builtin / apply.con commit apply: move check_to_create_blob() closer to its sole caller (813ebf8)
   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        "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        unsigned long deflate_origlen;
 188        int lines_added, lines_deleted;
 189        int score;
 190        unsigned int is_toplevel_relative:1;
 191        unsigned int inaccurate_eof:1;
 192        unsigned int is_binary:1;
 193        unsigned int is_copy:1;
 194        unsigned int is_rename:1;
 195        unsigned int recount:1;
 196        struct fragment *fragments;
 197        char *result;
 198        size_t resultsize;
 199        char old_sha1_prefix[41];
 200        char new_sha1_prefix[41];
 201        struct patch *next;
 202};
 203
 204static void free_fragment_list(struct fragment *list)
 205{
 206        while (list) {
 207                struct fragment *next = list->next;
 208                if (list->free_patch)
 209                        free((char *)list->patch);
 210                free(list);
 211                list = next;
 212        }
 213}
 214
 215static void free_patch(struct patch *patch)
 216{
 217        free_fragment_list(patch->fragments);
 218        free(patch->def_name);
 219        free(patch->old_name);
 220        free(patch->new_name);
 221        free(patch->result);
 222        free(patch);
 223}
 224
 225static void free_patch_list(struct patch *list)
 226{
 227        while (list) {
 228                struct patch *next = list->next;
 229                free_patch(list);
 230                list = next;
 231        }
 232}
 233
 234/*
 235 * A line in a file, len-bytes long (includes the terminating LF,
 236 * except for an incomplete line at the end if the file ends with
 237 * one), and its contents hashes to 'hash'.
 238 */
 239struct line {
 240        size_t len;
 241        unsigned hash : 24;
 242        unsigned flag : 8;
 243#define LINE_COMMON     1
 244#define LINE_PATCHED    2
 245};
 246
 247/*
 248 * This represents a "file", which is an array of "lines".
 249 */
 250struct image {
 251        char *buf;
 252        size_t len;
 253        size_t nr;
 254        size_t alloc;
 255        struct line *line_allocated;
 256        struct line *line;
 257};
 258
 259/*
 260 * Records filenames that have been touched, in order to handle
 261 * the case where more than one patches touch the same file.
 262 */
 263
 264static struct string_list fn_table;
 265
 266static uint32_t hash_line(const char *cp, size_t len)
 267{
 268        size_t i;
 269        uint32_t h;
 270        for (i = 0, h = 0; i < len; i++) {
 271                if (!isspace(cp[i])) {
 272                        h = h * 3 + (cp[i] & 0xff);
 273                }
 274        }
 275        return h;
 276}
 277
 278/*
 279 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 280 * whitespace difference. Returns 1 if they match, 0 otherwise
 281 */
 282static int fuzzy_matchlines(const char *s1, size_t n1,
 283                            const char *s2, size_t n2)
 284{
 285        const char *last1 = s1 + n1 - 1;
 286        const char *last2 = s2 + n2 - 1;
 287        int result = 0;
 288
 289        /* ignore line endings */
 290        while ((*last1 == '\r') || (*last1 == '\n'))
 291                last1--;
 292        while ((*last2 == '\r') || (*last2 == '\n'))
 293                last2--;
 294
 295        /* skip leading whitespace */
 296        while (isspace(*s1) && (s1 <= last1))
 297                s1++;
 298        while (isspace(*s2) && (s2 <= last2))
 299                s2++;
 300        /* early return if both lines are empty */
 301        if ((s1 > last1) && (s2 > last2))
 302                return 1;
 303        while (!result) {
 304                result = *s1++ - *s2++;
 305                /*
 306                 * Skip whitespace inside. We check for whitespace on
 307                 * both buffers because we don't want "a b" to match
 308                 * "ab"
 309                 */
 310                if (isspace(*s1) && isspace(*s2)) {
 311                        while (isspace(*s1) && s1 <= last1)
 312                                s1++;
 313                        while (isspace(*s2) && s2 <= last2)
 314                                s2++;
 315                }
 316                /*
 317                 * If we reached the end on one side only,
 318                 * lines don't match
 319                 */
 320                if (
 321                    ((s2 > last2) && (s1 <= last1)) ||
 322                    ((s1 > last1) && (s2 <= last2)))
 323                        return 0;
 324                if ((s1 > last1) && (s2 > last2))
 325                        break;
 326        }
 327
 328        return !result;
 329}
 330
 331static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
 332{
 333        ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
 334        img->line_allocated[img->nr].len = len;
 335        img->line_allocated[img->nr].hash = hash_line(bol, len);
 336        img->line_allocated[img->nr].flag = flag;
 337        img->nr++;
 338}
 339
 340/*
 341 * "buf" has the file contents to be patched (read from various sources).
 342 * attach it to "image" and add line-based index to it.
 343 * "image" now owns the "buf".
 344 */
 345static void prepare_image(struct image *image, char *buf, size_t len,
 346                          int prepare_linetable)
 347{
 348        const char *cp, *ep;
 349
 350        memset(image, 0, sizeof(*image));
 351        image->buf = buf;
 352        image->len = len;
 353
 354        if (!prepare_linetable)
 355                return;
 356
 357        ep = image->buf + image->len;
 358        cp = image->buf;
 359        while (cp < ep) {
 360                const char *next;
 361                for (next = cp; next < ep && *next != '\n'; next++)
 362                        ;
 363                if (next < ep)
 364                        next++;
 365                add_line_info(image, cp, next - cp, 0);
 366                cp = next;
 367        }
 368        image->line = image->line_allocated;
 369}
 370
 371static void clear_image(struct image *image)
 372{
 373        free(image->buf);
 374        free(image->line_allocated);
 375        memset(image, 0, sizeof(*image));
 376}
 377
 378/* fmt must contain _one_ %s and no other substitution */
 379static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
 380{
 381        struct strbuf sb = STRBUF_INIT;
 382
 383        if (patch->old_name && patch->new_name &&
 384            strcmp(patch->old_name, patch->new_name)) {
 385                quote_c_style(patch->old_name, &sb, NULL, 0);
 386                strbuf_addstr(&sb, " => ");
 387                quote_c_style(patch->new_name, &sb, NULL, 0);
 388        } else {
 389                const char *n = patch->new_name;
 390                if (!n)
 391                        n = patch->old_name;
 392                quote_c_style(n, &sb, NULL, 0);
 393        }
 394        fprintf(output, fmt, sb.buf);
 395        fputc('\n', output);
 396        strbuf_release(&sb);
 397}
 398
 399#define SLOP (16)
 400
 401static void read_patch_file(struct strbuf *sb, int fd)
 402{
 403        if (strbuf_read(sb, fd, 0) < 0)
 404                die_errno("git apply: failed to read");
 405
 406        /*
 407         * Make sure that we have some slop in the buffer
 408         * so that we can do speculative "memcmp" etc, and
 409         * see to it that it is NUL-filled.
 410         */
 411        strbuf_grow(sb, SLOP);
 412        memset(sb->buf + sb->len, 0, SLOP);
 413}
 414
 415static unsigned long linelen(const char *buffer, unsigned long size)
 416{
 417        unsigned long len = 0;
 418        while (size--) {
 419                len++;
 420                if (*buffer++ == '\n')
 421                        break;
 422        }
 423        return len;
 424}
 425
 426static int is_dev_null(const char *str)
 427{
 428        return !memcmp("/dev/null", str, 9) && isspace(str[9]);
 429}
 430
 431#define TERM_SPACE      1
 432#define TERM_TAB        2
 433
 434static int name_terminate(const char *name, int namelen, int c, int terminate)
 435{
 436        if (c == ' ' && !(terminate & TERM_SPACE))
 437                return 0;
 438        if (c == '\t' && !(terminate & TERM_TAB))
 439                return 0;
 440
 441        return 1;
 442}
 443
 444/* remove double slashes to make --index work with such filenames */
 445static char *squash_slash(char *name)
 446{
 447        int i = 0, j = 0;
 448
 449        if (!name)
 450                return NULL;
 451
 452        while (name[i]) {
 453                if ((name[j++] = name[i++]) == '/')
 454                        while (name[i] == '/')
 455                                i++;
 456        }
 457        name[j] = '\0';
 458        return name;
 459}
 460
 461static char *find_name_gnu(const char *line, const char *def, int p_value)
 462{
 463        struct strbuf name = STRBUF_INIT;
 464        char *cp;
 465
 466        /*
 467         * Proposed "new-style" GNU patch/diff format; see
 468         * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
 469         */
 470        if (unquote_c_style(&name, line, NULL)) {
 471                strbuf_release(&name);
 472                return NULL;
 473        }
 474
 475        for (cp = name.buf; p_value; p_value--) {
 476                cp = strchr(cp, '/');
 477                if (!cp) {
 478                        strbuf_release(&name);
 479                        return NULL;
 480                }
 481                cp++;
 482        }
 483
 484        strbuf_remove(&name, 0, cp - name.buf);
 485        if (root)
 486                strbuf_insert(&name, 0, root, root_len);
 487        return squash_slash(strbuf_detach(&name, NULL));
 488}
 489
 490static size_t sane_tz_len(const char *line, size_t len)
 491{
 492        const char *tz, *p;
 493
 494        if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
 495                return 0;
 496        tz = line + len - strlen(" +0500");
 497
 498        if (tz[1] != '+' && tz[1] != '-')
 499                return 0;
 500
 501        for (p = tz + 2; p != line + len; p++)
 502                if (!isdigit(*p))
 503                        return 0;
 504
 505        return line + len - tz;
 506}
 507
 508static size_t tz_with_colon_len(const char *line, size_t len)
 509{
 510        const char *tz, *p;
 511
 512        if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
 513                return 0;
 514        tz = line + len - strlen(" +08:00");
 515
 516        if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
 517                return 0;
 518        p = tz + 2;
 519        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 520            !isdigit(*p++) || !isdigit(*p++))
 521                return 0;
 522
 523        return line + len - tz;
 524}
 525
 526static size_t date_len(const char *line, size_t len)
 527{
 528        const char *date, *p;
 529
 530        if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
 531                return 0;
 532        p = date = line + len - strlen("72-02-05");
 533
 534        if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 535            !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
 536            !isdigit(*p++) || !isdigit(*p++))   /* Not a date. */
 537                return 0;
 538
 539        if (date - line >= strlen("19") &&
 540            isdigit(date[-1]) && isdigit(date[-2]))     /* 4-digit year */
 541                date -= strlen("19");
 542
 543        return line + len - date;
 544}
 545
 546static size_t short_time_len(const char *line, size_t len)
 547{
 548        const char *time, *p;
 549
 550        if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
 551                return 0;
 552        p = time = line + len - strlen(" 07:01:32");
 553
 554        /* Permit 1-digit hours? */
 555        if (*p++ != ' ' ||
 556            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 557            !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
 558            !isdigit(*p++) || !isdigit(*p++))   /* Not a time. */
 559                return 0;
 560
 561        return line + len - time;
 562}
 563
 564static size_t fractional_time_len(const char *line, size_t len)
 565{
 566        const char *p;
 567        size_t n;
 568
 569        /* Expected format: 19:41:17.620000023 */
 570        if (!len || !isdigit(line[len - 1]))
 571                return 0;
 572        p = line + len - 1;
 573
 574        /* Fractional seconds. */
 575        while (p > line && isdigit(*p))
 576                p--;
 577        if (*p != '.')
 578                return 0;
 579
 580        /* Hours, minutes, and whole seconds. */
 581        n = short_time_len(line, p - line);
 582        if (!n)
 583                return 0;
 584
 585        return line + len - p + n;
 586}
 587
 588static size_t trailing_spaces_len(const char *line, size_t len)
 589{
 590        const char *p;
 591
 592        /* Expected format: ' ' x (1 or more)  */
 593        if (!len || line[len - 1] != ' ')
 594                return 0;
 595
 596        p = line + len;
 597        while (p != line) {
 598                p--;
 599                if (*p != ' ')
 600                        return line + len - (p + 1);
 601        }
 602
 603        /* All spaces! */
 604        return len;
 605}
 606
 607static size_t diff_timestamp_len(const char *line, size_t len)
 608{
 609        const char *end = line + len;
 610        size_t n;
 611
 612        /*
 613         * Posix: 2010-07-05 19:41:17
 614         * GNU: 2010-07-05 19:41:17.620000023 -0500
 615         */
 616
 617        if (!isdigit(end[-1]))
 618                return 0;
 619
 620        n = sane_tz_len(line, end - line);
 621        if (!n)
 622                n = tz_with_colon_len(line, end - line);
 623        end -= n;
 624
 625        n = short_time_len(line, end - line);
 626        if (!n)
 627                n = fractional_time_len(line, end - line);
 628        end -= n;
 629
 630        n = date_len(line, end - line);
 631        if (!n) /* No date.  Too bad. */
 632                return 0;
 633        end -= n;
 634
 635        if (end == line)        /* No space before date. */
 636                return 0;
 637        if (end[-1] == '\t') {  /* Success! */
 638                end--;
 639                return line + len - end;
 640        }
 641        if (end[-1] != ' ')     /* No space before date. */
 642                return 0;
 643
 644        /* Whitespace damage. */
 645        end -= trailing_spaces_len(line, end - line);
 646        return line + len - end;
 647}
 648
 649static char *null_strdup(const char *s)
 650{
 651        return s ? xstrdup(s) : NULL;
 652}
 653
 654static char *find_name_common(const char *line, const char *def,
 655                              int p_value, const char *end, int terminate)
 656{
 657        int len;
 658        const char *start = NULL;
 659
 660        if (p_value == 0)
 661                start = line;
 662        while (line != end) {
 663                char c = *line;
 664
 665                if (!end && isspace(c)) {
 666                        if (c == '\n')
 667                                break;
 668                        if (name_terminate(start, line-start, c, terminate))
 669                                break;
 670                }
 671                line++;
 672                if (c == '/' && !--p_value)
 673                        start = line;
 674        }
 675        if (!start)
 676                return squash_slash(null_strdup(def));
 677        len = line - start;
 678        if (!len)
 679                return squash_slash(null_strdup(def));
 680
 681        /*
 682         * Generally we prefer the shorter name, especially
 683         * if the other one is just a variation of that with
 684         * something else tacked on to the end (ie "file.orig"
 685         * or "file~").
 686         */
 687        if (def) {
 688                int deflen = strlen(def);
 689                if (deflen < len && !strncmp(start, def, deflen))
 690                        return squash_slash(xstrdup(def));
 691        }
 692
 693        if (root) {
 694                char *ret = xmalloc(root_len + len + 1);
 695                strcpy(ret, root);
 696                memcpy(ret + root_len, start, len);
 697                ret[root_len + len] = '\0';
 698                return squash_slash(ret);
 699        }
 700
 701        return squash_slash(xmemdupz(start, len));
 702}
 703
 704static char *find_name(const char *line, char *def, int p_value, int terminate)
 705{
 706        if (*line == '"') {
 707                char *name = find_name_gnu(line, def, p_value);
 708                if (name)
 709                        return name;
 710        }
 711
 712        return find_name_common(line, def, p_value, NULL, terminate);
 713}
 714
 715static char *find_name_traditional(const char *line, char *def, int p_value)
 716{
 717        size_t len = strlen(line);
 718        size_t date_len;
 719
 720        if (*line == '"') {
 721                char *name = find_name_gnu(line, def, p_value);
 722                if (name)
 723                        return name;
 724        }
 725
 726        len = strchrnul(line, '\n') - line;
 727        date_len = diff_timestamp_len(line, len);
 728        if (!date_len)
 729                return find_name_common(line, def, p_value, NULL, TERM_TAB);
 730        len -= date_len;
 731
 732        return find_name_common(line, def, p_value, line + len, 0);
 733}
 734
 735static int count_slashes(const char *cp)
 736{
 737        int cnt = 0;
 738        char ch;
 739
 740        while ((ch = *cp++))
 741                if (ch == '/')
 742                        cnt++;
 743        return cnt;
 744}
 745
 746/*
 747 * Given the string after "--- " or "+++ ", guess the appropriate
 748 * p_value for the given patch.
