */
3003        if (patch->new_name != NULL) {
3004                item = string_list_insert(&fn_table, patch->new_name);
3005                item->util = patch;
3006        }
3007
3008        /*
3009         * store a failure on rename/deletion cases because
3010         * later chunks shouldn't patch old names
3011         */
3012        if ((patch->new_name == NULL) || (patch->is_rename)) {
3013                item = string_list_insert(&fn_table, patch->old_name);
3014                item->util = PATH_WAS_DELETED;
3015        }
3016}
3017
3018static void prepare_fn_table(struct patch *patch)
3019{
3020        /*
3021         * store information about incoming file deletion
3022         */
3023        while (patch) {
3024                if ((patch->new_name == NULL) || (patch->is_rename)) {
3025                        struct string_list_item *item;
3026                        item = string_list_insert(&fn_table, patch->old_name);
3027                        item->util = PATH_TO_BE_DELETED;
3028                }
3029                patch = patch->next;
3030        }
3031}
3032
3033static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
3034{
3035        struct strbuf buf = STRBUF_INIT;
3036        struct image image;
3037        size_t len;
3038        char *img;
3039        struct patch *tpatch;
3040
3041        if (!(patch->is_copy || patch->is_rename) &&
3042            (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
3043                if (was_deleted(tpatch)) {
3044                        return error(_("patch %s has been renamed/deleted"),
3045                                patch->old_name);
3046                }
3047                /* We have a patched copy in memory; use that. */
3048                strbuf_add(&buf, tpatch->result, tpatch->resultsize);
3049        } else if (cached) {
3050                if (read_file_or_gitlink(ce, &buf))
3051                        return error(_("read of %s failed"), patch->old_name);
3052        } else if (patch->old_name) {
3053                if (S_ISGITLINK(patch->old_mode)) {
3054                        if (ce) {
3055                                read_file_or_gitlink(ce, &buf);
3056                        } else {
3057                                /*
3058                                 * There is no way to apply subproject
3059                                 * patch without looking at the index.
3060                                 * NEEDSWORK: shouldn't this be flagged
3061                                 * as an error???
3062                                 */
3063                                free_fragment_list(patch->fragments);
3064                                patch->fragments = NULL;
3065                        }
3066                } else {
3067                        if (read_old_data(st, patch->old_name, &buf))
3068                                return error(_("read of %s failed"), patch->old_name);
3069                }
3070        }
3071
3072        img = strbuf_detach(&buf, &len);
3073        prepare_image(&image, img, len, !patch->is_binary);
3074
3075        if (apply_fragments(&image, patch) < 0)
3076                return -1; /* note with --reject this succeeds. */
3077        patch->result = image.buf;
3078        patch->resultsize = image.len;
3079        add_to_fn_table(patch);
3080        free(image.line_allocated);
3081
3082        if (0 < patch->is_delete && patch->resultsize)
3083                return error(_("removal patch leaves file contents"));
3084
3085        return 0;
3086}
3087
3088static int check_to_create_blob(const char *new_name, int ok_if_exists)
3089{
3090        struct stat nst;
3091        if (!lstat(new_name, &nst)) {
3092                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3093                        return 0;
3094                /*
3095                 * A leading component of new_name might be a symlink
3096                 * that is going to be removed with this patch, but
3097                 * still pointing at somewhere that has the path.
3098                 * In such a case, path "new_name" does not exist as
3099                 * far as git is concerned.
3100                 */
3101                if (has_symlink_leading_path(new_name, strlen(new_name)))
3102                        return 0;
3103
3104                return error(_("%s: already exists in working directory"), new_name);
3105        }
3106        else if ((errno != ENOENT) && (errno != ENOTDIR))
3107                return error("%s: %s", new_name, strerror(errno));
3108        return 0;
3109}
3110
3111static int verify_index_match(struct cache_entry *ce, struct stat *st)
3112{
3113        if (S_ISGITLINK(ce->ce_mode)) {
3114                if (!S_ISDIR(st->st_mode))
3115                        return -1;
3116                return 0;
3117        }
3118        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3119}
3120
3121static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3122{
3123        const char *old_name = patch->old_name;
3124        struct patch *tpatch = NULL;
3125        int stat_ret = 0;
3126        unsigned st_mode = 0;
3127
3128        /*
3129         * Make sure that we do not have local modifications from the
3130         * index when we are looking at the index.  Also make sure
3131         * we have the preimage file to be patched in the work tree,
3132         * unless --cached, which tells git to apply only in the index.
