fast-import.con commit Make sure objects/pack exists before creating a new pack (6e180cd)
   1/*
   2Format of STDIN stream:
   3
   4  stream ::= cmd*;
   5
   6  cmd ::= new_blob
   7        | new_commit
   8        | new_tag
   9        | reset_branch
  10        | checkpoint
  11        | progress
  12        ;
  13
  14  new_blob ::= 'blob' lf
  15    mark?
  16    file_content;
  17  file_content ::= data;
  18
  19  new_commit ::= 'commit' sp ref_str lf
  20    mark?
  21    ('author' sp name '<' email '>' when lf)?
  22    'committer' sp name '<' email '>' when lf
  23    commit_msg
  24    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  25    ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
  26    file_change*
  27    lf?;
  28  commit_msg ::= data;
  29
  30  file_change ::= file_clr
  31    | file_del
  32    | file_rnm
  33    | file_cpy
  34    | file_obm
  35    | file_inm;
  36  file_clr ::= 'deleteall' lf;
  37  file_del ::= 'D' sp path_str lf;
  38  file_rnm ::= 'R' sp path_str sp path_str lf;
  39  file_cpy ::= 'C' sp path_str sp path_str lf;
  40  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  41  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  42    data;
  43
  44  new_tag ::= 'tag' sp tag_str lf
  45    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
  46    ('tagger' sp name '<' email '>' when lf)?
  47    tag_msg;
  48  tag_msg ::= data;
  49
  50  reset_branch ::= 'reset' sp ref_str lf
  51    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  52    lf?;
  53
  54  checkpoint ::= 'checkpoint' lf
  55    lf?;
  56
  57  progress ::= 'progress' sp not_lf* lf
  58    lf?;
  59
  60     # note: the first idnum in a stream should be 1 and subsequent
  61     # idnums should not have gaps between values as this will cause
  62     # the stream parser to reserve space for the gapped values.  An
  63     # idnum can be updated in the future to a new object by issuing
  64     # a new mark directive with the old idnum.
  65     #
  66  mark ::= 'mark' sp idnum lf;
  67  data ::= (delimited_data | exact_data)
  68    lf?;
  69
  70    # note: delim may be any string but must not contain lf.
  71    # data_line may contain any data but must not be exactly
  72    # delim.
  73  delimited_data ::= 'data' sp '<<' delim lf
  74    (data_line lf)*
  75    delim lf;
  76
  77     # note: declen indicates the length of binary_data in bytes.
  78     # declen does not include the lf preceeding the binary data.
  79     #
  80  exact_data ::= 'data' sp declen lf
  81    binary_data;
  82
  83     # note: quoted strings are C-style quoting supporting \c for
  84     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  85     # is the signed byte value in octal.  Note that the only
  86     # characters which must actually be escaped to protect the
  87     # stream formatting is: \, " and LF.  Otherwise these values
  88     # are UTF8.
  89     #
  90  ref_str     ::= ref;
  91  sha1exp_str ::= sha1exp;
  92  tag_str     ::= tag;
  93  path_str    ::= path    | '"' quoted(path)    '"' ;
  94  mode        ::= '100644' | '644'
  95                | '100755' | '755'
  96                | '120000'
  97                ;
  98
  99  declen ::= # unsigned 32 bit value, ascii base10 notation;
 100  bigint ::= # unsigned integer value, ascii base10 notation;
 101  binary_data ::= # file content, not interpreted;
 102
 103  when         ::= raw_when | rfc2822_when;
 104  raw_when     ::= ts sp tz;
 105  rfc2822_when ::= # Valid RFC 2822 date and time;
 106
 107  sp ::= # ASCII space character;
 108  lf ::= # ASCII newline (LF) character;
 109
 110     # note: a colon (':') must precede the numerical value assigned to
 111     # an idnum.  This is to distinguish it from a ref or tag name as
 112     # GIT does not permit ':' in ref or tag strings.
 113     #
 114  idnum   ::= ':' bigint;
 115  path    ::= # GIT style file path, e.g. "a/b/c";
 116  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 117  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 118  sha1exp ::= # Any valid GIT SHA1 expression;
 119  hexsha1 ::= # SHA1 in hexadecimal format;
 120
 121     # note: name and email are UTF8 strings, however name must not
 122     # contain '<' or lf and email must not contain any of the
 123     # following: '<', '>', lf.
 124     #
 125  name  ::= # valid GIT author/committer name;
 126  email ::= # valid GIT author/committer email;
 127  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 128  tz    ::= # GIT style timezone;
 129
 130     # note: comments may appear anywhere in the input, except
 131     # within a data command.  Any form of the data command
 132     # always escapes the related input from comment processing.