 749 */
 750static int guess_p_value(const char *nameline)
 751{
 752        char *name, *cp;
 753        int val = -1;
 754
 755        if (is_dev_null(nameline))
 756                return -1;
 757        name = find_name_traditional(nameline, NULL, 0);
 758        if (!name)
 759                return -1;
 760        cp = strchr(name, '/');
 761        if (!cp)
 762                val = 0;
 763        else if (prefix) {
 764                /*
 765                 * Does it begin with "a/$our-prefix" and such?  Then this is
 766                 * very likely to apply to our directory.
 767                 */
 768                if (!strncmp(name, prefix, prefix_length))
 769                        val = count_slashes(prefix);
 770                else {
 771                        cp++;
 772                        if (!strncmp(cp, prefix, prefix_length))
 773                                val = count_slashes(prefix) + 1;
 774                }
 775        }
 776        free(name);
 777        return val;
 778}
 779
 780/*
 781 * Does the ---/+++ line has the POSIX timestamp after the last HT?
 782 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 783 * this such a timestamp?
 784 */
 785static int has_epoch_timestamp(const char *nameline)
 786{
 787        /*
 788         * We are only interested in epoch timestamp; any non-zero
 789         * fraction cannot be one, hence "(\.0+)?" in the regexp below.
 790         * For the same reason, the date must be either 1969-12-31 or
 791         * 1970-01-01, and the seconds part must be "00".
 792         */
 793        const char stamp_regexp[] =
 794                "^(1969-12-31|1970-01-01)"
 795                " "
 796                "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
 797                " "
 798                "([-+][0-2][0-9]:?[0-5][0-9])\n";
 799        const char *timestamp = NULL, *cp, *colon;
 800        static regex_t *stamp;
 801        regmatch_t m[10];
 802        int zoneoffset;
 803        int hourminute;
 804        int status;
 805
 806        for (cp = nameline; *cp != '\n'; cp++) {
 807                if (*cp == '\t')
 808                        timestamp = cp + 1;
 809        }
 810        if (!timestamp)
 811                return 0;
 812        if (!stamp) {
 813                stamp = xmalloc(sizeof(*stamp));
 814                if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
 815                        warning(_("Cannot prepare timestamp regexp %s"),
 816                                stamp_regexp);
 817                        return 0;
 818                }
 819        }
 820
 821        status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
 822        if (status) {
 823                if (status != REG_NOMATCH)
 824                        warning(_("regexec returned %d for input: %s"),
 825                                status, timestamp);
 826                return 0;
 827        }
 828
 829        zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
 830        if (*colon == ':')
 831                zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
 832        else
 833                zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
 834        if (timestamp[m[3].rm_so] == '-')
 835                zoneoffset = -zoneoffset;
 836
 837        /*
 838         * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
 839         * (west of GMT) or 1970-01-01 (east of GMT)
 840         */
 841        if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
 842            (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
 843                return 0;
 844
 845        hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
 846                      strtol(timestamp + 14, NULL, 10) -
 847                      zoneoffset);
 848
 849        return ((zoneoffset < 0 && hourminute == 1440) ||
 850                (0 <= zoneoffset && !hourminute));
 851}
 852
 853/*
 854 * Get the name etc info from the ---/+++ lines of a traditional patch header
 855 *
 856 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 857 * files, we can happily check the index for a match, but for creating a
 858 * new file we should try to match whatever "patch" does. I have no idea.
 859 */
 860static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 861{
 862        char *name;
 863
 864        first += 4;     /* skip "--- " */
 865        second += 4;    /* skip "+++ " */
 866        if (!p_value_known) {
 867                int p, q;
 868                p = guess_p_value(first);
 869                q = guess_p_value(second);
 870                if (p < 0) p = q;
 871                if (0 <= p && p == q) {
 872                        p_value = p;
 873                        p_value_known = 1;
 874                }
 875        }
 876        if (is_dev_null(first)) {
 877                patch->is_new = 1;
 878                patch->is_delete = 0;
 879                name = find_name_traditional(second, NULL, p_value);
 880                patch->new_name = name;
 881        } else if (is_dev_null(second)) {
 882                patch->is_new = 0;
 883                patch->is_delete = 1;
 884                name = find_name_traditional(first, NULL, p_value);
 885                patch->old_name = name;
 886        } else {
 887                char *first_name;
 888                first_name = find_name_traditional(first, NULL, p_value);
 889                name = find_name_traditional(second, first_name, p_value);
 890                free(first_name);
 891                if (has_epoch_timestamp(first)) {
 892                        patch->is_new = 1;
 893                        patch->is_delete = 0;
 894                        patch->new_name = name;
 895                } else if (has_epoch_timestamp(second)) {
 896                        patch->is_new = 0;
 897                        patch->is_delete = 1;
 898                        patch->old_name = name;
 899                } else {
 900                        patch->old_name = name;
 901                        patch->new_name = xstrdup(name);
 902                }
 903        }
 904        if (!name)
 905                die(_("unable to find filename in patch at line %d"), linenr);
 906}
 907
 908static int gitdiff_hdrend(const char *line, struct patch *patch)
 909{
 910        return -1;
 911}
 912
 913/*
 914 * We're anal about diff header consistency, to make
 915 * sure that we don't end up having strange ambiguous
 916 * patches floating around.
 917 *
 918 * As a result, gitdiff_{old|new}name() will check
 919 * their names against any previous information, just
 920 * to make sure..
 921 */
 922static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
 923{
 924        if (!orig_name && !isnull)
 925                return find_name(line, NULL, p_value, TERM_TAB);
 926
 927        if (orig_name) {
 928                int len;
 929                const char *name;
 930                char *another;
 931                name = orig_name;
 932                len = strlen(name);
 933                if (isnull)
 934                        die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
 935                another = find_name(line, NULL, p_value, TERM_TAB);
 936                if (!another || memcmp(another, name, len + 1))
 937                        die(_("git apply: bad git-diff - inconsistent %s filename on line %d"), oldnew, linenr);
 938                free(another);
 939                return orig_name;
 940        }
 941        else {
 942                /* expect "/dev/null" */
 943                if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 944                        die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
 945                return NULL;
 946        }
 947}
 948
 949static int gitdiff_oldname(const char *line, struct patch *patch)
 950{
 951        char *orig = patch->old_name;
 952        patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
 953        if (orig != patch->old_name)
 954                free(orig);
 955        return 0;
 956}
 957
 958static int gitdiff_newname(const char *line, struct patch *patch)
 959{
 960        char *orig = patch->new_name;
 961        patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
 962        if (orig != patch->new_name)
 963                free(orig);
 964        return 0;
 965}
 966
 967static int gitdiff_oldmode(const char *line, struct patch *patch)
 968{
 969        patch->old_mode = strtoul(line, NULL, 8);
 970        return 0;
 971}
 972
 973static int gitdiff_newmode(const char *line, struct patch *patch)
 974{
 975        patch->new_mode = strtoul(line, NULL, 8);
 976        return 0;
 977}
 978
 979static int gitdiff_delete(const char *line, struct patch *patch)
 980{
 981        patch->is_delete = 1;
 982        free(patch->old_name);
 983        patch->old_name = null_strdup(patch->def_name);
 984        return gitdiff_oldmode(line, patch);
 985}
 986
 987static int gitdiff_newfile(const char *line, struct patch *patch)
 988{
 989        patch->is_new = 1;
 990        free(patch->new_name);
 991        patch->new_name = null_strdup(patch->def_name);
 992        return gitdiff_newmode(line, patch);
 993}
 994
 995static int gitdiff_copysrc(const char *line, struct patch *patch)
 996{
 997        patch->is_copy = 1;
 998        free(patch->old_name);
 999        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1000        return 0;
1001}
1002
1003static int gitdiff_copydst(const char *line, struct patch *patch)
1004{
1005        patch->is_copy = 1;
1006        free(patch->new_name);
1007        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1008        return 0;
1009}
1010
1011static int gitdiff_renamesrc(const char *line, struct patch *patch)
1012{
1013        patch->is_rename = 1;
1014        free(patch->old_name);
1015        patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1016        return 0;
1017}
1018
1019static int gitdiff_renamedst(const char *line, struct patch *patch)
1020{
1021        patch->is_rename = 1;
1022        free(patch->new_name);
1023        patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1024        return 0;
1025}
1026
1027static int gitdiff_similarity(const char *line, struct patch *patch)
1028{
1029        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1030                patch->score = 0;
1031        return 0;
1032}
1033
1034static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1035{
1036        if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1037                patch->score = 0;
1038        return 0;
1039}
1040
1041static int gitdiff_index(const char *line, struct patch *patch)
1042{
1043        /*
1044         * index line is N hexadecimal, "..", N hexadecimal,
1045         * and optional space with octal mode.
1046         */
1047        const char *ptr, *eol;
1048        int len;
1049
1050        ptr = strchr(line, '.');
1051        if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1052                return 0;
1053        len = ptr - line;
1054        memcpy(patch->old_sha1_prefix, line, len);
1055        patch->old_sha1_prefix[len] = 0;
1056
1057        line = ptr + 2;
1058        ptr = strchr(line, ' ');
1059        eol = strchr(line, '\n');
1060
1061        if (!ptr || eol < ptr)
1062                ptr = eol;
1063        len = ptr - line;
1064
1065        if (40 < len)
1066                return 0;
1067        memcpy(patch->new_sha1_prefix, line, len);
1068        patch->new_sha1_prefix[len] = 0;
1069        if (*ptr == ' ')
1070                patch->old_mode = strtoul(ptr+1, NULL, 8);
1071        return 0;
1072}
1073
1074/*
1075 * This is normal for a diff that doesn't change anything: we'll fall through
1076 * into the next diff. Tell the parser to break out.
1077 */
1078static int gitdiff_unrecognized(const char *line, struct patch *patch)
1079{
1080        return -1;
1081}
1082
1083static const char *stop_at_slash(const char *line, int llen)
1084{
1085        int nslash = p_value;
1086        int i;
1087
1088        for (i = 0; i < llen; i++) {
1089                int ch = line[i];
1090                if (ch == '/' && --nslash <= 0)
1091                        return &line[i];
1092        }
1093        return NULL;
1094}
1095
1096/*
1097 * This is to extract the same name that appears on "diff --git"
1098 * line.  We do not find and return anything if it is a rename
1099 * patch, and it is OK because we will find the name elsewhere.
1100 * We need to reliably find name only when it is mode-change only,
1101 * creation or deletion of an empty file.  In any of these cases,
1102 * both sides are the same name under a/ and b/ respectively.
1103 */
1104static char *git_header_name(const char *line, int llen)
1105{
1106        const char *name;
1107        const char *second = NULL;
1108        size_t len, line_len;
1109
1110        line += strlen("diff --git ");
1111        llen -= strlen("diff --git ");
1112
1113        if (*line == '"') {
1114                const char *cp;
1115                struct strbuf first = STRBUF_INIT;
1116                struct strbuf sp = STRBUF_INIT;
1117
1118                if (unquote_c_style(&first, line, &second))
1119                        goto free_and_fail1;
1120
1121                /* advance to the first slash */
1122                cp = stop_at_slash(first.buf, first.len);
1123                /* we do not accept absolute paths */
1124                if (!cp || cp == first.buf)
1125                        goto free_and_fail1;
1126                strbuf_remove(&first, 0, cp + 1 - first.buf);
1127
1128                /*
1129                 * second points at one past closing dq of name.
1130                 * find the second name.
1131                 */
1132                while ((second < line + llen) && isspace(*second))
1133                        second++;
1134
1135                if (line + llen <= second)
1136                        goto free_and_fail1;
1137                if (*second == '"') {
1138                        if (unquote_c_style(&sp, second, NULL))
1139                                goto free_and_fail1;
1140                        cp = stop_at_slash(sp.buf, sp.len);
1141                        if (!cp || cp == sp.buf)
1142                                goto free_and_fail1;
1143                        /* They must match, otherwise ignore */
1144                        if (strcmp(cp + 1, first.buf))
1145                                goto free_and_fail1;
1146                        strbuf_release(&sp);
1147                        return strbuf_detach(&first, NULL);
1148                }
1149
1150                /* unquoted second */
1151                cp = stop_at_slash(second, line + llen - second);
1152                if (!cp || cp == second)
1153                        goto free_and_fail1;
1154                cp++;
1155                if (line + llen - cp != first.len + 1 ||
1156                    memcmp(first.buf, cp, first.len))
1157                        goto free_and_fail1;
1158                return strbuf_detach(&first, NULL);
1159
1160        free_and_fail1:
1161                strbuf_release(&first);
1162                strbuf_release(&sp);
1163                return NULL;
1164        }
1165
1166        /* unquoted first name */
1167        name = stop_at_slash(line, llen);
1168        if (!name || name == line)
1169                return NULL;
1170        name++;
1171
1172        /*
1173         * since the first name is unquoted, a dq if exists must be
1174         * the beginning of the second name.
1175         */
1176        for (second = name; second < line + llen; second++) {
1177                if (*second == '"') {
1178                        struct strbuf sp = STRBUF_INIT;
1179                        const char *np;
1180
1181                        if (unquote_c_style(&sp, second, NULL))
1182                                goto free_and_fail2;
1183
1184                        np = stop_at_slash(sp.buf, sp.len);
1185                        if (!np || np == sp.buf)
1186                                goto free_and_fail2;
1187                        np++;
1188
1189                        len = sp.buf + sp.len - np;
1190                        if (len < second - name &&
1191                            !strncmp(np, name, len) &&
1192                            isspace(name[len])) {
1193                                /* Good */
1194                                strbuf_remove(&sp, 0, np - sp.buf);
1195                                return strbuf_detach(&sp, NULL);
1196                        }
1197
1198                free_and_fail2:
1199                        strbuf_release(&sp);
1200                        return NULL;
1201                }
1202        }
1203
1204        /*
1205         * Accept a name only if it shows up twice, exactly the same
1206         * form.
1207         */
1208        second = strchr(name, '\n');
1209        if (!second)
1210                return NULL;
1211        line_len = second - name;
1212        for (len = 0 ; ; len++) {
1213                switch (name[len]) {
1214                default:
1215                        continue;
1216                case '\n':
1217                        return NULL;
1218                case '\t': case ' ':
1219                        second = stop_at_slash(name + len, line_len - len);
1220                        if (!second)
1221                                return NULL;
1222                        second++;
1223                        if (second[len] == '\n' && !strncmp(name, second, len)) {
1224                                return xmemdupz(name, len);
1225                        }
1226                }
1227        }
1228}
1229
1230/* Verify that we recognize the lines following a git header */
1231static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1232{
1233        unsigned long offset;
1234
1235        /* A git diff has explicit new/delete information, so we don't guess */
1236        patch->is_new = 0;
1237        patch->is_delete = 0;
1238
1239        /*
1240         * Some things may not have the old name in the
1241         * rest of the headers anywhere (pure mode changes,
1242         * or removing or adding empty files), so we get
1243         * the default name from the header.