3133         */
3134        if (!old_name)
3135                return 0;
3136
3137        assert(patch->is_new <= 0);
3138
3139        if (!(patch->is_copy || patch->is_rename) &&
3140            (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
3141                if (was_deleted(tpatch))
3142                        return error(_("%s: has been deleted/renamed"), old_name);
3143                st_mode = tpatch->new_mode;
3144        } else if (!cached) {
3145                stat_ret = lstat(old_name, st);
3146                if (stat_ret && errno != ENOENT)
3147                        return error(_("%s: %s"), old_name, strerror(errno));
3148        }
3149
3150        if (to_be_deleted(tpatch))
3151                tpatch = NULL;
3152
3153        if (check_index && !tpatch) {
3154                int pos = cache_name_pos(old_name, strlen(old_name));
3155                if (pos < 0) {
3156                        if (patch->is_new < 0)
3157                                goto is_new;
3158                        return error(_("%s: does not exist in index"), old_name);
3159                }
3160                *ce = active_cache[pos];
3161                if (stat_ret < 0) {
3162                        struct checkout costate;
3163                        /* checkout */
3164                        memset(&costate, 0, sizeof(costate));
3165                        costate.base_dir = "";
3166                        costate.refresh_cache = 1;
3167                        if (checkout_entry(*ce, &costate, NULL) ||
3168                            lstat(old_name, st))
3169                                return -1;
3170                }
3171                if (!cached && verify_index_match(*ce, st))
3172                        return error(_("%s: does not match index"), old_name);
3173                if (cached)
3174                        st_mode = (*ce)->ce_mode;
3175        } else if (stat_ret < 0) {
3176                if (patch->is_new < 0)
3177                        goto is_new;
3178                return error(_("%s: %s"), old_name, strerror(errno));
3179        }
3180
3181        if (!cached && !tpatch)
3182                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3183
3184        if (patch->is_new < 0)
3185                patch->is_new = 0;
3186        if (!patch->old_mode)
3187                patch->old_mode = st_mode;
3188        if ((st_mode ^ patch->old_mode) & S_IFMT)
3189                return error(_("%s: wrong type"), old_name);
3190        if (st_mode != patch->old_mode)
3191                warning(_("%s has type %o, expected %o"),
3192                        old_name, st_mode, patch->old_mode);
3193        if (!patch->new_mode && !patch->is_delete)
3194                patch->new_mode = st_mode;
3195        return 0;
3196
3197 is_new:
3198        patch->is_new = 1;
3199        patch->is_delete = 0;
3200        free(patch->old_name);
3201        patch->old_name = NULL;
3202        return 0;
3203}
3204
3205/*
3206 * Check and apply the patch in-core; leave the result in patch->result
3207 * for the caller to write it out to the final destination.
3208 */
3209static int check_patch(struct patch *patch)
3210{
3211        struct stat st;
3212        const char *old_name = patch->old_name;
3213        const char *new_name = patch->new_name;
3214        const char *name = old_name ? old_name : new_name;
3215        struct cache_entry *ce = NULL;
3216        struct patch *tpatch;
3217        int ok_if_exists;
3218        int status;
3219
3220        patch->rejected = 1; /* we will drop this after we succeed */
3221
3222        status = check_preimage(patch, &ce, &st);
3223        if (status)
3224                return status;
3225        old_name = patch->old_name;
3226
3227        if ((tpatch = in_fn_table(new_name)) &&
3228                        (was_deleted(tpatch) || to_be_deleted(tpatch)))
3229                /*
3230                 * A type-change diff is always split into a patch to
3231                 * delete old, immediately followed by a patch to
3232                 * create new (see diff.c::run_diff()); in such a case
3233                 * it is Ok that the entry to be deleted by the
3234                 * previous patch is still in the working tree and in
3235                 * the index.