 133     #
 134     # In case it is not clear, the '#' that starts the comment
 135     # must be the first character on that the line (an lf have
 136     # preceeded it).
 137     #
 138  comment ::= '#' not_lf* lf;
 139  not_lf  ::= # Any byte that is not ASCII newline (LF);
 140*/
 141
 142#include "builtin.h"
 143#include "cache.h"
 144#include "object.h"
 145#include "blob.h"
 146#include "tree.h"
 147#include "commit.h"
 148#include "delta.h"
 149#include "pack.h"
 150#include "refs.h"
 151#include "csum-file.h"
 152#include "quote.h"
 153
 154#define PACK_ID_BITS 16
 155#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 156#define DEPTH_BITS 13
 157#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 158
 159struct object_entry
 160{
 161        struct object_entry *next;
 162        uint32_t offset;
 163        uint32_t type : TYPE_BITS,
 164                pack_id : PACK_ID_BITS,
 165                depth : DEPTH_BITS;
 166        unsigned char sha1[20];
 167};
 168
 169struct object_entry_pool
 170{
 171        struct object_entry_pool *next_pool;
 172        struct object_entry *next_free;
 173        struct object_entry *end;
 174        struct object_entry entries[FLEX_ARRAY]; /* more */
 175};
 176
 177struct mark_set
 178{
 179        union {
 180                struct object_entry *marked[1024];
 181                struct mark_set *sets[1024];
 182        } data;
 183        unsigned int shift;
 184};
 185
 186struct last_object
 187{
 188        struct strbuf data;
 189        uint32_t offset;
 190        unsigned int depth;
 191        unsigned no_swap : 1;
 192};
 193
 194struct mem_pool
 195{
 196        struct mem_pool *next_pool;
 197        char *next_free;
 198        char *end;
 199        uintmax_t space[FLEX_ARRAY]; /* more */
 200};
 201
 202struct atom_str
 203{
 204        struct atom_str *next_atom;
 205        unsigned short str_len;
 206        char str_dat[FLEX_ARRAY]; /* more */
 207};
 208
 209struct tree_content;
 210struct tree_entry
 211{
 212        struct tree_content *tree;
 213        struct atom_str* name;
 214        struct tree_entry_ms
 215        {
 216                uint16_t mode;
 217                unsigned char sha1[20];
 218        } versions[2];
 219};
 220
 221struct tree_content
 222{
 223        unsigned int entry_capacity; /* must match avail_tree_content */
 224        unsigned int entry_count;
 225        unsigned int delta_depth;
 226        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 227};
 228
 229struct avail_tree_content
 230{
 231        unsigned int entry_capacity; /* must match tree_content */
 232        struct avail_tree_content *next_avail;
 233};
 234
 235struct branch
 236{
 237        struct branch *table_next_branch;
 238        struct branch *active_next_branch;
 239        const char *name;
 240        struct tree_entry branch_tree;
 241        uintmax_t last_commit;
 242        unsigned active : 1;
 243        unsigned pack_id : PACK_ID_BITS;
 244        unsigned char sha1[20];
 245};
 246
 247struct tag
 248{
 249        struct tag *next_tag;
 250        const char *name;
 251        unsigned int pack_id;
 252        unsigned char sha1[20];
 253};
 254
 255struct hash_list
 256{
 257        struct hash_list *next;
 258        unsigned char sha1[20];
 259};
 260
 261typedef enum {
 262        WHENSPEC_RAW = 1,
 263        WHENSPEC_RFC2822,
 264        WHENSPEC_NOW,
 265} whenspec_type;
 266
 267struct recent_command
 268{
 269        struct recent_command *prev;
 270        struct recent_command *next;
 271        char *buf;
 272};
 273
 274/* Configured limits on output */
 275static unsigned long max_depth = 10;
 276static off_t max_packsize = (1LL << 32) - 1;
 277static int force_update;
 278static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 279static int pack_compression_seen;
 280
 281/* Stats and misc. counters */
 282static uintmax_t alloc_count;
 283static uintmax_t marks_set_count;
 284static uintmax_t object_count_by_type[1 << TYPE_BITS];
 285static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 286static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 287static unsigned long object_count;
 288static unsigned long branch_count;
 289static unsigned long branch_load_count;
 290static int failure;
 291static FILE *pack_edges;
 292
 293/* Memory pools */
 294static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 295static size_t total_allocd;
 296static struct mem_pool *mem_pool;
 297
 298/* Atom management */
 299static unsigned int atom_table_sz = 4451;
 300static unsigned int atom_cnt;
 301static struct atom_str **atom_table;
 302
 303/* The .pack file being generated */