1244         */
1245        patch->def_name = git_header_name(line, len);
1246        if (patch->def_name && root) {
1247                char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1248                strcpy(s, root);
1249                strcpy(s + root_len, patch->def_name);
1250                free(patch->def_name);
1251                patch->def_name = s;
1252        }
1253
1254        line += len;
1255        size -= len;
1256        linenr++;
1257        for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1258                static const struct opentry {
1259                        const char *str;
1260                        int (*fn)(const char *, struct patch *);
1261                } optable[] = {
1262                        { "@@ -", gitdiff_hdrend },
1263                        { "--- ", gitdiff_oldname },
1264                        { "+++ ", gitdiff_newname },
1265                        { "old mode ", gitdiff_oldmode },
1266                        { "new mode ", gitdiff_newmode },
1267                        { "deleted file mode ", gitdiff_delete },
1268                        { "new file mode ", gitdiff_newfile },
1269                        { "copy from ", gitdiff_copysrc },
1270                        { "copy to ", gitdiff_copydst },
1271                        { "rename old ", gitdiff_renamesrc },
1272                        { "rename new ", gitdiff_renamedst },
1273                        { "rename from ", gitdiff_renamesrc },
1274                        { "rename to ", gitdiff_renamedst },
1275                        { "similarity index ", gitdiff_similarity },
1276                        { "dissimilarity index ", gitdiff_dissimilarity },
1277                        { "index ", gitdiff_index },
1278                        { "", gitdiff_unrecognized },
1279                };
1280                int i;
1281
1282                len = linelen(line, size);
1283                if (!len || line[len-1] != '\n')
1284                        break;
1285                for (i = 0; i < ARRAY_SIZE(optable); i++) {
1286                        const struct opentry *p = optable + i;
1287                        int oplen = strlen(p->str);
1288                        if (len < oplen || memcmp(p->str, line, oplen))
1289                                continue;
1290                        if (p->fn(line + oplen, patch) < 0)
1291                                return offset;
1292                        break;
1293                }
1294        }
1295
1296        return offset;
1297}
1298
1299static int parse_num(const char *line, unsigned long *p)
1300{
1301        char *ptr;
1302
1303        if (!isdigit(*line))
1304                return 0;
1305        *p = strtoul(line, &ptr, 10);
1306        return ptr - line;
1307}
1308
1309static int parse_range(const char *line, int len, int offset, const char *expect,
1310                       unsigned long *p1, unsigned long *p2)
1311{
1312        int digits, ex;
1313
1314        if (offset < 0 || offset >= len)
1315                return -1;
1316        line += offset;
1317        len -= offset;
1318
1319        digits = parse_num(line, p1);
1320        if (!digits)
1321                return -1;
1322
1323        offset += digits;
1324        line += digits;
1325        len -= digits;
1326
1327        *p2 = 1;
1328        if (*line == ',') {
1329                digits = parse_num(line+1, p2);
1330                if (!digits)
1331                        return -1;
1332
1333                offset += digits+1;
1334                line += digits+1;
1335                len -= digits+1;
1336        }
1337
1338        ex = strlen(expect);
1339        if (ex > len)
1340                return -1;
1341        if (memcmp(line, expect, ex))
1342                return -1;
1343
1344        return offset + ex;
1345}
1346
1347static void recount_diff(const char *line, int size, struct fragment *fragment)
1348{
1349        int oldlines = 0, newlines = 0, ret = 0;
1350
1351        if (size < 1) {
1352                warning("recount: ignore empty hunk");
1353                return;
1354        }
1355
1356        for (;;) {
1357                int len = linelen(line, size);
1358                size -= len;
1359                line += len;
1360
1361                if (size < 1)
1362                        break;
1363
1364                switch (*line) {
1365                case ' ': case '\n':
1366                        newlines++;
1367                        /* fall through */
1368                case '-':
1369                        oldlines++;
1370                        continue;
1371                case '+':
1372                        newlines++;
1373                        continue;
1374                case '\\':
1375                        continue;
1376                case '@':
1377                        ret = size < 3 || prefixcmp(line, "@@ ");
1378                        break;
1379                case 'd':
1380                        ret = size < 5 || prefixcmp(line, "diff ");
1381                        break;
1382                default:
1383                        ret = -1;
1384                        break;
1385                }
1386                if (ret) {
1387                        warning(_("recount: unexpected line: %.*s"),
1388                                (int)linelen(line, size), line);
1389                        return;
1390                }
1391                break;
1392        }
1393        fragment->oldlines = oldlines;
1394        fragment->newlines = newlines;
1395}
1396
1397/*
1398 * Parse a unified diff fragment header of the
1399 * form "@@ -a,b +c,d @@"
1400 */
1401static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1402{
1403        int offset;
1404
1405        if (!len || line[len-1] != '\n')
1406                return -1;
1407
1408        /* Figure out the number of lines in a fragment */
1409        offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1410        offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1411
1412        return offset;
1413}
1414
1415static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1416{
1417        unsigned long offset, len;
1418
1419        patch->is_toplevel_relative = 0;
1420        patch->is_rename = patch->is_copy = 0;
1421        patch->is_new = patch->is_delete = -1;
1422        patch->old_mode = patch->new_mode = 0;
1423        patch->old_name = patch->new_name = NULL;
1424        for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1425                unsigned long nextlen;
1426
1427                len = linelen(line, size);
1428                if (!len)
1429                        break;
1430
1431                /* Testing this early allows us to take a few shortcuts.. */
1432                if (len < 6)
1433                        continue;
1434
1435                /*
1436                 * Make sure we don't find any unconnected patch fragments.
1437                 * That's a sign that we didn't find a header, and that a
1438                 * patch has become corrupted/broken up.
1439                 */
1440                if (!memcmp("@@ -", line, 4)) {
1441                        struct fragment dummy;
1442                        if (parse_fragment_header(line, len, &dummy) < 0)
1443                                continue;
1444                        die(_("patch fragment without header at line %d: %.*s"),
1445                            linenr, (int)len-1, line);
1446                }
1447
1448                if (size < len + 6)
1449                        break;
1450
1451                /*
1452                 * Git patch? It might not have a real patch, just a rename
1453                 * or mode change, so we handle that specially
1454                 */
1455                if (!memcmp("diff --git ", line, 11)) {
1456                        int git_hdr_len = parse_git_header(line, len, size, patch);
1457                        if (git_hdr_len <= len)
1458                                continue;
1459                        if (!patch->old_name && !patch->new_name) {
1460                                if (!patch->def_name)
1461                                        die(Q_("git diff header lacks filename information when removing "
1462                                               "%d leading pathname component (line %d)",
1463                                               "git diff header lacks filename information when removing "
1464                                               "%d leading pathname components (line %d)",
1465                                               p_value),
1466                                            p_value, linenr);
1467                                patch->old_name = xstrdup(patch->def_name);
1468                                patch->new_name = xstrdup(patch->def_name);
1469                        }
1470                        if (!patch->is_delete && !patch->new_name)
1471                                die("git diff header lacks filename information "
1472                                    "(line %d)", linenr);
1473                        patch->is_toplevel_relative = 1;
1474                        *hdrsize = git_hdr_len;
1475                        return offset;
1476                }
1477
1478                /* --- followed by +++ ? */
1479                if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1480                        continue;
1481
1482                /*
1483                 * We only accept unified patches, so we want it to
1484                 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1485                 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1486                 */
1487                nextlen = linelen(line + len, size - len);
1488                if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1489                        continue;
1490
1491                /* Ok, we'll consider it a patch */
1492                parse_traditional_patch(line, line+len, patch);
1493                *hdrsize = len + nextlen;
1494                linenr += 2;
1495                return offset;
1496        }
1497        return -1;
1498}
1499
1500static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1501{
1502        char *err;
1503
1504        if (!result)
1505                return;
1506
1507        whitespace_error++;
1508        if (squelch_whitespace_errors &&
1509            squelch_whitespace_errors < whitespace_error)
1510                return;
1511
1512        err = whitespace_error_string(result);
1513        fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1514                patch_input_file, linenr, err, len, line);
1515        free(err);
1516}
1517
1518static void check_whitespace(const char *line, int len, unsigned ws_rule)
1519{
1520        unsigned result = ws_check(line + 1, len - 1, ws_rule);
1521
1522        record_ws_error(result, line + 1, len - 2, linenr);
1523}
1524
1525/*
1526 * Parse a unified diff. Note that this really needs to parse each
1527 * fragment separately, since the only way to know the difference
1528 * between a "---" that is part of a patch, and a "---" that starts
1529 * the next patch is to look at the line counts..
1530 */
1531static int parse_fragment(const char *line, unsigned long size,
1532                          struct patch *patch, struct fragment *fragment)
1533{
1534        int added, deleted;
1535        int len = linelen(line, size), offset;
1536        unsigned long oldlines, newlines;
1537        unsigned long leading, trailing;
1538
1539        offset = parse_fragment_header(line, len, fragment);
1540        if (offset < 0)
1541                return -1;
1542        if (offset > 0 && patch->recount)
1543                recount_diff(line + offset, size - offset, fragment);
1544        oldlines = fragment->oldlines;
1545        newlines = fragment->newlines;
1546        leading = 0;
1547        trailing = 0;
1548
1549        /* Parse the thing.. */
1550        line += len;
1551        size -= len;
1552        linenr++;
1553        added = deleted = 0;
1554        for (offset = len;
1555             0 < size;
1556             offset += len, size -= len, line += len, linenr++) {
1557                if (!oldlines && !newlines)
1558                        break;
1559                len = linelen(line, size);
1560                if (!len || line[len-1] != '\n')
1561                        return -1;
1562                switch (*line) {
1563                default:
1564                        return -1;
1565                case '\n': /* newer GNU diff, an empty context line */
1566                case ' ':
1567                        oldlines--;
1568                        newlines--;
1569                        if (!deleted && !added)
1570                                leading++;
1571                        trailing++;
1572                        break;
1573                case '-':
1574                        if (apply_in_reverse &&
1575                            ws_error_action != nowarn_ws_error)
1576                                check_whitespace(line, len, patch->ws_rule);
1577                        deleted++;
1578                        oldlines--;
1579                        trailing = 0;
1580                        break;
1581                case '+':
1582                        if (!apply_in_reverse &&
1583                            ws_error_action != nowarn_ws_error)
1584                                check_whitespace(line, len, patch->ws_rule);
1585                        added++;
1586                        newlines--;
1587                        trailing = 0;
1588                        break;
1589
1590                /*
1591                 * We allow "\ No newline at end of file". Depending
1592                 * on locale settings when the patch was produced we
1593                 * don't know what this line looks like. The only
1594                 * thing we do know is that it begins with "\ ".
1595                 * Checking for 12 is just for sanity check -- any
1596                 * l10n of "\ No newline..." is at least that long.
1597                 */
1598                case '\\':
1599                        if (len < 12 || memcmp(line, "\\ ", 2))
1600                                return -1;
1601                        break;
1602                }
1603        }
1604        if (oldlines || newlines)
1605                return -1;
1606        fragment->leading = leading;
1607        fragment->trailing = trailing;
1608
1609        /*
1610         * If a fragment ends with an incomplete line, we failed to include
1611         * it in the above loop because we hit oldlines == newlines == 0
1612         * before seeing it.
1613         */
1614        if (12 < size && !memcmp(line, "\\ ", 2))
1615                offset += linelen(line, size);
1616
1617        patch->lines_added += added;
1618        patch->lines_deleted += deleted;
1619
1620        if (0 < patch->is_new && oldlines)
1621                return error(_("new file depends on old contents"));
1622        if (0 < patch->is_delete && newlines)
1623                return error(_("deleted file still has contents"));
1624        return offset;
1625}
1626
1627/*
1628 * We have seen "diff --git a/... b/..." header (or a traditional patch
1629 * header).  Read hunks that belong to this patch into fragments and hang
1630 * them to the given patch structure.
1631 *
1632 * The (fragment->patch, fragment->size) pair points into the memory given
1633 * by the caller, not a copy, when we return.
1634 */
1635static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1636{
1637        unsigned long offset = 0;
1638        unsigned long oldlines = 0, newlines = 0, context = 0;
1639        struct fragment **fragp = &patch->fragments;
1640
1641        while (size > 4 && !memcmp(line, "@@ -", 4)) {
1642                struct fragment *fragment;
1643                int len;
1644
1645                fragment = xcalloc(1, sizeof(*fragment));
1646                fragment->linenr = linenr;
1647                len = parse_fragment(line, size, patch, fragment);
1648                if (len <= 0)
1649                        die(_("corrupt patch at line %d"), linenr);
1650                fragment->patch = line;
1651                fragment->size = len;
1652                oldlines += fragment->oldlines;
1653                newlines += fragment->newlines;
1654                context += fragment->leading + fragment->trailing;
1655
1656                *fragp = fragment;
1657                fragp = &fragment->next;
1658
1659                offset += len;
1660                line += len;
1661                size -= len;
1662        }
1663
1664        /*
1665         * If something was removed (i.e. we have old-lines) it cannot
1666         * be creation, and if something was added it cannot be
1667         * deletion.  However, the reverse is not true; --unified=0
1668         * patches that only add are not necessarily creation even
1669         * though they do not have any old lines, and ones that only
1670         * delete are not necessarily deletion.
1671         *
1672         * Unfortunately, a real creation/deletion patch do _not_ have
1673         * any context line by definition, so we cannot safely tell it
1674         * apart with --unified=0 insanity.  At least if the patch has
1675         * more than one hunk it is not creation or deletion.
1676         */
1677        if (patch->is_new < 0 &&
1678            (oldlines || (patch->fragments && patch->fragments->next)))
1679                patch->is_new = 0;
1680        if (patch->is_delete < 0 &&
1681            (newlines || (patch->fragments && patch->fragments->next)))
1682                patch->is_delete = 0;
1683
1684        if (0 < patch->is_new && oldlines)
1685                die(_("new file %s depends on old contents"), patch->new_name);
1686        if (0 < patch->is_delete && newlines)
1687                die(_("deleted file %s still has contents"), patch->old_name);
1688        if (!patch->is_delete && !newlines && context)
1689                fprintf_ln(stderr,
1690                           _("** warning: "
1691                             "file %s becomes empty but is not deleted"),
1692                           patch->new_name);
1693
1694        return offset;
1695}
1696
1697static inline int metadata_changes(struct patch *patch)
1698{
1699        return  patch->is_rename > 0 ||
1700                patch->is_copy > 0 ||
1701                patch->is_new > 0 ||
1702                patch->is_delete ||
1703                (patch->old_mode && patch->new_mode &&
1704                 patch->old_mode != patch->new_mode);
1705}
1706
1707static char *inflate_it(const void *data, unsigned long size,
1708                        unsigned long inflated_size)
1709{
1710        git_zstream stream;
1711        void *out;
1712        int st;
1713
1714        memset(&stream, 0, sizeof(stream));
1715
1716        stream.next_in = (unsigned char *)data;
1717        stream.avail_in = size;
1718        stream.next_out = out = xmalloc(inflated_size);
1719        stream.avail_out = inflated_size;
1720        git_inflate_init(&stream);
1721        st = git_inflate(&stream, Z_FINISH);
1722        git_inflate_end(&stream);
1723        if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1724                free(out);
1725                return NULL;
1726        }
1727        return out;
1728}
1729
1730/*
1731 * Read a binary hunk and return a new fragment; fragment->patch
1732 * points at an allocated memory that the caller must free, so
1733 * it is marked as "->free_patch = 1".
1734 */
1735static struct fragment *parse_binary_hunk(char **buf_p,
1736                                          unsigned long *sz_p,
1737                                          int *status_p,
1738                                          int *used_p)
1739{
1740        /*
1741         * Expect a line that begins with binary patch method ("literal"
1742         * or "delta"), followed by the length of data before deflating.
1743         * a sequence of 'length-byte' followed by base-85 encoded data
1744         * should follow, terminated by a newline.
1745         *
1746         * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1747         * and we would limit the patch line to 66 characters,
1748         * so one line can fit up to 13 groups that would decode
1749         * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1750         * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1751         */
1752        int llen, used;
1753        unsigned long size = *sz_p;
1754        char *buffer = *buf_p;
1755        int patch_method;
1756        unsigned long origlen;
1757        char *data = NULL;
1758        int hunk_size = 0;
1759        struct fragment *frag;
1760
1761        llen = linelen(buffer, size);
1762        used = llen;
1763
1764        *status_p = 0;
1765
1766        if (!prefixcmp(buffer, "delta ")) {
1767                patch_method = BINARY_DELTA_DEFLATED;
1768                origlen = strtoul(buffer + 6, NULL, 10);
1769        }
1770        else if (!prefixcmp(buffer, "literal ")) {
1771                patch_method = BINARY_LITERAL_DEFLATED;
1772                origlen = strtoul(buffer + 8, NULL, 10);
1773        }
1774        else
1775                return NULL;
1776
1777        linenr++;
1778        buffer += llen;
1779        while (1) {
1780                int byte_length, max_byte_length, newsize;
1781                llen = linelen(buffer, size);
1782                used += llen;
1783                linenr++;
1784                if (llen == 1) {
1785                        /* consume the blank line */
1786                        buffer++;
1787                        size--;
1788                        break;
1789                }
1790                /*
1791                 * Minimum line is "A00000\n" which is 7-byte long,
1792                 * and the line length must be multiple of 5 plus 2.