3236                 */
3237                ok_if_exists = 1;
3238        else
3239                ok_if_exists = 0;
3240
3241        if (new_name &&
3242            ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3243                if (check_index &&
3244                    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3245                    !ok_if_exists)
3246                        return error(_("%s: already exists in index"), new_name);
3247                if (!cached) {
3248                        int err = check_to_create_blob(new_name, ok_if_exists);
3249                        if (err)
3250                                return err;
3251                }
3252                if (!patch->new_mode) {
3253                        if (0 < patch->is_new)
3254                                patch->new_mode = S_IFREG | 0644;
3255                        else
3256                                patch->new_mode = patch->old_mode;
3257                }
3258        }
3259
3260        if (new_name && old_name) {
3261                int same = !strcmp(old_name, new_name);
3262                if (!patch->new_mode)
3263                        patch->new_mode = patch->old_mode;
3264                if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3265                        if (same)
3266                                return error(_("new mode (%o) of %s does not "
3267                                               "match old mode (%o)"),
3268                                        patch->new_mode, new_name,
3269                                        patch->old_mode);
3270                        else
3271                                return error(_("new mode (%o) of %s does not "
3272                                               "match old mode (%o) of %s"),
3273                                        patch->new_mode, new_name,
3274                                        patch->old_mode, old_name);
3275                }
3276        }
3277
3278        if (apply_data(patch, &st, ce) < 0)
3279                return error(_("%s: patch does not apply"), name);
3280        patch->rejected = 0;
3281        return 0;
3282}
3283
3284static int check_patch_list(struct patch *patch)
3285{
3286        int err = 0;
3287
3288        prepare_fn_table(patch);
3289        while (patch) {
3290                if (apply_verbosely)
3291                        say_patch_name(stderr,
3292                                       _("Checking patch %s..."), patch);
3293                err |= check_patch(patch);
3294                patch = patch->next;
3295        }
3296        return err;
3297}
3298
3299/* This function tries to read the sha1 from the current index */
3300static int get_current_sha1(const char *path, unsigned char *sha1)
3301{
3302        int pos;
3303
3304        if (read_cache() < 0)
3305                return -1;
3306        pos = cache_name_pos(path, strlen(path));
3307        if (pos < 0)
3308                return -1;
3309        hashcpy(sha1, active_cache[pos]->sha1);
3310        return 0;
3311}
3312
3313/* Build an index that contains the just the files needed for a 3way merge */
3314static void build_fake_ancestor(struct patch *list, const char *filename)
3315{
3316        struct patch *patch;
3317        struct index_state result = { NULL };
3318        int fd;
3319
3320        /* Once we start supporting the reverse patch, it may be
3321         * worth showing the new sha1 prefix, but until then...
3322         */
3323        for (patch = list; patch; patch = patch->next) {
3324                const unsigned char *sha1_ptr;
3325                unsigned char sha1[20];
3326                struct cache_entry *ce;
3327                const char *name;
3328
3329                name = patch->old_name ? patch->old_name : patch->new_name;
3330                if (0 < patch->is_new)
3331                        continue;
3332                else if (get_sha1(patch->old_sha1_prefix, sha1))
3333                        /* git diff has no index line for mode/type changes */
3334                        if (!patch->lines_added && !patch->lines_deleted) {
3335                                if (get_current_sha1(patch->old_name, sha1))
3336                                        die("mode change for %s, which is not "
3337                                                "in current HEAD", name);
3338                                sha1_ptr = sha1;
3339                        } else
3340                                die("sha1 information is lacking or useless "
3341                                        "(%s).", name);
3342                else
3343                        sha1_ptr = sha1;
3344
3345                ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3346                if (!ce)
3347                        die(_("make_cache_entry failed for path '%s'"), name);
3348                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3349                        die ("Could not add %s to temporary index", name);
3350        }
3351
3352        fd = open(filename, O_WRONLY | O_CREAT, 0666);
3353        if (fd < 0 || write_index(&result, fd) || close(fd))
3354                die ("Could not write temporary index to %s", filename);
3355
3356        discard_index(&result);
3357}
3358
3359static void stat_patch_list(struct patch *patch)
3360{
3361        int files, adds, dels;
3362
3363        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3364                files++;
3365                adds += patch->lines_added;
3366                dels += patch->lines_deleted;
3367                show_stats(patch);
3368        }
3369
3370        print_stat_summary(stdout, files, adds, dels);
3371}
3372
3373static void numstat_patch_list(struct patch *patch)
3374{
3375        for ( ; patch; patch = patch->next) {
3376                const char *name;
3377                name = patch->new_name ? patch->new_name : patch->old_name;
3378                if (patch->is_binary)
3379                        printf("-\t-\t");
3380                else
3381                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3382                write_name_quoted(name, stdout, line_termination);
3383        }
3384}
3385
3386static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3387{