 304static unsigned int pack_id;
 305static struct packed_git *pack_data;
 306static struct packed_git **all_packs;
 307static unsigned long pack_size;
 308
 309/* Table of objects we've written. */
 310static unsigned int object_entry_alloc = 5000;
 311static struct object_entry_pool *blocks;
 312static struct object_entry *object_table[1 << 16];
 313static struct mark_set *marks;
 314static const char* mark_file;
 315
 316/* Our last blob */
 317static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 318
 319/* Tree management */
 320static unsigned int tree_entry_alloc = 1000;
 321static void *avail_tree_entry;
 322static unsigned int avail_tree_table_sz = 100;
 323static struct avail_tree_content **avail_tree_table;
 324static struct strbuf old_tree = STRBUF_INIT;
 325static struct strbuf new_tree = STRBUF_INIT;
 326
 327/* Branch data */
 328static unsigned long max_active_branches = 5;
 329static unsigned long cur_active_branches;
 330static unsigned long branch_table_sz = 1039;
 331static struct branch **branch_table;
 332static struct branch *active_branches;
 333
 334/* Tag data */
 335static struct tag *first_tag;
 336static struct tag *last_tag;
 337
 338/* Input stream parsing */
 339static whenspec_type whenspec = WHENSPEC_RAW;
 340static struct strbuf command_buf = STRBUF_INIT;
 341static int unread_command_buf;
 342static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 343static struct recent_command *cmd_tail = &cmd_hist;
 344static struct recent_command *rc_free;
 345static unsigned int cmd_save = 100;
 346static uintmax_t next_mark;
 347static struct strbuf new_data = STRBUF_INIT;
 348
 349static void write_branch_report(FILE *rpt, struct branch *b)
 350{
 351        fprintf(rpt, "%s:\n", b->name);
 352
 353        fprintf(rpt, "  status      :");
 354        if (b->active)
 355                fputs(" active", rpt);
 356        if (b->branch_tree.tree)
 357                fputs(" loaded", rpt);
 358        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 359                fputs(" dirty", rpt);
 360        fputc('\n', rpt);
 361
 362        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 363        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 364        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 365        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 366
 367        fputs("  last pack   : ", rpt);
 368        if (b->pack_id < MAX_PACK_ID)
 369                fprintf(rpt, "%u", b->pack_id);
 370        fputc('\n', rpt);
 371
 372        fputc('\n', rpt);
 373}
 374
 375static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 376
 377static void write_crash_report(const char *err)
 378{
 379        char *loc = git_path("fast_import_crash_%d", getpid());
 380        FILE *rpt = fopen(loc, "w");
 381        struct branch *b;
 382        unsigned long lu;
 383        struct recent_command *rc;
 384
 385        if (!rpt) {
 386                error("can't write crash report %s: %s", loc, strerror(errno));
 387                return;
 388        }
 389
 390        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 391
 392        fprintf(rpt, "fast-import crash report:\n");
 393        fprintf(rpt, "    fast-import process: %d\n", getpid());
 394        fprintf(rpt, "    parent process     : %d\n", getppid());
 395        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 396        fputc('\n', rpt);
 397
 398        fputs("fatal: ", rpt);
 399        fputs(err, rpt);
 400        fputc('\n', rpt);
 401
 402        fputc('\n', rpt);
 403        fputs("Most Recent Commands Before Crash\n", rpt);
 404        fputs("---------------------------------\n", rpt);
 405        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 406                if (rc->next == &cmd_hist)
 407                        fputs("* ", rpt);
 408                else
 409                        fputs("  ", rpt);
 410                fputs(rc->buf, rpt);
 411                fputc('\n', rpt);
 412        }
 413
 414        fputc('\n', rpt);
 415        fputs("Active Branch LRU\n", rpt);
 416        fputs("-----------------\n", rpt);
 417        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 418                cur_active_branches,
 419                max_active_branches);
 420        fputc('\n', rpt);
 421        fputs("  pos  clock name\n", rpt);
 422        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 423        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 424                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 425                        ++lu, b->last_commit, b->name);
 426
 427        fputc('\n', rpt);
 428        fputs("Inactive Branches\n", rpt);