1793                 */
1794                if ((llen < 7) || (llen-2) % 5)
1795                        goto corrupt;
1796                max_byte_length = (llen - 2) / 5 * 4;
1797                byte_length = *buffer;
1798                if ('A' <= byte_length && byte_length <= 'Z')
1799                        byte_length = byte_length - 'A' + 1;
1800                else if ('a' <= byte_length && byte_length <= 'z')
1801                        byte_length = byte_length - 'a' + 27;
1802                else
1803                        goto corrupt;
1804                /* if the input length was not multiple of 4, we would
1805                 * have filler at the end but the filler should never
1806                 * exceed 3 bytes
1807                 */
1808                if (max_byte_length < byte_length ||
1809                    byte_length <= max_byte_length - 4)
1810                        goto corrupt;
1811                newsize = hunk_size + byte_length;
1812                data = xrealloc(data, newsize);
1813                if (decode_85(data + hunk_size, buffer + 1, byte_length))
1814                        goto corrupt;
1815                hunk_size = newsize;
1816                buffer += llen;
1817                size -= llen;
1818        }
1819
1820        frag = xcalloc(1, sizeof(*frag));
1821        frag->patch = inflate_it(data, hunk_size, origlen);
1822        frag->free_patch = 1;
1823        if (!frag->patch)
1824                goto corrupt;
1825        free(data);
1826        frag->size = origlen;
1827        *buf_p = buffer;
1828        *sz_p = size;
1829        *used_p = used;
1830        frag->binary_patch_method = patch_method;
1831        return frag;
1832
1833 corrupt:
1834        free(data);
1835        *status_p = -1;
1836        error(_("corrupt binary patch at line %d: %.*s"),
1837              linenr-1, llen-1, buffer);
1838        return NULL;
1839}
1840
1841static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1842{
1843        /*
1844         * We have read "GIT binary patch\n"; what follows is a line
1845         * that says the patch method (currently, either "literal" or
1846         * "delta") and the length of data before deflating; a
1847         * sequence of 'length-byte' followed by base-85 encoded data
1848         * follows.
1849         *
1850         * When a binary patch is reversible, there is another binary
1851         * hunk in the same format, starting with patch method (either
1852         * "literal" or "delta") with the length of data, and a sequence
1853         * of length-byte + base-85 encoded data, terminated with another
1854         * empty line.  This data, when applied to the postimage, produces
1855         * the preimage.
1856         */
1857        struct fragment *forward;
1858        struct fragment *reverse;
1859        int status;
1860        int used, used_1;
1861
1862        forward = parse_binary_hunk(&buffer, &size, &status, &used);
1863        if (!forward && !status)
1864                /* there has to be one hunk (forward hunk) */
1865                return error(_("unrecognized binary patch at line %d"), linenr-1);
1866        if (status)
1867                /* otherwise we already gave an error message */
1868                return status;
1869
1870        reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1871        if (reverse)
1872                used += used_1;
1873        else if (status) {
1874                /*
1875                 * Not having reverse hunk is not an error, but having
1876                 * a corrupt reverse hunk is.
1877                 */
1878                free((void*) forward->patch);
1879                free(forward);
1880                return status;
1881        }
1882        forward->next = reverse;
1883        patch->fragments = forward;
1884        patch->is_binary = 1;
1885        return used;
1886}
1887
1888/*
1889 * Read the patch text in "buffer" taht extends for "size" bytes; stop
1890 * reading after seeing a single patch (i.e. changes to a single file).
1891 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1892 * Return the number of bytes consumed, so that the caller can call us
1893 * again for the next patch.
1894 */
1895static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1896{
1897        int hdrsize, patchsize;
1898        int offset = find_header(buffer, size, &hdrsize, patch);
1899
1900        if (offset < 0)
1901                return offset;
1902
1903        patch->ws_rule = whitespace_rule(patch->new_name
1904                                         ? patch->new_name
1905                                         : patch->old_name);
1906
1907        patchsize = parse_single_patch(buffer + offset + hdrsize,
1908                                       size - offset - hdrsize, patch);
1909
1910        if (!patchsize) {
1911                static const char *binhdr[] = {
1912                        "Binary files ",
1913                        "Files ",
1914                        NULL,
1915                };
1916                static const char git_binary[] = "GIT binary patch\n";
1917                int i;
1918                int hd = hdrsize + offset;
1919                unsigned long llen = linelen(buffer + hd, size - hd);
1920
1921                if (llen == sizeof(git_binary) - 1 &&
1922                    !memcmp(git_binary, buffer + hd, llen)) {
1923                        int used;
1924                        linenr++;
1925                        used = parse_binary(buffer + hd + llen,
1926                                            size - hd - llen, patch);
1927                        if (used)
1928                                patchsize = used + llen;
1929                        else
1930                                patchsize = 0;
1931                }
1932                else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1933                        for (i = 0; binhdr[i]; i++) {
1934                                int len = strlen(binhdr[i]);
1935                                if (len < size - hd &&
1936                                    !memcmp(binhdr[i], buffer + hd, len)) {
1937                                        linenr++;
1938                                        patch->is_binary = 1;
1939                                        patchsize = llen;
1940                                        break;
1941                                }
1942                        }
1943                }
1944
1945                /* Empty patch cannot be applied if it is a text patch
1946                 * without metadata change.  A binary patch appears
1947                 * empty to us here.
1948                 */
1949                if ((apply || check) &&
1950                    (!patch->is_binary && !metadata_changes(patch)))
1951                        die(_("patch with only garbage at line %d"), linenr);
1952        }
1953
1954        return offset + hdrsize + patchsize;
1955}
1956
1957#define swap(a,b) myswap((a),(b),sizeof(a))
1958
1959#define myswap(a, b, size) do {         \
1960        unsigned char mytmp[size];      \
1961        memcpy(mytmp, &a, size);                \
1962        memcpy(&a, &b, size);           \
1963        memcpy(&b, mytmp, size);                \
1964} while (0)
1965
1966static void reverse_patches(struct patch *p)
1967{
1968        for (; p; p = p->next) {
1969                struct fragment *frag = p->fragments;
1970
1971                swap(p->new_name, p->old_name);
1972                swap(p->new_mode, p->old_mode);
1973                swap(p->is_new, p->is_delete);
1974                swap(p->lines_added, p->lines_deleted);
1975                swap(p->old_sha1_prefix, p->new_sha1_prefix);
1976
1977                for (; frag; frag = frag->next) {
1978                        swap(frag->newpos, frag->oldpos);
1979                        swap(frag->newlines, frag->oldlines);
1980                }
1981        }
1982}
1983
1984static const char pluses[] =
1985"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1986static const char minuses[]=
1987"----------------------------------------------------------------------";
1988
1989static void show_stats(struct patch *patch)
1990{
1991        struct strbuf qname = STRBUF_INIT;
1992        char *cp = patch->new_name ? patch->new_name : patch->old_name;
1993        int max, add, del;
1994
1995        quote_c_style(cp, &qname, NULL, 0);
1996
1997        /*
1998         * "scale" the filename
1999         */
2000        max = max_len;
2001        if (max > 50)
2002                max = 50;
2003
2004        if (qname.len > max) {
2005                cp = strchr(qname.buf + qname.len + 3 - max, '/');
2006                if (!cp)
2007                        cp = qname.buf + qname.len + 3 - max;
2008                strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2009        }
2010
2011        if (patch->is_binary) {
2012                printf(" %-*s |  Bin\n", max, qname.buf);
2013                strbuf_release(&qname);
2014                return;
2015        }
2016
2017        printf(" %-*s |", max, qname.buf);
2018        strbuf_release(&qname);
2019
2020        /*
2021         * scale the add/delete
2022         */
2023        max = max + max_change > 70 ? 70 - max : max_change;
2024        add = patch->lines_added;
2025        del = patch->lines_deleted;
2026
2027        if (max_change > 0) {
2028                int total = ((add + del) * max + max_change / 2) / max_change;
2029                add = (add * max + max_change / 2) / max_change;
2030                del = total - add;
2031        }
2032        printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2033                add, pluses, del, minuses);
2034}
2035
2036static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2037{
2038        switch (st->st_mode & S_IFMT) {
2039        case S_IFLNK:
2040                if (strbuf_readlink(buf, path, st->st_size) < 0)
2041                        return error(_("unable to read symlink %s"), path);
2042                return 0;
2043        case S_IFREG:
2044                if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2045                        return error(_("unable to open or read %s"), path);
2046                convert_to_git(path, buf->buf, buf->len, buf, 0);
2047                return 0;
2048        default:
2049                return -1;
2050        }
2051}
2052
2053/*
2054 * Update the preimage, and the common lines in postimage,
2055 * from buffer buf of length len. If postlen is 0 the postimage
2056 * is updated in place, otherwise it's updated on a new buffer
2057 * of length postlen
2058 */
2059
2060static void update_pre_post_images(struct image *preimage,
2061                                   struct image *postimage,
2062                                   char *buf,
2063                                   size_t len, size_t postlen)
2064{
2065        int i, ctx;
2066        char *new, *old, *fixed;
2067        struct image fixed_preimage;
2068
2069        /*
2070         * Update the preimage with whitespace fixes.  Note that we
2071         * are not losing preimage->buf -- apply_one_fragment() will
2072         * free "oldlines".
2073         */
2074        prepare_image(&fixed_preimage, buf, len, 1);
2075        assert(fixed_preimage.nr == preimage->nr);
2076        for (i = 0; i < preimage->nr; i++)
2077                fixed_preimage.line[i].flag = preimage->line[i].flag;
2078        free(preimage->line_allocated);
2079        *preimage = fixed_preimage;
2080
2081        /*
2082         * Adjust the common context lines in postimage. This can be
2083         * done in-place when we are just doing whitespace fixing,
2084         * which does not make the string grow, but needs a new buffer
2085         * when ignoring whitespace causes the update, since in this case
2086         * we could have e.g. tabs converted to multiple spaces.
2087         * We trust the caller to tell us if the update can be done
2088         * in place (postlen==0) or not.
2089         */
2090        old = postimage->buf;
2091        if (postlen)
2092                new = postimage->buf = xmalloc(postlen);
2093        else
2094                new = old;
2095        fixed = preimage->buf;
2096        for (i = ctx = 0; i < postimage->nr; i++) {
2097                size_t len = postimage->line[i].len;
2098                if (!(postimage->line[i].flag & LINE_COMMON)) {
2099                        /* an added line -- no counterparts in preimage */
2100                        memmove(new, old, len);
2101                        old += len;
2102                        new += len;
2103                        continue;
2104                }
2105
2106                /* a common context -- skip it in the original postimage */
2107                old += len;
2108
2109                /* and find the corresponding one in the fixed preimage */
2110                while (ctx < preimage->nr &&
2111                       !(preimage->line[ctx].flag & LINE_COMMON)) {
2112                        fixed += preimage->line[ctx].len;
2113                        ctx++;
2114                }
2115                if (preimage->nr <= ctx)
2116                        die(_("oops"));
2117
2118                /* and copy it in, while fixing the line length */
2119                len = preimage->line[ctx].len;
2120                memcpy(new, fixed, len);
2121                new += len;
2122                fixed += len;
2123                postimage->line[i].len = len;
2124                ctx++;
2125        }
2126
2127        /* Fix the length of the whole thing */
2128        postimage->len = new - postimage->buf;
2129}
2130
2131static int match_fragment(struct image *img,
2132                          struct image *preimage,
2133                          struct image *postimage,
2134                          unsigned long try,
2135                          int try_lno,
2136                          unsigned ws_rule,
2137                          int match_beginning, int match_end)
2138{
2139        int i;
2140        char *fixed_buf, *buf, *orig, *target;
2141        struct strbuf fixed;
2142        size_t fixed_len;
2143        int preimage_limit;
2144
2145        if (preimage->nr + try_lno <= img->nr) {
2146                /*
2147                 * The hunk falls within the boundaries of img.
2148                 */
2149                preimage_limit = preimage->nr;
2150                if (match_end && (preimage->nr + try_lno != img->nr))
2151                        return 0;
2152        } else if (ws_error_action == correct_ws_error &&
2153                   (ws_rule & WS_BLANK_AT_EOF)) {
2154                /*
2155                 * This hunk extends beyond the end of img, and we are
2156                 * removing blank lines at the end of the file.  This
2157                 * many lines from the beginning of the preimage must
2158                 * match with img, and the remainder of the preimage
2159                 * must be blank.
2160                 */
2161                preimage_limit = img->nr - try_lno;
2162        } else {
2163                /*
2164                 * The hunk extends beyond the end of the img and
2165                 * we are not removing blanks at the end, so we
2166                 * should reject the hunk at this position.
2167                 */
2168                return 0;
2169        }
2170
2171        if (match_beginning && try_lno)
2172                return 0;
2173
2174        /* Quick hash check */
2175        for (i = 0; i < preimage_limit; i++)
2176                if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2177                    (preimage->line[i].hash != img->line[try_lno + i].hash))
2178                        return 0;
2179
2180        if (preimage_limit == preimage->nr) {
2181                /*
2182                 * Do we have an exact match?  If we were told to match
2183                 * at the end, size must be exactly at try+fragsize,
2184                 * otherwise try+fragsize must be still within the preimage,
2185                 * and either case, the old piece should match the preimage
2186                 * exactly.
2187                 */
2188                if ((match_end
2189                     ? (try + preimage->len == img->len)
2190                     : (try + preimage->len <= img->len)) &&
2191                    !memcmp(img->buf + try, preimage->buf, preimage->len))
2192                        return 1;
2193        } else {
2194                /*
2195                 * The preimage extends beyond the end of img, so
2196                 * there cannot be an exact match.
2197                 *
2198                 * There must be one non-blank context line that match
2199                 * a line before the end of img.
2200                 */
2201                char *buf_end;
2202
2203                buf = preimage->buf;
2204                buf_end = buf;
2205                for (i = 0; i < preimage_limit; i++)
2206                        buf_end += preimage->line[i].len;
2207
2208                for ( ; buf < buf_end; buf++)
2209                        if (!isspace(*buf))
2210                                break;
2211                if (buf == buf_end)
2212                        return 0;
2213        }
2214
2215        /*
2216         * No exact match. If we are ignoring whitespace, run a line-by-line
2217         * fuzzy matching. We collect all the line length information because
2218         * we need it to adjust whitespace if we match.
2219         */
2220        if (ws_ignore_action == ignore_ws_change) {
2221                size_t imgoff = 0;
2222                size_t preoff = 0;
2223                size_t postlen = postimage->len;
2224                size_t extra_chars;
2225                char *preimage_eof;
2226                char *preimage_end;
2227                for (i = 0; i < preimage_limit; i++) {
2228                        size_t prelen = preimage->line[i].len;
2229                        size_t imglen = img->line[try_lno+i].len;
2230
2231                        if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2232                                              preimage->buf + preoff, prelen))
2233                                return 0;
2234                        if (preimage->line[i].flag & LINE_COMMON)
2235                                postlen += imglen - prelen;
2236                        imgoff += imglen;
2237                        preoff += prelen;
2238                }
2239
2240                /*
2241                 * Ok, the preimage matches with whitespace fuzz.
2242                 *
2243                 * imgoff now holds the true length of the target that
2244                 * matches the preimage before the end of the file.