3388        if (mode)
3389                printf(" %s mode %06o %s\n", newdelete, mode, name);
3390        else
3391                printf(" %s %s\n", newdelete, name);
3392}
3393
3394static void show_mode_change(struct patch *p, int show_name)
3395{
3396        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3397                if (show_name)
3398                        printf(" mode change %06o => %06o %s\n",
3399                               p->old_mode, p->new_mode, p->new_name);
3400                else
3401                        printf(" mode change %06o => %06o\n",
3402                               p->old_mode, p->new_mode);
3403        }
3404}
3405
3406static void show_rename_copy(struct patch *p)
3407{
3408        const char *renamecopy = p->is_rename ? "rename" : "copy";
3409        const char *old, *new;
3410
3411        /* Find common prefix */
3412        old = p->old_name;
3413        new = p->new_name;
3414        while (1) {
3415                const char *slash_old, *slash_new;
3416                slash_old = strchr(old, '/');
3417                slash_new = strchr(new, '/');
3418                if (!slash_old ||
3419                    !slash_new ||
3420                    slash_old - old != slash_new - new ||
3421                    memcmp(old, new, slash_new - new))
3422                        break;
3423                old = slash_old + 1;
3424                new = slash_new + 1;
3425        }
3426        /* p->old_name thru old is the common prefix, and old and new
3427         * through the end of names are renames
3428         */
3429        if (old != p->old_name)
3430                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3431                       (int)(old - p->old_name), p->old_name,
3432                       old, new, p->score);
3433        else
3434                printf(" %s %s => %s (%d%%)\n", renamecopy,
3435                       p->old_name, p->new_name, p->score);
3436        show_mode_change(p, 0);
3437}
3438
3439static void summary_patch_list(struct patch *patch)
3440{
3441        struct patch *p;
3442
3443        for (p = patch; p; p = p->next) {
3444                if (p->is_new)
3445                        show_file_mode_name("create", p->new_mode, p->new_name);
3446                else if (p->is_delete)
3447                        show_file_mode_name("delete", p->old_mode, p->old_name);
3448                else {
3449                        if (p->is_rename || p->is_copy)
3450                                show_rename_copy(p);
3451                        else {
3452                                if (p->score) {
3453                                        printf(" rewrite %s (%d%%)\n",
3454                                               p->new_name, p->score);
3455                                        show_mode_change(p, 0);
3456                                }
3457                                else
3458                                        show_mode_change(p, 1);
3459                        }
3460                }
3461        }
3462}
3463
3464static void patch_stats(struct patch *patch)
3465{
3466        int lines = patch->lines_added + patch->lines_deleted;
3467
3468        if (lines > max_change)
3469                max_change = lines;
3470        if (patch->old_name) {
3471                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3472                if (!len)
3473                        len = strlen(patch->old_name);
3474                if (len > max_len)
3475                        max_len = len;
3476        }
3477        if (patch->new_name) {
3478                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3479                if (!len)
3480                        len = strlen(patch->new_name);
3481                if (len > max_len)
3482                        max_len = len;
3483        }
3484}
3485
3486static void remove_file(struct patch *patch, int rmdir_empty)
3487{
3488        if (update_index) {
3489                if (remove_file_from_cache(patch->old_name) < 0)
3490                        die(_("unable to remove %s from index"), patch->old_name);
3491        }
3492        if (!cached) {
3493                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3494                        remove_path(patch->old_name);
3495                }
3496        }
3497}
3498
3499static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3500{
3501        struct stat st;
3502        struct cache_entry *ce;
3503        int namelen = strlen(path);
3504        unsigned ce_size = cache_entry_size(namelen);
3505
3506        if (!update_index)
3507                return;
3508
3509        ce = xcalloc(1, ce_size);
3510        memcpy(ce->name, path, namelen);
3511        ce->ce_mode = create_ce_mode(mode);
3512        ce->ce_flags = namelen;
3513        if (S_ISGITLINK(mode)) {
3514                const char *s = buf;
3515
3516                if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3517                        die(_("corrupt patch for subproject %s"), path);
3518        } else {
3519                if (!cached) {
3520                        if (lstat(path, &st) < 0)
3521                                die_errno(_("unable to stat newly created file '%s'"),
3522                                          path);
3523                        fill_stat_cache_info(ce, &st);
3524                }
3525                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3526                        die(_("unable to create backing store for newly created file %s"), path);
3527        }
3528        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3529                die(_("unable to add cache entry for %s"), path);
3530}
3531
3532static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3533{
3534        int fd;
3535        struct strbuf nbuf = STRBUF_INIT;
3536
3537        if (S_ISGITLINK(mode)) {
3538                struct stat st;
3539                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3540                        return 0;
3541                return mkdir(path, 0777);
3542        }
3543
3544        if (has_symlinks && S_ISLNK(mode))
3545                /* Although buf:size is counted string, it also is NUL
3546                 * terminated.
3547                 */
3548                return symlink(buf, path);
3549
3550        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3551        if (fd < 0)
3552                return -1;
3553
3554        if (convert_to_working_tree(path, buf, size, &nbuf)) {
3555                size = nbuf.len;
3556                buf  = nbuf.buf;
3557        }
3558        write_or_die(fd, buf, size);
3559        strbuf_release(&nbuf);
3560
3561        if (close(fd) < 0)
3562                die_errno(_("closing file '%s'"), path);
3563        return 0;
3564}
3565
3566/*
3567 * We optimistically assume that the directories exist,
3568 * which is true 99% of the time anyway. If they don't,
3569 * we create them and try again.
3570 */
3571static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3572{
3573        if (cached)
3574                return;
3575        if (!try_create_file(path, mode, buf, size))
3576                return;
3577
3578        if (errno == ENOENT) {
3579                if (safe_create_leading_directories(path))
3580                        return;
3581                if (!try_create_file(path, mode, buf, size))
3582                        return;
3583        }
3584
3585        if (errno == EEXIST || errno == EACCES) {
3586                /* We may be trying to create a file where a directory
3587                 * used to be.
3588                 */
3589                struct stat st;
3590                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3591                        errno = EEXIST;
3592        }
3593
3594        if (errno == EEXIST) {
3595                unsigned int nr = getpid();
3596
3597                for (;;) {
3598                        char newpath[PATH_MAX];
3599                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3600                        if (!try_create_file(newpath, mode, buf, size)) {
3601                                if (!rename(newpath, path))
3602                                        return;
3603                                unlink_or_warn(newpath);
3604                                break;
3605                        }
3606                        if (errno != EEXIST)
3607                                break;
3608                        ++nr;
3609                }
3610        }
3611        die_errno(_("unable to write file '%s' mode %o"), path, mode);
3612}
3613
3614static void create_file(struct patch *patch)
3615{
3616        char *path = patch->new_name;
3617        unsigned mode = patch->new_mode;
3618        unsigned long size = patch->resultsize;
3619        char *buf = patch->result;
3620
3621        if (!mode)
3622                mode = S_IFREG | 0644;
3623        create_one_file(path, mode, buf, size);
3624        add_index_file(path, mode, buf, size);
3625}
3626
3627/* phase zero is to remove, phase one is to create */
3628static void write_out_one_result(struct patch *patch, int phase)
3629{
3630        if (patch->is_delete > 0) {
3631                if (phase == 0)
3632                        remove_file(patch, 1);
3633                return;
3634        }
3635        if (patch->is_new > 0 || patch->is_copy) {
3636                if (phase == 1)
3637                        create_file(patch);
3638                return;
3639        }
3640        /*
3641         * Rename or modification boils down to the same
3642         * thing: remove the old, write the new
3643         */
3644        if (phase == 0)
3645                remove_file(patch, patch->is_rename);
3646        if (phase == 1)
3647                create_file(patch);
3648}
3649
3650static int write_out_one_reject(struct patch *patch)
3651{
3652        FILE *rej;
3653        char namebuf[PATH_MAX];
3654        struct fragment *frag;
3655        int cnt = 0;
3656        struct strbuf sb = STRBUF_INIT;
3657
3658        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3659                if (!frag->rejected)
3660                        continue;
3661                cnt++;
3662        }
3663
3664        if (!cnt) {
3665                if (apply_verbosely)
3666                        say_patch_name(stderr,
3667                                       _("Applied patch %s cleanly."), patch);
3668                return 0;
3669        }
3670
3671        /* This should not happen, because a removal patch that leaves
3672         * contents are marked "rejected" at the patch level.
3673         */
3674        if (!patch->new_name)
3675                die(_("internal error"));
3676
3677        /* Say this even without --verbose */
3678        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
3679                            "Applying patch %%s with %d rejects...",
3680                            cnt),
3681                    cnt);
3682        say_patch_name(stderr, sb.buf, patch);
3683        strbuf_release(&sb);
3684
3685        cnt = strlen(patch->new_name);
3686        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3687                cnt = ARRAY_SIZE(namebuf) - 5;
3688                warning(_("truncating .rej filename to %.*s.rej"),
3689                        cnt - 1, patch->new_name);
3690        }
3691        memcpy(namebuf, patch->new_name, cnt);
3692        memcpy(namebuf + cnt, ".rej", 5);
3693
3694        rej = fopen(namebuf, "w");
3695        if (!rej)
3696                return error(_("cannot open %s: %s"), namebuf, strerror(errno));
3697
3698        /* Normal git tools never deal with .rej, so do not pretend
3699         * this is a git patch by saying --git nor give extended
3700         * headers.  While at it, maybe please "kompare" that wants
3701         * the trailing TAB and some garbage at the end of line ;-).