2245                 *
2246                 * Count the number of characters in the preimage that fall
2247                 * beyond the end of the file and make sure that all of them
2248                 * are whitespace characters. (This can only happen if
2249                 * we are removing blank lines at the end of the file.)
2250                 */
2251                buf = preimage_eof = preimage->buf + preoff;
2252                for ( ; i < preimage->nr; i++)
2253                        preoff += preimage->line[i].len;
2254                preimage_end = preimage->buf + preoff;
2255                for ( ; buf < preimage_end; buf++)
2256                        if (!isspace(*buf))
2257                                return 0;
2258
2259                /*
2260                 * Update the preimage and the common postimage context
2261                 * lines to use the same whitespace as the target.
2262                 * If whitespace is missing in the target (i.e.
2263                 * if the preimage extends beyond the end of the file),
2264                 * use the whitespace from the preimage.
2265                 */
2266                extra_chars = preimage_end - preimage_eof;
2267                strbuf_init(&fixed, imgoff + extra_chars);
2268                strbuf_add(&fixed, img->buf + try, imgoff);
2269                strbuf_add(&fixed, preimage_eof, extra_chars);
2270                fixed_buf = strbuf_detach(&fixed, &fixed_len);
2271                update_pre_post_images(preimage, postimage,
2272                                fixed_buf, fixed_len, postlen);
2273                return 1;
2274        }
2275
2276        if (ws_error_action != correct_ws_error)
2277                return 0;
2278
2279        /*
2280         * The hunk does not apply byte-by-byte, but the hash says
2281         * it might with whitespace fuzz. We haven't been asked to
2282         * ignore whitespace, we were asked to correct whitespace
2283         * errors, so let's try matching after whitespace correction.
2284         *
2285         * The preimage may extend beyond the end of the file,
2286         * but in this loop we will only handle the part of the
2287         * preimage that falls within the file.
2288         */
2289        strbuf_init(&fixed, preimage->len + 1);
2290        orig = preimage->buf;
2291        target = img->buf + try;
2292        for (i = 0; i < preimage_limit; i++) {
2293                size_t oldlen = preimage->line[i].len;
2294                size_t tgtlen = img->line[try_lno + i].len;
2295                size_t fixstart = fixed.len;
2296                struct strbuf tgtfix;
2297                int match;
2298
2299                /* Try fixing the line in the preimage */
2300                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2301
2302                /* Try fixing the line in the target */
2303                strbuf_init(&tgtfix, tgtlen);
2304                ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2305
2306                /*
2307                 * If they match, either the preimage was based on
2308                 * a version before our tree fixed whitespace breakage,
2309                 * or we are lacking a whitespace-fix patch the tree
2310                 * the preimage was based on already had (i.e. target
2311                 * has whitespace breakage, the preimage doesn't).
2312                 * In either case, we are fixing the whitespace breakages
2313                 * so we might as well take the fix together with their
2314                 * real change.
2315                 */
2316                match = (tgtfix.len == fixed.len - fixstart &&
2317                         !memcmp(tgtfix.buf, fixed.buf + fixstart,
2318                                             fixed.len - fixstart));
2319
2320                strbuf_release(&tgtfix);
2321                if (!match)
2322                        goto unmatch_exit;
2323
2324                orig += oldlen;
2325                target += tgtlen;
2326        }
2327
2328
2329        /*
2330         * Now handle the lines in the preimage that falls beyond the
2331         * end of the file (if any). They will only match if they are
2332         * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2333         * false).
2334         */
2335        for ( ; i < preimage->nr; i++) {
2336                size_t fixstart = fixed.len; /* start of the fixed preimage */
2337                size_t oldlen = preimage->line[i].len;
2338                int j;
2339
2340                /* Try fixing the line in the preimage */
2341                ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2342
2343                for (j = fixstart; j < fixed.len; j++)
2344                        if (!isspace(fixed.buf[j]))
2345                                goto unmatch_exit;
2346
2347                orig += oldlen;
2348        }
2349
2350        /*
2351         * Yes, the preimage is based on an older version that still
2352         * has whitespace breakages unfixed, and fixing them makes the
2353         * hunk match.  Update the context lines in the postimage.
2354         */
2355        fixed_buf = strbuf_detach(&fixed, &fixed_len);
2356        update_pre_post_images(preimage, postimage,
2357                               fixed_buf, fixed_len, 0);
2358        return 1;
2359
2360 unmatch_exit:
2361        strbuf_release(&fixed);
2362        return 0;
2363}
2364
2365static int find_pos(struct image *img,
2366                    struct image *preimage,
2367                    struct image *postimage,
2368                    int line,
2369                    unsigned ws_rule,
2370                    int match_beginning, int match_end)
2371{
2372        int i;
2373        unsigned long backwards, forwards, try;
2374        int backwards_lno, forwards_lno, try_lno;
2375
2376        /*
2377         * If match_beginning or match_end is specified, there is no
2378         * point starting from a wrong line that will never match and
2379         * wander around and wait for a match at the specified end.
2380         */
2381        if (match_beginning)
2382                line = 0;
2383        else if (match_end)
2384                line = img->nr - preimage->nr;
2385
2386        /*
2387         * Because the comparison is unsigned, the following test
2388         * will also take care of a negative line number that can
2389         * result when match_end and preimage is larger than the target.
2390         */
2391        if ((size_t) line > img->nr)
2392                line = img->nr;
2393
2394        try = 0;
2395        for (i = 0; i < line; i++)
2396                try += img->line[i].len;
2397
2398        /*
2399         * There's probably some smart way to do this, but I'll leave
2400         * that to the smart and beautiful people. I'm simple and stupid.
2401         */
2402        backwards = try;
2403        backwards_lno = line;
2404        forwards = try;
2405        forwards_lno = line;
2406        try_lno = line;
2407
2408        for (i = 0; ; i++) {
2409                if (match_fragment(img, preimage, postimage,
2410                                   try, try_lno, ws_rule,
2411                                   match_beginning, match_end))
2412                        return try_lno;
2413
2414        again:
2415                if (backwards_lno == 0 && forwards_lno == img->nr)
2416                        break;
2417
2418                if (i & 1) {
2419                        if (backwards_lno == 0) {
2420                                i++;
2421                                goto again;
2422                        }
2423                        backwards_lno--;
2424                        backwards -= img->line[backwards_lno].len;
2425                        try = backwards;
2426                        try_lno = backwards_lno;
2427                } else {
2428                        if (forwards_lno == img->nr) {
2429                                i++;
2430                                goto again;
2431                        }
2432                        forwards += img->line[forwards_lno].len;
2433                        forwards_lno++;
2434                        try = forwards;
2435                        try_lno = forwards_lno;
2436                }
2437
2438        }
2439        return -1;
2440}
2441
2442static void remove_first_line(struct image *img)
2443{
2444        img->buf += img->line[0].len;
2445        img->len -= img->line[0].len;
2446        img->line++;
2447        img->nr--;
2448}
2449
2450static void remove_last_line(struct image *img)
2451{
2452        img->len -= img->line[--img->nr].len;
2453}
2454
2455/*
2456 * The change from "preimage" and "postimage" has been found to
2457 * apply at applied_pos (counts in line numbers) in "img".
2458 * Update "img" to remove "preimage" and replace it with "postimage".
2459 */
2460static void update_image(struct image *img,
2461                         int applied_pos,
2462                         struct image *preimage,
2463                         struct image *postimage)
2464{
2465        /*
2466         * remove the copy of preimage at offset in img
2467         * and replace it with postimage
2468         */
2469        int i, nr;
2470        size_t remove_count, insert_count, applied_at = 0;
2471        char *result;
2472        int preimage_limit;
2473
2474        /*
2475         * If we are removing blank lines at the end of img,
2476         * the preimage may extend beyond the end.
2477         * If that is the case, we must be careful only to
2478         * remove the part of the preimage that falls within
2479         * the boundaries of img. Initialize preimage_limit
2480         * to the number of lines in the preimage that falls
2481         * within the boundaries.
2482         */
2483        preimage_limit = preimage->nr;
2484        if (preimage_limit > img->nr - applied_pos)
2485                preimage_limit = img->nr - applied_pos;
2486
2487        for (i = 0; i < applied_pos; i++)
2488                applied_at += img->line[i].len;
2489
2490        remove_count = 0;
2491        for (i = 0; i < preimage_limit; i++)
2492                remove_count += img->line[applied_pos + i].len;
2493        insert_count = postimage->len;
2494
2495        /* Adjust the contents */
2496        result = xmalloc(img->len + insert_count - remove_count + 1);
2497        memcpy(result, img->buf, applied_at);
2498        memcpy(result + applied_at, postimage->buf, postimage->len);
2499        memcpy(result + applied_at + postimage->len,
2500               img->buf + (applied_at + remove_count),
2501               img->len - (applied_at + remove_count));
2502        free(img->buf);
2503        img->buf = result;
2504        img->len += insert_count - remove_count;
2505        result[img->len] = '\0';
2506
2507        /* Adjust the line table */
2508        nr = img->nr + postimage->nr - preimage_limit;
2509        if (preimage_limit < postimage->nr) {
2510                /*
2511                 * NOTE: this knows that we never call remove_first_line()
2512                 * on anything other than pre/post image.
2513                 */
2514                img->line = xrealloc(img->line, nr * sizeof(*img->line));
2515                img->line_allocated = img->line;
2516        }
2517        if (preimage_limit != postimage->nr)
2518                memmove(img->line + applied_pos + postimage->nr,
2519                        img->line + applied_pos + preimage_limit,
2520                        (img->nr - (applied_pos + preimage_limit)) *
2521                        sizeof(*img->line));
2522        memcpy(img->line + applied_pos,
2523               postimage->line,
2524               postimage->nr * sizeof(*img->line));
2525        if (!allow_overlap)
2526                for (i = 0; i < postimage->nr; i++)
2527                        img->line[applied_pos + i].flag |= LINE_PATCHED;
2528        img->nr = nr;
2529}
2530
2531/*
2532 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2533 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
2534 * replace the part of "img" with "postimage" text.
2535 */
2536static int apply_one_fragment(struct image *img, struct fragment *frag,
2537                              int inaccurate_eof, unsigned ws_rule,
2538                              int nth_fragment)
2539{
2540        int match_beginning, match_end;
2541        const char *patch = frag->patch;
2542        int size = frag->size;
2543        char *old, *oldlines;
2544        struct strbuf newlines;
2545        int new_blank_lines_at_end = 0;
2546        int found_new_blank_lines_at_end = 0;
2547        int hunk_linenr = frag->linenr;
2548        unsigned long leading, trailing;
2549        int pos, applied_pos;
2550        struct image preimage;
2551        struct image postimage;
2552
2553        memset(&preimage, 0, sizeof(preimage));
2554        memset(&postimage, 0, sizeof(postimage));
2555        oldlines = xmalloc(size);
2556        strbuf_init(&newlines, size);
2557
2558        old = oldlines;
2559        while (size > 0) {
2560                char first;
2561                int len = linelen(patch, size);
2562                int plen;
2563                int added_blank_line = 0;
2564                int is_blank_context = 0;
2565                size_t start;
2566
2567                if (!len)
2568                        break;
2569
2570                /*
2571                 * "plen" is how much of the line we should use for
2572                 * the actual patch data. Normally we just remove the
2573                 * first character on the line, but if the line is
2574                 * followed by "\ No newline", then we also remove the
2575                 * last one (which is the newline, of course).
2576                 */
2577                plen = len - 1;
2578                if (len < size && patch[len] == '\\')
2579                        plen--;
2580                first = *patch;
2581                if (apply_in_reverse) {
2582                        if (first == '-')
2583                                first = '+';
2584                        else if (first == '+')
2585                                first = '-';
2586                }
2587
2588                switch (first) {
2589                case '\n':
2590                        /* Newer GNU diff, empty context line */
2591                        if (plen < 0)
2592                                /* ... followed by '\No newline'; nothing */
2593                                break;
2594                        *old++ = '\n';
2595                        strbuf_addch(&newlines, '\n');
2596                        add_line_info(&preimage, "\n", 1, LINE_COMMON);
2597                        add_line_info(&postimage, "\n", 1, LINE_COMMON);
2598                        is_blank_context = 1;
2599                        break;
2600                case ' ':
2601                        if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2602                            ws_blank_line(patch + 1, plen, ws_rule))
2603                                is_blank_context = 1;
2604                case '-':
2605                        memcpy(old, patch + 1, plen);
2606                        add_line_info(&preimage, old, plen,
2607                                      (first == ' ' ? LINE_COMMON : 0));
2608                        old += plen;
2609                        if (first == '-')
2610                                break;
2611                /* Fall-through for ' ' */
2612                case '+':
2613                        /* --no-add does not add new lines */
2614                        if (first == '+' && no_add)
2615                                break;
2616
2617                        start = newlines.len;
2618                        if (first != '+' ||
2619                            !whitespace_error ||
2620                            ws_error_action != correct_ws_error) {
2621                                strbuf_add(&newlines, patch + 1, plen);
2622                        }
2623                        else {
2624                                ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2625                        }
2626                        add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2627                                      (first == '+' ? 0 : LINE_COMMON));
2628                        if (first == '+' &&
2629                            (ws_rule & WS_BLANK_AT_EOF) &&
2630                            ws_blank_line(patch + 1, plen, ws_rule))
2631                                added_blank_line = 1;
2632                        break;
2633                case '@': case '\\':
2634                        /* Ignore it, we already handled it */
2635                        break;
2636                default:
2637                        if (apply_verbosely)
2638                                error(_("invalid start of line: '%c'"), first);
2639                        return -1;
2640                }
2641                if (added_blank_line) {
2642                        if (!new_blank_lines_at_end)
2643                                found_new_blank_lines_at_end = hunk_linenr;
2644                        new_blank_lines_at_end++;
2645                }
2646                else if (is_blank_context)
2647                        ;
2648                else
2649                        new_blank_lines_at_end = 0;
2650                patch += len;
2651                size -= len;
2652                hunk_linenr++;
2653        }
2654        if (inaccurate_eof &&
2655            old > oldlines && old[-1] == '\n' &&
2656            newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2657                old--;
2658                strbuf_setlen(&newlines, newlines.len - 1);
2659        }
2660
2661        leading = frag->leading;
2662        trailing = frag->trailing;
2663
2664        /*
2665         * A hunk to change lines at the beginning would begin with
2666         * @@ -1,L +N,M @@
2667         * but we need to be careful.  -U0 that inserts before the second
2668         * line also has this pattern.
2669         *
2670         * And a hunk to add to an empty file would begin with
2671         * @@ -0,0 +N,M @@
2672         *
2673         * In other words, a hunk that is (frag->oldpos <= 1) with or
2674         * without leading context must match at the beginning.
2675         */
2676        match_beginning = (!frag->oldpos ||
2677                           (frag->oldpos == 1 && !unidiff_zero));
2678
2679        /*
2680         * A hunk without trailing lines must match at the end.
2681         * However, we simply cannot tell if a hunk must match end
2682         * from the lack of trailing lines if the patch was generated
2683         * with unidiff without any context.
2684         */
2685        match_end = !unidiff_zero && !trailing;
2686
2687        pos = frag->newpos ? (frag->newpos - 1) : 0;
2688        preimage.buf = oldlines;
2689        preimage.len = old - oldlines;
2690        postimage.buf = newlines.buf;
2691        postimage.len = newlines.len;
2692        preimage.line = preimage.line_allocated;
2693        postimage.line = postimage.line_allocated;
2694
2695        for (;;) {
2696
2697                applied_pos = find_pos(img, &preimage, &postimage, pos,
2698                                       ws_rule, match_beginning, match_end);
2699
2700                if (applied_pos >= 0)
2701                        break;
2702
2703                /* Am I at my context limits? */
2704                if ((leading <= p_context) && (trailing <= p_context))
2705                        break;
2706                if (match_beginning || match_end) {
2707                        match_beginning = match_end = 0;
2708                        continue;
2709                }
2710
2711                /*
2712                 * Reduce the number of context lines; reduce both
2713                 * leading and trailing if they are equal otherwise
2714                 * just reduce the larger context.