3702         */
3703        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3704                patch->new_name, patch->new_name);
3705        for (cnt = 1, frag = patch->fragments;
3706             frag;
3707             cnt++, frag = frag->next) {
3708                if (!frag->rejected) {
3709                        fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
3710                        continue;
3711                }
3712                fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
3713                fprintf(rej, "%.*s", frag->size, frag->patch);
3714                if (frag->patch[frag->size-1] != '\n')
3715                        fputc('\n', rej);
3716        }
3717        fclose(rej);
3718        return -1;
3719}
3720
3721static int write_out_results(struct patch *list)
3722{
3723        int phase;
3724        int errs = 0;
3725        struct patch *l;
3726
3727        for (phase = 0; phase < 2; phase++) {
3728                l = list;
3729                while (l) {
3730                        if (l->rejected)
3731                                errs = 1;
3732                        else {
3733                                write_out_one_result(l, phase);
3734                                if (phase == 1 && write_out_one_reject(l))
3735                                        errs = 1;
3736                        }
3737                        l = l->next;
3738                }
3739        }
3740        return errs;
3741}
3742
3743static struct lock_file lock_file;
3744
3745static struct string_list limit_by_name;
3746static int has_include;
3747static void add_name_limit(const char *name, int exclude)
3748{
3749        struct string_list_item *it;
3750
3751        it = string_list_append(&limit_by_name, name);
3752        it->util = exclude ? NULL : (void *) 1;
3753}
3754
3755static int use_patch(struct patch *p)
3756{
3757        const char *pathname = p->new_name ? p->new_name : p->old_name;
3758        int i;
3759
3760        /* Paths outside are not touched regardless of "--include" */
3761        if (0 < prefix_length) {
3762                int pathlen = strlen(pathname);
3763                if (pathlen <= prefix_length ||
3764                    memcmp(prefix, pathname, prefix_length))
3765                        return 0;
3766        }
3767
3768        /* See if it matches any of exclude/include rule */
3769        for (i = 0; i < limit_by_name.nr; i++) {
3770                struct string_list_item *it = &limit_by_name.items[i];
3771                if (!fnmatch(it->string, pathname, 0))
3772                        return (it->util != NULL);
3773        }
3774
3775        /*
3776         * If we had any include, a path that does not match any rule is
3777         * not used.  Otherwise, we saw bunch of exclude rules (or none)
3778         * and such a path is used.
3779         */
3780        return !has_include;
3781}
3782
3783
3784static void prefix_one(char **name)
3785{
3786        char *old_name = *name;
3787        if (!old_name)
3788                return;
3789        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3790        free(old_name);
3791}
3792
3793static void prefix_patches(struct patch *p)
3794{
3795        if (!prefix || p->is_toplevel_relative)
3796                return;
3797        for ( ; p; p = p->next) {
3798                prefix_one(&p->new_name);
3799                prefix_one(&p->old_name);
3800        }
3801}
3802
3803#define INACCURATE_EOF  (1<<0)
3804#define RECOUNT         (1<<1)
3805
3806static int apply_patch(int fd, const char *filename, int options)
3807{
3808        size_t offset;
3809        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
3810        struct patch *list = NULL, **listp = &list;
3811        int skipped_patch = 0;
3812
3813        patch_input_file = filename;
3814        read_patch_file(&buf, fd);
3815        offset = 0;
3816        while (offset < buf.len) {
3817                struct patch *patch;
3818                int nr;
3819
3820                patch = xcalloc(1, sizeof(*patch));
3821                patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3822                patch->recount =  !!(options & RECOUNT);
3823                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3824                if (nr < 0)
3825                        break;
3826                if (apply_in_reverse)
3827                        reverse_patches(patch);
3828                if (prefix)
3829                        prefix_patches(patch);
3830                if (use_patch(patch)) {
3831                        patch_stats(patch);
3832                        *listp = patch;
3833                        listp = &patch->next;
3834                }
3835                else {
3836                        free_patch(patch);
3837                        skipped_patch++;
3838                }
3839                offset += nr;
3840        }
3841
3842        if (!list && !skipped_patch)
3843                die(_("unrecognized input"));
3844
3845        if (whitespace_error && (ws_error_action == die_on_ws_error))
3846                apply = 0;
3847
3848        update_index = check_index && apply;
3849        if (update_index && newfd < 0)
3850                newfd = hold_locked_index(&lock_file, 1);
3851
3852        if (check_index) {
3853                if (read_cache() < 0)
3854                        die(_("unable to read index file"));
3855        }
3856
3857        if ((check || apply) &&
3858            check_patch_list(list) < 0 &&
3859            !apply_with_reject)
3860                exit(1);
3861
3862        if (apply && write_out_results(list))
3863                exit(1);
3864
3865        if (fake_ancestor)