2715                 */
2716                if (leading >= trailing) {
2717                        remove_first_line(&preimage);
2718                        remove_first_line(&postimage);
2719                        pos--;
2720                        leading--;
2721                }
2722                if (trailing > leading) {
2723                        remove_last_line(&preimage);
2724                        remove_last_line(&postimage);
2725                        trailing--;
2726                }
2727        }
2728
2729        if (applied_pos >= 0) {
2730                if (new_blank_lines_at_end &&
2731                    preimage.nr + applied_pos >= img->nr &&
2732                    (ws_rule & WS_BLANK_AT_EOF) &&
2733                    ws_error_action != nowarn_ws_error) {
2734                        record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2735                                        found_new_blank_lines_at_end);
2736                        if (ws_error_action == correct_ws_error) {
2737                                while (new_blank_lines_at_end--)
2738                                        remove_last_line(&postimage);
2739                        }
2740                        /*
2741                         * We would want to prevent write_out_results()
2742                         * from taking place in apply_patch() that follows
2743                         * the callchain led us here, which is:
2744                         * apply_patch->check_patch_list->check_patch->
2745                         * apply_data->apply_fragments->apply_one_fragment
2746                         */
2747                        if (ws_error_action == die_on_ws_error)
2748                                apply = 0;
2749                }
2750
2751                if (apply_verbosely && applied_pos != pos) {
2752                        int offset = applied_pos - pos;
2753                        if (apply_in_reverse)
2754                                offset = 0 - offset;
2755                        fprintf_ln(stderr,
2756                                   Q_("Hunk #%d succeeded at %d (offset %d line).",
2757                                      "Hunk #%d succeeded at %d (offset %d lines).",
2758                                      offset),
2759                                   nth_fragment, applied_pos + 1, offset);
2760                }
2761
2762                /*
2763                 * Warn if it was necessary to reduce the number
2764                 * of context lines.
2765                 */
2766                if ((leading != frag->leading) ||
2767                    (trailing != frag->trailing))
2768                        fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2769                                             " to apply fragment at %d"),
2770                                   leading, trailing, applied_pos+1);
2771                update_image(img, applied_pos, &preimage, &postimage);
2772        } else {
2773                if (apply_verbosely)
2774                        error(_("while searching for:\n%.*s"),
2775                              (int)(old - oldlines), oldlines);
2776        }
2777
2778        free(oldlines);
2779        strbuf_release(&newlines);
2780        free(preimage.line_allocated);
2781        free(postimage.line_allocated);
2782
2783        return (applied_pos < 0);
2784}
2785
2786static int apply_binary_fragment(struct image *img, struct patch *patch)
2787{
2788        struct fragment *fragment = patch->fragments;
2789        unsigned long len;
2790        void *dst;
2791
2792        if (!fragment)
2793                return error(_("missing binary patch data for '%s'"),
2794                             patch->new_name ?
2795                             patch->new_name :
2796                             patch->old_name);
2797
2798        /* Binary patch is irreversible without the optional second hunk */
2799        if (apply_in_reverse) {
2800                if (!fragment->next)
2801                        return error("cannot reverse-apply a binary patch "
2802                                     "without the reverse hunk to '%s'",
2803                                     patch->new_name
2804                                     ? patch->new_name : patch->old_name);
2805                fragment = fragment->next;
2806        }
2807        switch (fragment->binary_patch_method) {
2808        case BINARY_DELTA_DEFLATED:
2809                dst = patch_delta(img->buf, img->len, fragment->patch,
2810                                  fragment->size, &len);
2811                if (!dst)
2812                        return -1;
2813                clear_image(img);
2814                img->buf = dst;
2815                img->len = len;
2816                return 0;
2817        case BINARY_LITERAL_DEFLATED:
2818                clear_image(img);
2819                img->len = fragment->size;
2820                img->buf = xmalloc(img->len+1);
2821                memcpy(img->buf, fragment->patch, img->len);
2822                img->buf[img->len] = '\0';
2823                return 0;
2824        }
2825        return -1;
2826}
2827
2828/*
2829 * Replace "img" with the result of applying the binary patch.
2830 * The binary patch data itself in patch->fragment is still kept
2831 * but the preimage prepared by the caller in "img" is freed here
2832 * or in the helper function apply_binary_fragment() this calls.
2833 */
2834static int apply_binary(struct image *img, struct patch *patch)
2835{
2836        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2837        unsigned char sha1[20];
2838
2839        /*
2840         * For safety, we require patch index line to contain
2841         * full 40-byte textual SHA1 for old and new, at least for now.
2842         */
2843        if (strlen(patch->old_sha1_prefix) != 40 ||
2844            strlen(patch->new_sha1_prefix) != 40 ||
2845            get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2846            get_sha1_hex(patch->new_sha1_prefix, sha1))
2847                return error("cannot apply binary patch to '%s' "
2848                             "without full index line", name);
2849
2850        if (patch->old_name) {
2851                /*
2852                 * See if the old one matches what the patch
2853                 * applies to.
2854                 */
2855                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2856                if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2857                        return error("the patch applies to '%s' (%s), "
2858                                     "which does not match the "
2859                                     "current contents.",
2860                                     name, sha1_to_hex(sha1));
2861        }
2862        else {
2863                /* Otherwise, the old one must be empty. */
2864                if (img->len)
2865                        return error("the patch applies to an empty "
2866                                     "'%s' but it is not empty", name);
2867        }
2868
2869        get_sha1_hex(patch->new_sha1_prefix, sha1);
2870        if (is_null_sha1(sha1)) {
2871                clear_image(img);
2872                return 0; /* deletion patch */
2873        }
2874
2875        if (has_sha1_file(sha1)) {
2876                /* We already have the postimage */
2877                enum object_type type;
2878                unsigned long size;
2879                char *result;
2880
2881                result = read_sha1_file(sha1, &type, &size);
2882                if (!result)
2883                        return error("the necessary postimage %s for "
2884                                     "'%s' cannot be read",
2885                                     patch->new_sha1_prefix, name);
2886                clear_image(img);
2887                img->buf = result;
2888                img->len = size;
2889        } else {
2890                /*
2891                 * We have verified buf matches the preimage;
2892                 * apply the patch data to it, which is stored
2893                 * in the patch->fragments->{patch,size}.
2894                 */
2895                if (apply_binary_fragment(img, patch))
2896                        return error(_("binary patch does not apply to '%s'"),
2897                                     name);
2898
2899                /* verify that the result matches */
2900                hash_sha1_file(img->buf, img->len, blob_type, sha1);
2901                if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2902                        return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2903                                name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2904        }
2905
2906        return 0;
2907}
2908
2909static int apply_fragments(struct image *img, struct patch *patch)
2910{
2911        struct fragment *frag = patch->fragments;
2912        const char *name = patch->old_name ? patch->old_name : patch->new_name;
2913        unsigned ws_rule = patch->ws_rule;
2914        unsigned inaccurate_eof = patch->inaccurate_eof;
2915        int nth = 0;
2916
2917        if (patch->is_binary)
2918                return apply_binary(img, patch);
2919
2920        while (frag) {
2921                nth++;
2922                if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2923                        error(_("patch failed: %s:%ld"), name, frag->oldpos);
2924                        if (!apply_with_reject)
2925                                return -1;
2926                        frag->rejected = 1;
2927                }
2928                frag = frag->next;
2929        }
2930        return 0;
2931}
2932
2933static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
2934{
2935        if (S_ISGITLINK(mode)) {
2936                strbuf_grow(buf, 100);
2937                strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
2938        } else {
2939                enum object_type type;
2940                unsigned long sz;
2941                char *result;
2942
2943                result = read_sha1_file(sha1, &type, &sz);
2944                if (!result)
2945                        return -1;
2946                /* XXX read_sha1_file NUL-terminates */
2947                strbuf_attach(buf, result, sz, sz + 1);
2948        }
2949        return 0;
2950}
2951
2952static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2953{
2954        if (!ce)
2955                return 0;
2956        return read_blob_object(buf, ce->sha1, ce->ce_mode);
2957}
2958
2959static struct patch *in_fn_table(const char *name)
2960{
2961        struct string_list_item *item;
2962
2963        if (name == NULL)
2964                return NULL;
2965
2966        item = string_list_lookup(&fn_table, name);
2967        if (item != NULL)
2968                return (struct patch *)item->util;
2969
2970        return NULL;
2971}
2972
2973/*
2974 * item->util in the filename table records the status of the path.
2975 * Usually it points at a patch (whose result records the contents
2976 * of it after applying it), but it could be PATH_WAS_DELETED for a
2977 * path that a previously applied patch has already removed, or
2978 * PATH_TO_BE_DELETED for a path that a later patch would remove.
2979 *
2980 * The latter is needed to deal with a case where two paths A and B
2981 * are swapped by first renaming A to B and then renaming B to A;
2982 * moving A to B should not be prevented due to presense of B as we
2983 * will remove it in a later patch.
2984 */
2985#define PATH_TO_BE_DELETED ((struct patch *) -2)
2986#define PATH_WAS_DELETED ((struct patch *) -1)
2987
2988static int to_be_deleted(struct patch *patch)
2989{
2990        return patch == PATH_TO_BE_DELETED;
2991}
2992
2993static int was_deleted(struct patch *patch)
2994{
2995        return patch == PATH_WAS_DELETED;
2996}
2997
2998static void add_to_fn_table(struct patch *patch)
2999{
3000        struct string_list_item *item;
3001
3002        /*
3003         * Always add new_name unless patch is a deletion
3004         * This should cover the cases for normal diffs,
3005         * file creations and copies
3006         */
3007        if (patch->new_name != NULL) {
3008                item = string_list_insert(&fn_table, patch->new_name);
3009                item->util = patch;
3010        }
3011
3012        /*
3013         * store a failure on rename/deletion cases because
3014         * later chunks shouldn't patch old names
3015         */
3016        if ((patch->new_name == NULL) || (patch->is_rename)) {
3017                item = string_list_insert(&fn_table, patch->old_name);
3018                item->util = PATH_WAS_DELETED;
3019        }
3020}
3021
3022static void prepare_fn_table(struct patch *patch)
3023{
3024        /*
3025         * store information about incoming file deletion
3026         */
3027        while (patch) {
3028                if ((patch->new_name == NULL) || (patch->is_rename)) {
3029                        struct string_list_item *item;
3030                        item = string_list_insert(&fn_table, patch->old_name);
3031                        item->util = PATH_TO_BE_DELETED;
3032                }
3033                patch = patch->next;
3034        }
3035}
3036
3037static int checkout_target(struct cache_entry *ce, struct stat *st)
3038{
3039        struct checkout costate;
3040
3041        memset(&costate, 0, sizeof(costate));
3042        costate.base_dir = "";
3043        costate.refresh_cache = 1;
3044        if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3045                return error(_("cannot checkout %s"), ce->name);
3046        return 0;
3047}
3048
3049static struct patch *previous_patch(struct patch *patch, int *gone)
3050{
3051        struct patch *previous;
3052
3053        *gone = 0;
3054        if (patch->is_copy || patch->is_rename)
3055                return NULL; /* "git" patches do not depend on the order */
3056
3057        previous = in_fn_table(patch->old_name);
3058        if (!previous)
3059                return NULL;
3060
3061        if (to_be_deleted(previous))
3062                return NULL; /* the deletion hasn't happened yet */
3063
3064        if (was_deleted(previous))
3065                *gone = 1;
3066
3067        return previous;
3068}
3069
3070#define SUBMODULE_PATCH_WITHOUT_INDEX 1
3071
3072static int load_patch_target(struct strbuf *buf,
3073                             struct cache_entry *ce,
3074                             struct stat *st,
3075                             const char *name,
3076                             unsigned expected_mode)
3077{
3078        if (cached) {
3079                if (read_file_or_gitlink(ce, buf))
3080                        return error(_("read of %s failed"), name);
3081        } else if (name) {
3082                if (S_ISGITLINK(expected_mode)) {
3083                        if (ce)
3084                                return read_file_or_gitlink(ce, buf);
3085                        else
3086                                return SUBMODULE_PATCH_WITHOUT_INDEX;
3087                } else {
3088                        if (read_old_data(st, name, buf))
3089                                return error(_("read of %s failed"), name);
3090                }
3091        }
3092        return 0;
3093}
3094
3095/*
3096 * We are about to apply "patch"; populate the "image" with the
3097 * current version we have, from the working tree or from the index,
3098 * depending on the situation e.g. --cached/--index.  If we are
3099 * applying a non-git patch that incrementally updates the tree,
3100 * we read from the result of a previous diff.
3101 */
3102static int load_preimage(struct image *image,
3103                         struct patch *patch, struct stat *st, struct cache_entry *ce)
3104{
3105        struct strbuf buf = STRBUF_INIT;
3106        size_t len;
3107        char *img;
3108        struct patch *previous;
3109        int status;
3110
3111        previous = previous_patch(patch, &status);
3112        if (status)
3113                return error(_("path %s has been renamed/deleted"),
3114                             patch->old_name);
3115        if (previous) {
3116                /* We have a patched copy in memory; use that. */
3117                strbuf_add(&buf, previous->result, previous->resultsize);
3118        } else {
3119                status = load_patch_target(&buf, ce, st,
3120                                           patch->old_name, patch->old_mode);
3121                if (status < 0)
3122                        return status;
3123                else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3124                        /*
3125                         * There is no way to apply subproject
3126                         * patch without looking at the index.
3127                         * NEEDSWORK: shouldn't this be flagged
3128                         * as an error???
3129                         */
3130                        free_fragment_list(patch->fragments);
3131                        patch->fragments = NULL;
3132                } else if (status) {
3133                        return error(_("read of %s failed"), patch->old_name);
3134                }
3135        }
3136
3137        img = strbuf_detach(&buf, &len);
3138        prepare_image(image, img, len, !patch->is_binary);
3139        return 0;
3140}
3141
3142static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
3143{
3144        struct image image;
3145
3146        if (load_preimage(&image, patch, st, ce) < 0)
3147                return -1;
3148
3149        if (apply_fragments(&image, patch) < 0)
3150                return -1; /* note with --reject this succeeds. */
3151        patch->result = image.buf;
3152        patch->resultsize = image.len;
3153        add_to_fn_table(patch);
3154        free(image.line_allocated);
3155
3156        if (0 < patch->is_delete && patch->resultsize)
3157                return error(_("removal patch leaves file contents"));
3158
3159        return 0;
3160}
3161
3162static int verify_index_match(struct cache_entry *ce, struct stat *st)
3163{
3164        if (S_ISGITLINK(ce->ce_mode)) {
3165                if (!S_ISDIR(st->st_mode))
3166                        return -1;
3167                return 0;
3168        }
3169        return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3170}
3171
3172/*
3173 * If "patch" that we are looking at modifies or deletes what we have,
3174 * we would want it not to lose any local modification we have, either
3175 * in the working tree or in the index.
3176 *
3177 * This also decides if a non-git patch is a creation patch or a
3178 * modification to an existing empty file.  We do not check the state
3179 * of the current tree for a creation patch in this function; the caller
3180 * check_patch() separately makes sure (and errors out otherwise) that
3181 * the path the patch creates does not exist in the current tree.
3182 */
3183static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3184{
3185        const char *old_name = patch->old_name;
3186        struct patch *previous = NULL;
3187        int stat_ret = 0, status;
3188        unsigned st_mode = 0;
3189
3190        if (!old_name)
3191                return 0;
3192
3193        assert(patch->is_new <= 0);
3194        previous = previous_patch(patch, &status);
3195
3196        if (status)
3197                return error(_("path %s has been renamed/deleted"), old_name);
3198        if (previous) {
3199                st_mode = previous->new_mode;
3200        } else if (!cached) {
3201                stat_ret = lstat(old_name, st);
3202                if (stat_ret && errno != ENOENT)
3203                        return error(_("%s: %s"), old_name, strerror(errno));
3204        }
3205
3206        if (check_index && !previous) {
3207                int pos = cache_name_pos(old_name, strlen(old_name));
3208                if (pos < 0) {
3209                        if (patch->is_new < 0)
3210                                goto is_new;
3211                        return error(_("%s: does not exist in index"), old_name);
3212                }
3213                *ce = active_cache[pos];
3214                if (stat_ret < 0) {
3215                        if (checkout_target(*ce, st))
3216                                return -1;
3217                }
3218                if (!cached && verify_index_match(*ce, st))
3219                        return error(_("%s: does not match index"), old_name);
3220                if (cached)
3221                        st_mode = (*ce)->ce_mode;
3222        } else if (stat_ret < 0) {
3223                if (patch->is_new < 0)
3224                        goto is_new;
3225                return error(_("%s: %s"), old_name, strerror(errno));
3226        }
3227
3228        if (!cached && !previous)
3229                st_mode = ce_mode_from_stat(*ce, st->st_mode);
3230
3231        if (patch->is_new < 0)
3232                patch->is_new = 0;
3233        if (!patch->old_mode)
3234                patch->old_mode = st_mode;
3235        if ((st_mode ^ patch->old_mode) & S_IFMT)
3236                return error(_("%s: wrong type"), old_name);
3237        if (st_mode != patch->old_mode)
3238                warning(_("%s has type %o, expected %o"),
3239                        old_name, st_mode, patch->old_mode);
3240        if (!patch->new_mode && !patch->is_delete)
3241                patch->new_mode = st_mode;
3242        return 0;
3243
3244 is_new:
3245        patch->is_new = 1;
3246        patch->is_delete = 0;
3247        free(patch->old_name);
3248        patch->old_name = NULL;
3249        return 0;
3250}
3251
3252static int check_to_create_blob(const char *new_name, int ok_if_exists)
3253{
3254        struct stat nst;
3255        if (!lstat(new_name, &nst)) {
3256                if (S_ISDIR(nst.st_mode) || ok_if_exists)
3257                        return 0;
3258                /*
3259                 * A leading component of new_name might be a symlink
3260                 * that is going to be removed with this patch, but
3261                 * still pointing at somewhere that has the path.
3262                 * In such a case, path "new_name" does not exist as
3263                 * far as git is concerned.
3264                 */
3265                if (has_symlink_leading_path(new_name, strlen(new_name)))
3266                        return 0;
3267
3268                return error(_("%s: already exists in working directory"), new_name);
3269        }
3270        else if ((errno != ENOENT) && (errno != ENOTDIR))
3271                return error("%s: %s", new_name, strerror(errno));
3272        return 0;
3273}
3274
3275/*
3276 * Check and apply the patch in-core; leave the result in patch->result
3277 * for the caller to write it out to the final destination.
3278 */
3279static int check_patch(struct patch *patch)
3280{
3281        struct stat st;
3282        const char *old_name = patch->old_name;
3283        const char *new_name = patch->new_name;
3284        const char *name = old_name ? old_name : new_name;
3285        struct cache_entry *ce = NULL;
3286        struct patch *tpatch;
3287        int ok_if_exists;
3288        int status;
3289
3290        patch->rejected = 1; /* we will drop this after we succeed */
3291
3292        status = check_preimage(patch, &ce, &st);
3293        if (status)
3294                return status;
3295        old_name = patch->old_name;
3296
3297        /*
3298         * A type-change diff is always split into a patch to delete
3299         * old, immediately followed by a patch to create new (see
3300         * diff.c::run_diff()); in such a case it is Ok that the entry
3301         * to be deleted by the previous patch is still in the working
3302         * tree and in the index.
3303         *
3304         * A patch to swap-rename between A and B would first rename A
3305         * to B and then rename B to A.  While applying the first one,
3306         * the presense of B should not stop A from getting renamed to
3307         * B; ask to_be_deleted() about the later rename.  Removal of
3308         * B and rename from A to B is handled the same way by asking
3309         * was_deleted().
3310         */
3311        if ((tpatch = in_fn_table(new_name)) &&
3312            (was_deleted(tpatch) || to_be_deleted(tpatch)))
3313                ok_if_exists = 1;
3314        else
3315                ok_if_exists = 0;
3316
3317        if (new_name &&
3318            ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3319                if (check_index &&
3320                    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3321                    !ok_if_exists)
3322                        return error(_("%s: already exists in index"), new_name);
3323                if (!cached) {
3324                        int err = check_to_create_blob(new_name, ok_if_exists);
3325                        if (err)
3326                                return err;
3327                }
3328                if (!patch->new_mode) {
3329                        if (0 < patch->is_new)
3330                                patch->new_mode = S_IFREG | 0644;
3331                        else
3332                                patch->new_mode = patch->old_mode;
3333                }
3334        }
3335
3336        if (new_name && old_name) {
3337                int same = !strcmp(old_name, new_name);
3338                if (!patch->new_mode)
3339                        patch->new_mode = patch->old_mode;
3340                if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
3341                        return error(_("new mode (%o) of %s does not match old mode (%o)%s%s"),
3342                                patch->new_mode, new_name, patch->old_mode,
3343                                same ? "" : " of ", same ? "" : old_name);
3344        }
3345
3346        if (apply_data(patch, &st, ce) < 0)
3347                return error(_("%s: patch does not apply"), name);
3348        patch->rejected = 0;
3349        return 0;
3350}
3351
3352static int check_patch_list(struct patch *patch)
3353{
3354        int err = 0;
3355
3356        prepare_fn_table(patch);
3357        while (patch) {
3358                if (apply_verbosely)
3359                        say_patch_name(stderr,
3360                                       _("Checking patch %s..."), patch);
3361                err |= check_patch(patch);
3362                patch = patch->next;
3363        }
3364        return err;
3365}
3366
3367/* This function tries to read the sha1 from the current index */
3368static int get_current_sha1(const char *path, unsigned char *sha1)
3369{
3370        int pos;
3371
3372        if (read_cache() < 0)
3373                return -1;
3374        pos = cache_name_pos(path, strlen(path));
3375        if (pos < 0)
3376                return -1;
3377        hashcpy(sha1, active_cache[pos]->sha1);
3378        return 0;
3379}
3380
3381/* Build an index that contains the just the files needed for a 3way merge */
3382static void build_fake_ancestor(struct patch *list, const char *filename)
3383{
3384        struct patch *patch;
3385        struct index_state result = { NULL };
3386        int fd;
3387
3388        /* Once we start supporting the reverse patch, it may be
3389         * worth showing the new sha1 prefix, but until then...
3390         */
3391        for (patch = list; patch; patch = patch->next) {
3392                const unsigned char *sha1_ptr;
3393                unsigned char sha1[20];
3394                struct cache_entry *ce;
3395                const char *name;
3396
3397                name = patch->old_name ? patch->old_name : patch->new_name;
3398                if (0 < patch->is_new)
3399                        continue;
3400                else if (get_sha1(patch->old_sha1_prefix, sha1))
3401                        /* git diff has no index line for mode/type changes */
3402                        if (!patch->lines_added && !patch->lines_deleted) {
3403                                if (get_current_sha1(patch->old_name, sha1))
3404                                        die("mode change for %s, which is not "
3405                                                "in current HEAD", name);
3406                                sha1_ptr = sha1;
3407                        } else
3408                                die("sha1 information is lacking or useless "
3409                                        "(%s).", name);
3410                else
3411                        sha1_ptr = sha1;
3412
3413                ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3414                if (!ce)
3415                        die(_("make_cache_entry failed for path '%s'"), name);
3416                if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3417                        die ("Could not add %s to temporary index", name);
3418        }
3419
3420        fd = open(filename, O_WRONLY | O_CREAT, 0666);
3421        if (fd < 0 || write_index(&result, fd) || close(fd))
3422                die ("Could not write temporary index to %s", filename);
3423
3424        discard_index(&result);
3425}
3426
3427static void stat_patch_list(struct patch *patch)
3428{
3429        int files, adds, dels;
3430
3431        for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3432                files++;
3433                adds += patch->lines_added;
3434                dels += patch->lines_deleted;
3435                show_stats(patch);
3436        }
3437
3438        print_stat_summary(stdout, files, adds, dels);
3439}
3440
3441static void numstat_patch_list(struct patch *patch)
3442{
3443        for ( ; patch; patch = patch->next) {
3444                const char *name;
3445                name = patch->new_name ? patch->new_name : patch->old_name;
3446                if (patch->is_binary)
3447                        printf("-\t-\t");
3448                else
3449                        printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3450                write_name_quoted(name, stdout, line_termination);
3451        }
3452}
3453
3454static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3455{
3456        if (mode)
3457                printf(" %s mode %06o %s\n", newdelete, mode, name);
3458        else
3459                printf(" %s %s\n", newdelete, name);
3460}
3461
3462static void show_mode_change(struct patch *p, int show_name)
3463{
3464        if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3465                if (show_name)
3466                        printf(" mode change %06o => %06o %s\n",
3467                               p->old_mode, p->new_mode, p->new_name);
3468                else
3469                        printf(" mode change %06o => %06o\n",
3470                               p->old_mode, p->new_mode);
3471        }
3472}
3473
3474static void show_rename_copy(struct patch *p)
3475{
3476        const char *renamecopy = p->is_rename ? "rename" : "copy";
3477        const char *old, *new;
3478
3479        /* Find common prefix */
3480        old = p->old_name;
3481        new = p->new_name;
3482        while (1) {
3483                const char *slash_old, *slash_new;
3484                slash_old = strchr(old, '/');
3485                slash_new = strchr(new, '/');
3486                if (!slash_old ||
3487                    !slash_new ||
3488                    slash_old - old != slash_new - new ||
3489                    memcmp(old, new, slash_new - new))
3490                        break;
3491                old = slash_old + 1;
3492                new = slash_new + 1;
3493        }
3494        /* p->old_name thru old is the common prefix, and old and new
3495         * through the end of names are renames
3496         */
3497        if (old != p->old_name)
3498                printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3499                       (int)(old - p->old_name), p->old_name,
3500                       old, new, p->score);
3501        else
3502                printf(" %s %s => %s (%d%%)\n", renamecopy,
3503                       p->old_name, p->new_name, p->score);
3504        show_mode_change(p, 0);
3505}
3506
3507static void summary_patch_list(struct patch *patch)
3508{
3509        struct patch *p;
3510
3511        for (p = patch; p; p = p->next) {
3512                if (p->is_new)
3513                        show_file_mode_name("create", p->new_mode, p->new_name);
3514                else if (p->is_delete)
3515                        show_file_mode_name("delete", p->old_mode, p->old_name);
3516                else {
3517                        if (p->is_rename || p->is_copy)
3518                                show_rename_copy(p);
3519                        else {
3520                                if (p->score) {
3521                                        printf(" rewrite %s (%d%%)\n",
3522                                               p->new_name, p->score);
3523                                        show_mode_change(p, 0);
3524                                }
3525                                else
3526                                        show_mode_change(p, 1);
3527                        }
3528                }
3529        }
3530}
3531
3532static void patch_stats(struct patch *patch)
3533{
3534        int lines = patch->lines_added + patch->lines_deleted;
3535
3536        if (lines > max_change)
3537                max_change = lines;
3538        if (patch->old_name) {
3539                int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3540                if (!len)
3541                        len = strlen(patch->old_name);
3542                if (len > max_len)
3543                        max_len = len;
3544        }
3545        if (patch->new_name) {
3546                int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3547                if (!len)
3548                        len = strlen(patch->new_name);
3549                if (len > max_len)
3550                        max_len = len;
3551        }
3552}
3553
3554static void remove_file(struct patch *patch, int rmdir_empty)
3555{
3556        if (update_index) {
3557                if (remove_file_from_cache(patch->old_name) < 0)
3558                        die(_("unable to remove %s from index"), patch->old_name);
3559        }
3560        if (!cached) {
3561                if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3562                        remove_path(patch->old_name);
3563                }
3564        }
3565}
3566
3567static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3568{
3569        struct stat st;
3570        struct cache_entry *ce;
3571        int namelen = strlen(path);
3572        unsigned ce_size = cache_entry_size(namelen);
3573
3574        if (!update_index)
3575                return;
3576
3577        ce = xcalloc(1, ce_size);
3578        memcpy(ce->name, path, namelen);
3579        ce->ce_mode = create_ce_mode(mode);
3580        ce->ce_flags = namelen;
3581        if (S_ISGITLINK(mode)) {
3582                const char *s = buf;
3583
3584                if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3585                        die(_("corrupt patch for subproject %s"), path);
3586        } else {
3587                if (!cached) {
3588                        if (lstat(path, &st) < 0)
3589                                die_errno(_("unable to stat newly created file '%s'"),
3590                                          path);
3591                        fill_stat_cache_info(ce, &st);
3592                }
3593                if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3594                        die(_("unable to create backing store for newly created file %s"), path);
3595        }
3596        if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3597                die(_("unable to add cache entry for %s"), path);
3598}
3599
3600static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3601{
3602        int fd;
3603        struct strbuf nbuf = STRBUF_INIT;
3604
3605        if (S_ISGITLINK(mode)) {
3606                struct stat st;
3607                if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3608                        return 0;
3609                return mkdir(path, 0777);
3610        }
3611
3612        if (has_symlinks && S_ISLNK(mode))
3613                /* Although buf:size is counted string, it also is NUL
3614                 * terminated.
3615                 */
3616                return symlink(buf, path);
3617
3618        fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3619        if (fd < 0)
3620                return -1;
3621
3622        if (convert_to_working_tree(path, buf, size, &nbuf)) {
3623                size = nbuf.len;
3624                buf  = nbuf.buf;
3625        }
3626        write_or_die(fd, buf, size);
3627        strbuf_release(&nbuf);
3628
3629        if (close(fd) < 0)
3630                die_errno(_("closing file '%s'"), path);
3631        return 0;
3632}
3633
3634/*
3635 * We optimistically assume that the directories exist,
3636 * which is true 99% of the time anyway. If they don't,
3637 * we create them and try again.
3638 */
3639static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3640{
3641        if (cached)
3642                return;
3643        if (!try_create_file(path, mode, buf, size))
3644                return;
3645
3646        if (errno == ENOENT) {
3647                if (safe_create_leading_directories(path))
3648                        return;
3649                if (!try_create_file(path, mode, buf, size))
3650                        return;
3651        }
3652
3653        if (errno == EEXIST || errno == EACCES) {
3654                /* We may be trying to create a file where a directory
3655                 * used to be.
3656                 */
3657                struct stat st;
3658                if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3659                        errno = EEXIST;
3660        }
3661
3662        if (errno == EEXIST) {
3663                unsigned int nr = getpid();
3664
3665                for (;;) {
3666                        char newpath[PATH_MAX];
3667                        mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3668                        if (!try_create_file(newpath, mode, buf, size)) {
3669                                if (!rename(newpath, path))
3670                                        return;
3671                                unlink_or_warn(newpath);
3672                                break;
3673                        }
3674                        if (errno != EEXIST)
3675                                break;
3676                        ++nr;
3677                }
3678        }
3679        die_errno(_("unable to write file '%s' mode %o"), path, mode);
3680}
3681
3682static void create_file(struct patch *patch)
3683{
3684        char *path = patch->new_name;
3685        unsigned mode = patch->new_mode;
3686        unsigned long size = patch->resultsize;
3687        char *buf = patch->result;
3688
3689        if (!mode)
3690                mode = S_IFREG | 0644;
3691        create_one_file(path, mode, buf, size);
3692        add_index_file(path, mode, buf, size);
3693}
3694
3695/* phase zero is to remove, phase one is to create */
3696static void write_out_one_result(struct patch *patch, int phase)
3697{
3698        if (patch->is_delete > 0) {
3699                if (phase == 0)
3700                        remove_file(patch, 1);
3701                return;
3702        }
3703        if (patch->is_new > 0 || patch->is_copy) {
3704                if (phase == 1)
3705                        create_file(patch);
3706                return;
3707        }
3708        /*
3709         * Rename or modification boils down to the same
3710         * thing: remove the old, write the new
3711         */
3712        if (phase == 0)
3713                remove_file(patch, patch->is_rename);
3714        if (phase == 1)
3715                create_file(patch);
3716}
3717
3718static int write_out_one_reject(struct patch *patch)
3719{
3720        FILE *rej;
3721        char namebuf[PATH_MAX];
3722        struct fragment *frag;
3723        int cnt = 0;
3724        struct strbuf sb = STRBUF_INIT;
3725
3726        for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3727                if (!frag->rejected)
3728                        continue;
3729                cnt++;
3730        }
3731
3732        if (!cnt) {
3733                if (apply_verbosely)
3734                        say_patch_name(stderr,
3735                                       _("Applied patch %s cleanly."), patch);
3736                return 0;
3737        }
3738
3739        /* This should not happen, because a removal patch that leaves
3740         * contents are marked "rejected" at the patch level.