3866                build_fake_ancestor(list, fake_ancestor);
3867
3868        if (diffstat)
3869                stat_patch_list(list);
3870
3871        if (numstat)
3872                numstat_patch_list(list);
3873
3874        if (summary)
3875                summary_patch_list(list);
3876
3877        free_patch_list(list);
3878        strbuf_release(&buf);
3879        string_list_clear(&fn_table, 0);
3880        return 0;
3881}
3882
3883static int git_apply_config(const char *var, const char *value, void *cb)
3884{
3885        if (!strcmp(var, "apply.whitespace"))
3886                return git_config_string(&apply_default_whitespace, var, value);
3887        else if (!strcmp(var, "apply.ignorewhitespace"))
3888                return git_config_string(&apply_default_ignorewhitespace, var, value);
3889        return git_default_config(var, value, cb);
3890}
3891
3892static int option_parse_exclude(const struct option *opt,
3893                                const char *arg, int unset)
3894{
3895        add_name_limit(arg, 1);
3896        return 0;
3897}
3898
3899static int option_parse_include(const struct option *opt,
3900                                const char *arg, int unset)
3901{
3902        add_name_limit(arg, 0);
3903        has_include = 1;
3904        return 0;
3905}
3906
3907static int option_parse_p(const struct option *opt,
3908                          const char *arg, int unset)
3909{
3910        p_value = atoi(arg);
3911        p_value_known = 1;
3912        return 0;
3913}
3914
3915static int option_parse_z(const struct option *opt,
3916                          const char *arg, int unset)
3917{
3918        if (unset)
3919                line_termination = '\n';
3920        else
3921                line_termination = 0;
3922        return 0;
3923}
3924
3925static int option_parse_space_change(const struct option *opt,
3926                          const char *arg, int unset)
3927{
3928        if (unset)
3929                ws_ignore_action = ignore_ws_none;
3930        else
3931                ws_ignore_action = ignore_ws_change;
3932        return 0;
3933}
3934
3935static int option_parse_whitespace(const struct option *opt,
3936                                   const char *arg, int unset)
3937{
3938        const char **whitespace_option = opt->value;
3939
3940        *whitespace_option = arg;
3941        parse_whitespace_option(arg);
3942        return 0;
3943}
3944
3945static int option_parse_directory(const struct option *opt,
3946                                  const char *arg, int unset)
3947{
3948        root_len = strlen(arg);
3949        if (root_len && arg[root_len - 1] != '/') {
3950                char *new_root;
3951                root = new_root = xmalloc(root_len + 2);
3952                strcpy(new_root, arg);
3953                strcpy(new_root + root_len++, "/");
3954        } else
3955                root = arg;
3956        return 0;
3957}
3958
3959int cmd_apply(int argc, const char **argv, const char *prefix_)
3960{
3961        int i;
3962        int errs = 0;
3963        int is_not_gitdir = !startup_info->have_repository;
3964        int force_apply = 0;
3965
3966        const char *whitespace_option = NULL;
3967
3968        struct option builtin_apply_options[] = {
3969                { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
3970                        N_("don't apply changes matching the given path"),
3971                        0, option_parse_exclude },
3972                { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
3973                        N_("apply changes matching the given path"),
3974                        0, option_parse_include },
3975                { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
3976                        N_("remove <num> leading slashes from traditional diff paths"),
3977                        0, option_parse_p },
3978                OPT_BOOLEAN(0, "no-add", &no_add,
3979                        N_("ignore additions made by the patch")),
3980                OPT_BOOLEAN(0, "stat", &diffstat,
3981                        N_("instead of applying the patch, output diffstat for the input")),
3982                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
3983                OPT_NOOP_NOARG(0, "binary"),
3984                OPT_BOOLEAN(0, "numstat", &numstat,
3985                        N_("shows number of added and deleted lines in decimal notation")),
3986                OPT_BOOLEAN(0, "summary", &summary,
3987                        N_("instead of applying the patch, output a summary for the input")),
3988                OPT_BOOLEAN(0, "check", &check,
3989                        N_("instead of applying the patch, see if the patch is applicable")),
3990                OPT_BOOLEAN(0, "index", &check_index,
3991                        N_("make sure the patch is applicable to the current index")),
3992                OPT_BOOLEAN(0, "cached", &cached,
3993                        N_("apply a patch without touching the working tree")),
3994                OPT_BOOLEAN(0, "apply", &force_apply,
3995                        N_("also apply the patch (use with --stat/--summary/--check)")),
3996                OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3997                        N_("build a temporary index based on embedded index information")),
3998                { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3999                        N_("paths are separated with NUL character"),
4000                        PARSE_OPT_NOARG, option_parse_z },
4001                OPT_INTEGER('C', NULL, &p_context,
4002                                N_("ensure at least <n> lines of context match")),