3741         */
3742        if (!patch->new_name)
3743                die(_("internal error"));
3744
3745        /* Say this even without --verbose */
3746        strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
3747                            "Applying patch %%s with %d rejects...",
3748                            cnt),
3749                    cnt);
3750        say_patch_name(stderr, sb.buf, patch);
3751        strbuf_release(&sb);
3752
3753        cnt = strlen(patch->new_name);
3754        if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3755                cnt = ARRAY_SIZE(namebuf) - 5;
3756                warning(_("truncating .rej filename to %.*s.rej"),
3757                        cnt - 1, patch->new_name);
3758        }
3759        memcpy(namebuf, patch->new_name, cnt);
3760        memcpy(namebuf + cnt, ".rej", 5);
3761
3762        rej = fopen(namebuf, "w");
3763        if (!rej)
3764                return error(_("cannot open %s: %s"), namebuf, strerror(errno));
3765
3766        /* Normal git tools never deal with .rej, so do not pretend
3767         * this is a git patch by saying --git nor give extended
3768         * headers.  While at it, maybe please "kompare" that wants
3769         * the trailing TAB and some garbage at the end of line ;-).
3770         */
3771        fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3772                patch->new_name, patch->new_name);
3773        for (cnt = 1, frag = patch->fragments;
3774             frag;
3775             cnt++, frag = frag->next) {
3776                if (!frag->rejected) {
3777                        fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
3778                        continue;
3779                }
3780                fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
3781                fprintf(rej, "%.*s", frag->size, frag->patch);
3782                if (frag->patch[frag->size-1] != '\n')
3783                        fputc('\n', rej);
3784        }
3785        fclose(rej);
3786        return -1;
3787}
3788
3789static int write_out_results(struct patch *list)
3790{
3791        int phase;
3792        int errs = 0;
3793        struct patch *l;
3794
3795        for (phase = 0; phase < 2; phase++) {
3796                l = list;
3797                while (l) {
3798                        if (l->rejected)
3799                                errs = 1;
3800                        else {
3801                                write_out_one_result(l, phase);
3802                                if (phase == 1 && write_out_one_reject(l))
3803                                        errs = 1;
3804                        }
3805                        l = l->next;
3806                }
3807        }
3808        return errs;
3809}
3810
3811static struct lock_file lock_file;
3812
3813static struct string_list limit_by_name;
3814static int has_include;
3815static void add_name_limit(const char *name, int exclude)
3816{
3817        struct string_list_item *it;
3818
3819        it = string_list_append(&limit_by_name, name);
3820        it->util = exclude ? NULL : (void *) 1;
3821}
3822
3823static int use_patch(struct patch *p)
3824{
3825        const char *pathname = p->new_name ? p->new_name : p->old_name;
3826        int i;
3827
3828        /* Paths outside are not touched regardless of "--include" */
3829        if (0 < prefix_length) {
3830                int pathlen = strlen(pathname);
3831                if (pathlen <= prefix_length ||
3832                    memcmp(prefix, pathname, prefix_length))
3833                        return 0;
3834        }
3835
3836        /* See if it matches any of exclude/include rule */
3837        for (i = 0; i < limit_by_name.nr; i++) {
3838                struct string_list_item *it = &limit_by_name.items[i];
3839                if (!fnmatch(it->string, pathname, 0))
3840                        return (it->util != NULL);
3841        }
3842
3843        /*
3844         * If we had any include, a path that does not match any rule is
3845         * not used.  Otherwise, we saw bunch of exclude rules (or none)
3846         * and such a path is used.
3847         */
3848        return !has_include;
3849}
3850
3851
3852static void prefix_one(char **name)
3853{
3854        char *old_name = *name;
3855        if (!old_name)
3856                return;
3857        *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3858        free(old_name);
3859}
3860
3861static void prefix_patches(struct patch *p)
3862{
3863        if (!prefix || p->is_toplevel_relative)
3864                return;
3865        for ( ; p; p = p->next) {
3866                prefix_one(&p->new_name);
3867                prefix_one(&p->old_name);
3868        }
3869}
3870
3871#define INACCURATE_EOF  (1<<0)
3872#define RECOUNT         (1<<1)
3873
3874static int apply_patch(int fd, const char *filename, int options)
3875{
3876        size_t offset;
3877        struct strbuf buf = STRBUF_INIT; /* owns the patch text */
3878        struct patch *list = NULL, **listp = &list;
3879        int skipped_patch = 0;
3880
3881        patch_input_file = filename;
3882        read_patch_file(&buf, fd);
3883        offset = 0;
3884        while (offset < buf.len) {
3885                struct patch *patch;
3886                int nr;
3887
3888                patch = xcalloc(1, sizeof(*patch));
3889                patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3890                patch->recount =  !!(options & RECOUNT);
3891                nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3892                if (nr < 0)
3893                        break;
3894                if (apply_in_reverse)
3895                        reverse_patches(patch);
3896                if (prefix)
3897                        prefix_patches(patch);
3898                if (use_patch(patch)) {
3899                        patch_stats(patch);
3900                        *listp = patch;
3901                        listp = &patch->next;
3902                }
3903                else {
3904                        free_patch(patch);
3905                        skipped_patch++;
3906                }
3907                offset += nr;
3908        }
3909
3910        if (!list && !skipped_patch)
3911                die(_("unrecognized input"));
3912
3913        if (whitespace_error && (ws_error_action == die_on_ws_error))
3914                apply = 0;
3915
3916        update_index = check_index && apply;
3917        if (update_index && newfd < 0)
3918                newfd = hold_locked_index(&lock_file, 1);
3919
3920        if (check_index) {
3921                if (read_cache() < 0)
3922                        die(_("unable to read index file"));
3923        }
3924
3925        if ((check || apply) &&
3926            check_patch_list(list) < 0 &&
3927            !apply_with_reject)
3928                exit(1);
3929
3930        if (apply && write_out_results(list))
3931                exit(1);
3932
3933        if (fake_ancestor)
3934                build_fake_ancestor(list, fake_ancestor);
3935
3936        if (diffstat)
3937                stat_patch_list(list);
3938
3939        if (numstat)
3940                numstat_patch_list(list);
3941
3942        if (summary)
3943                summary_patch_list(list);
3944
3945        free_patch_list(list);
3946        strbuf_release(&buf);
3947        string_list_clear(&fn_table, 0);
3948        return 0;
3949}
3950
3951static int git_apply_config(const char *var, const char *value, void *cb)
3952{
3953        if (!strcmp(var, "apply.whitespace"))
3954                return git_config_string(&apply_default_whitespace, var, value);
3955        else if (!strcmp(var, "apply.ignorewhitespace"))
3956                return git_config_string(&apply_default_ignorewhitespace, var, value);
3957        return git_default_config(var, value, cb);
3958}
3959
3960static int option_parse_exclude(const struct option *opt,
3961                                const char *arg, int unset)
3962{
3963        add_name_limit(arg, 1);
3964        return 0;
3965}
3966
3967static int option_parse_include(const struct option *opt,
3968                                const char *arg, int unset)
3969{
3970        add_name_limit(arg, 0);
3971        has_include = 1;
3972        return 0;
3973}
3974
3975static int option_parse_p(const struct option *opt,
3976                          const char *arg, int unset)
3977{
3978        p_value = atoi(arg);
3979        p_value_known = 1;
3980        return 0;
3981}
3982
3983static int option_parse_z(const struct option *opt,
3984                          const char *arg, int unset)
3985{
3986        if (unset)
3987                line_termination = '\n';
3988        else
3989                line_termination = 0;
3990        return 0;
3991}
3992
3993static int option_parse_space_change(const struct option *opt,
3994                          const char *arg, int unset)
3995{
3996        if (unset)
3997                ws_ignore_action = ignore_ws_none;
3998        else
3999                ws_ignore_action = ignore_ws_change;
4000        return 0;
4001}
4002
4003static int option_parse_whitespace(const struct option *opt,
4004                                   const char *arg, int unset)
4005{
4006        const char **whitespace_option = opt->value;
4007
4008        *whitespace_option = arg;
4009        parse_whitespace_option(arg);
4010        return 0;
4011}
4012
4013static int option_parse_directory(const struct option *opt,
4014                                  const char *arg, int unset)
4015{
4016        root_len = strlen(arg);
4017        if (root_len && arg[root_len - 1] != '/') {
4018                char *new_root;
4019                root = new_root = xmalloc(root_len + 2);
4020                strcpy(new_root, arg);
4021                strcpy(new_root + root_len++, "/");
4022        } else
4023                root = arg;
4024        return 0;
4025}
4026
4027int cmd_apply(int argc, const char **argv, const char *prefix_)
4028{
4029        int i;
4030        int errs = 0;
4031        int is_not_gitdir = !startup_info->have_repository;
4032        int force_apply = 0;
4033
4034        const char *whitespace_option = NULL;
4035
4036        struct option builtin_apply_options[] = {
4037                { OPTION_CALLBACK, 0, "exclude", NULL, "path",
4038                        "don't apply changes matching the given path",
4039                        0, option_parse_exclude },
4040                { OPTION_CALLBACK, 0, "include", NULL, "path",
4041                        "apply changes matching the given path",
4042                        0, option_parse_include },
4043                { OPTION_CALLBACK, 'p', NULL, NULL, "num",
4044                        "remove <num> leading slashes from traditional diff paths",
4045                        0, option_parse_p },
4046                OPT_BOOLEAN(0, "no-add", &no_add,
4047                        "ignore additions made by the patch"),
4048                OPT_BOOLEAN(0, "stat", &diffstat,
4049                        "instead of applying the patch, output diffstat for the input"),
4050                OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4051                OPT_NOOP_NOARG(0, "binary"),
4052                OPT_BOOLEAN(0, "numstat", &numstat,
4053                        "shows number of added and deleted lines in decimal notation"),
4054                OPT_BOOLEAN(0, "summary", &summary,
4055                        "instead of applying the patch, output a summary for the input"),
4056                OPT_BOOLEAN(0, "check", &check,
4057                        "instead of applying the patch, see if the patch is applicable"),
4058                OPT_BOOLEAN(0, "index", &check_index,
4059                        "make sure the patch is applicable to the current index"),
4060                OPT_BOOLEAN(0, "cached", &cached,
4061                        "apply a patch without touching the working tree"),
4062                OPT_BOOLEAN(0, "apply", &force_apply,
4063                        "also apply the patch (use with --stat/--summary/--check)"),
4064                OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4065                        "build a temporary index based on embedded index information"),
4066                { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4067                        "paths are separated with NUL character",
4068                        PARSE_OPT_NOARG, option_parse_z },
4069                OPT_INTEGER('C', NULL, &p_context,
4070                                "ensure at least <n> lines of context match"),
4071                { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
4072                        "detect new or modified lines that have whitespace errors",
4073                        0, option_parse_whitespace },
4074                { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4075                        "ignore changes in whitespace when finding context",
4076                        PARSE_OPT_NOARG, option_parse_space_change },
4077                { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4078                        "ignore changes in whitespace when finding context",
4079                        PARSE_OPT_NOARG, option_parse_space_change },
4080                OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
4081                        "apply the patch in reverse"),
4082                OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
4083                        "don't expect at least one line of context"),
4084                OPT_BOOLEAN(0, "reject", &apply_with_reject,
4085                        "leave the rejected hunks in corresponding *.rej files"),
4086                OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
4087                        "allow overlapping hunks"),
4088                OPT__VERBOSE(&apply_verbosely, "be verbose"),
4089                OPT_BIT(0, "inaccurate-eof", &options,
4090                        "tolerate incorrectly detected missing new-line at the end of file",
4091                        INACCURATE_EOF),
4092                OPT_BIT(0, "recount", &options,
4093                        "do not trust the line counts in the hunk headers",
4094                        RECOUNT),
4095                { OPTION_CALLBACK, 0, "directory", NULL, "root",
4096                        "prepend <root> to all filenames",
4097                        0, option_parse_directory },
4098                OPT_END()
4099        };
4100
4101        prefix = prefix_;
4102        prefix_length = prefix ? strlen(prefix) : 0;
4103        git_config(git_apply_config, NULL);
4104        if (apply_default_whitespace)
4105                parse_whitespace_option(apply_default_whitespace);
4106        if (apply_default_ignorewhitespace)
4107                parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4108
4109        argc = parse_options(argc, argv, prefix, builtin_apply_options,
4110                        apply_usage, 0);
4111
4112        if (apply_with_reject)
4113                apply = apply_verbosely = 1;
4114        if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4115                apply = 0;
4116        if (check_index && is_not_gitdir)
4117                die(_("--index outside a repository"));
4118        if (cached) {
4119                if (is_not_gitdir)
4120                        die(_("--cached outside a repository"));
4121                check_index = 1;
4122        }
4123        for (i = 0; i < argc; i++) {
4124                const char *arg = argv[i];
4125                int fd;
4126
4127                if (!strcmp(arg, "-")) {
4128                        errs |= apply_patch(0, "<stdin>", options);
4129                        read_stdin = 0;
4130                        continue;
4131                } else if (0 < prefix_length)
4132                        arg = prefix_filename(prefix, prefix_length, arg);
4133
4134                fd = open(arg, O_RDONLY);
4135                if (fd < 0)
4136                        die_errno(_("can't open patch '%s'"), arg);
4137                read_stdin = 0;
4138                set_default_whitespace_mode(whitespace_option);
4139                errs |= apply_patch(fd, arg, options);
4140                close(fd);
4141        }
4142        set_default_whitespace_mode(whitespace_option);
4143        if (read_stdin)
4144                errs |= apply_patch(0, "<stdin>", options);
4145        if (whitespace_error) {
4146                if (squelch_whitespace_errors &&
4147                    squelch_whitespace_errors < whitespace_error) {
4148                        int squelched =
4149                                whitespace_error - squelch_whitespace_errors;
4150                        warning(Q_("squelched %d whitespace error",
4151                                   "squelched %d whitespace errors",
4152                                   squelched),
4153                                squelched);
4154                }
4155                if (ws_error_action == die_on_ws_error)
4156                        die(Q_("%d line adds whitespace errors.",
4157                               "%d lines add whitespace errors.",
4158                               whitespace_error),
4159                            whitespace_error);
4160                if (applied_after_fixing_ws && apply)
4161                        warning("%d line%s applied after"
4162                                " fixing whitespace errors.",
4163                                applied_after_fixing_ws,
4164                                applied_after_fixing_ws == 1 ? "" : "s");
4165                else if (whitespace_error)
4166                        warning(Q_("%d line adds whitespace errors.",
4167                                   "%d lines add whitespace errors.",
4168                                   whitespace_error),
4169                                whitespace_error);
4170        }
4171
4172        if (update_index) {
4173                if (write_cache(newfd, active_cache, active_nr) ||
4174                    commit_locked_index(&lock_file))
4175                        die(_("Unable to write new index file"));
4176        }
4177
4178        return !!errs;
4179}