4003                { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4004                        N_("detect new or modified lines that have whitespace errors"),
4005                        0, option_parse_whitespace },
4006                { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4007                        N_("ignore changes in whitespace when finding context"),
4008                        PARSE_OPT_NOARG, option_parse_space_change },
4009                { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4010                        N_("ignore changes in whitespace when finding context"),
4011                        PARSE_OPT_NOARG, option_parse_space_change },
4012                OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
4013                        N_("apply the patch in reverse")),
4014                OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
4015                        N_("don't expect at least one line of context")),
4016                OPT_BOOLEAN(0, "reject", &apply_with_reject,
4017                        N_("leave the rejected hunks in corresponding *.rej files")),
4018                OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
4019                        N_("allow overlapping hunks")),
4020                OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4021                OPT_BIT(0, "inaccurate-eof", &options,
4022                        N_("tolerate incorrectly detected missing new-line at the end of file"),
4023                        INACCURATE_EOF),
4024                OPT_BIT(0, "recount", &options,
4025                        N_("do not trust the line counts in the hunk headers"),
4026                        RECOUNT),
4027                { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4028                        N_("prepend <root> to all filenames"),
4029                        0, option_parse_directory },
4030                OPT_END()
4031        };
4032
4033        prefix = prefix_;
4034        prefix_length = prefix ? strlen(prefix) : 0;
4035        git_config(git_apply_config, NULL);
4036        if (apply_default_whitespace)
4037                parse_whitespace_option(apply_default_whitespace);
4038        if (apply_default_ignorewhitespace)
4039                parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4040
4041        argc = parse_options(argc, argv, prefix, builtin_apply_options,
4042                        apply_usage, 0);
4043
4044        if (apply_with_reject)
4045                apply = apply_verbosely = 1;
4046        if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4047                apply = 0;
4048        if (check_index && is_not_gitdir)
4049                die(_("--index outside a repository"));
4050        if (cached) {
4051                if (is_not_gitdir)
4052                        die(_("--cached outside a repository"));
4053                check_index = 1;
4054        }
4055        for (i = 0; i < argc; i++) {
4056                const char *arg = argv[i];
4057                int fd;
4058
4059                if (!strcmp(arg, "-")) {
4060                        errs |= apply_patch(0, "<stdin>", options);
4061                        read_stdin = 0;
4062                        continue;
4063                } else if (0 < prefix_length)
4064                        arg = prefix_filename(prefix, prefix_length, arg);
4065
4066                fd = open(arg, O_RDONLY);
4067                if (fd < 0)
4068                        die_errno(_("can't open patch '%s'"), arg);
4069                read_stdin = 0;
4070                set_default_whitespace_mode(whitespace_option);
4071                errs |= apply_patch(fd, arg, options);
4072                close(fd);
4073        }
4074        set_default_whitespace_mode(whitespace_option);
4075        if (read_stdin)
4076                errs |= apply_patch(0, "<stdin>", options);
4077        if (whitespace_error) {
4078                if (squelch_whitespace_errors &&
4079                    squelch_whitespace_errors < whitespace_error) {
4080                        int squelched =
4081                                whitespace_error - squelch_whitespace_errors;
4082                        warning(Q_("squelched %d whitespace error",
4083                                   "squelched %d whitespace errors",
4084                                   squelched),
4085                                squelched);
4086                }
4087                if (ws_error_action == die_on_ws_error)
4088                        die(Q_("%d line adds whitespace errors.",
4089                               "%d lines add whitespace errors.",
4090                               whitespace_error),
4091                            whitespace_error);
4092                if (applied_after_fixing_ws && apply)
4093                        warning("%d line%s applied after"
4094                                " fixing whitespace errors.",
4095                                applied_after_fixing_ws,
4096                                applied_after_fixing_ws == 1 ? "" : "s");
4097                else if (whitespace_error)
4098                        warning(Q_("%d line adds whitespace errors.",
4099                                   "%d lines add whitespace errors.",
4100                                   whitespace_error),
4101                                whitespace_error);
4102        }
4103
4104        if (update_index) {
4105                if (write_cache(newfd, active_cache, active_nr) ||
4106                    commit_locked_index(&lock_file))
4107                        die(_("Unable to write new index file"));
4108        }
4109
4110        return !!errs;
4111}