 429        fputs("-----------------\n", rpt);
 430        for (lu = 0; lu < branch_table_sz; lu++) {
 431                for (b = branch_table[lu]; b; b = b->table_next_branch)
 432                        write_branch_report(rpt, b);
 433        }
 434
 435        if (first_tag) {
 436                struct tag *tg;
 437                fputc('\n', rpt);
 438                fputs("Annotated Tags\n", rpt);
 439                fputs("--------------\n", rpt);
 440                for (tg = first_tag; tg; tg = tg->next_tag) {
 441                        fputs(sha1_to_hex(tg->sha1), rpt);
 442                        fputc(' ', rpt);
 443                        fputs(tg->name, rpt);
 444                        fputc('\n', rpt);
 445                }
 446        }
 447
 448        fputc('\n', rpt);
 449        fputs("Marks\n", rpt);
 450        fputs("-----\n", rpt);
 451        if (mark_file)
 452                fprintf(rpt, "  exported to %s\n", mark_file);
 453        else
 454                dump_marks_helper(rpt, 0, marks);
 455
 456        fputc('\n', rpt);
 457        fputs("-------------------\n", rpt);
 458        fputs("END OF CRASH REPORT\n", rpt);
 459        fclose(rpt);
 460}
 461
 462static void end_packfile(void);
 463static void unkeep_all_packs(void);
 464static void dump_marks(void);
 465
 466static NORETURN void die_nicely(const char *err, va_list params)
 467{
 468        static int zombie;
 469        char message[2 * PATH_MAX];
 470
 471        vsnprintf(message, sizeof(message), err, params);
 472        fputs("fatal: ", stderr);
 473        fputs(message, stderr);
 474        fputc('\n', stderr);
 475
 476        if (!zombie) {
 477                zombie = 1;
 478                write_crash_report(message);
 479                end_packfile();
 480                unkeep_all_packs();
 481                dump_marks();
 482        }
 483        exit(128);
 484}
 485
 486static void alloc_objects(unsigned int cnt)
 487{
 488        struct object_entry_pool *b;
 489
 490        b = xmalloc(sizeof(struct object_entry_pool)
 491                + cnt * sizeof(struct object_entry));
 492        b->next_pool = blocks;
 493        b->next_free = b->entries;
 494        b->end = b->entries + cnt;
 495        blocks = b;
 496        alloc_count += cnt;
 497}
 498
 499static struct object_entry *new_object(unsigned char *sha1)
 500{
 501        struct object_entry *e;
 502
 503        if (blocks->next_free == blocks->end)
 504                alloc_objects(object_entry_alloc);
 505
 506        e = blocks->next_free++;
 507        hashcpy(e->sha1, sha1);
 508        return e;
 509}
 510
 511static struct object_entry *find_object(unsigned char *sha1)
 512{
 513        unsigned int h = sha1[0] << 8 | sha1[1];
 514        struct object_entry *e;
 515        for (e = object_table[h]; e; e = e->next)
 516                if (!hashcmp(sha1, e->sha1))
 517                        return e;
 518        return NULL;
 519}
 520
 521static struct object_entry *insert_object(unsigned char *sha1)
 522{
 523        unsigned int h = sha1[0] << 8 | sha1[1];
 524        struct object_entry *e = object_table[h];
 525        struct object_entry *p = NULL;
 526
 527        while (e) {
 528                if (!hashcmp(sha1, e->sha1))
 529                        return e;
 530                p = e;
 531                e = e->next;
 532        }
 533
 534        e = new_object(sha1);
 535        e->next = NULL;
 536        e->offset = 0;
 537        if (p)
 538                p->next = e;
 539        else
 540                object_table[h] = e;
 541        return e;
 542}
 543
 544static unsigned int hc_str(const char *s, size_t len)
 545{
 546        unsigned int r = 0;
 547        while (len-- > 0)
 548                r = r * 31 + *s++;
 549        return r;
 550}
 551
 552static void *pool_alloc(size_t len)
 553{
 554        struct mem_pool *p;
 555        void *r;
 556
 557        for (p = mem_pool; p; p = p->next_pool)
 558                if ((p->end - p->next_free >= len))
 559                        break;
 560
 561        if (!p) {
 562                if (len >= (mem_pool_alloc/2)) {
 563                        total_allocd += len;
 564                        return xmalloc(len);
 565                }
 566                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 567                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 568                p->next_pool = mem_pool;
 569                p->next_free = (char *) p->space;
 570                p->end = p->next_free + mem_pool_alloc;
 571                mem_pool = p;
 572        }
 573
 574        r = p->next_free;
 575        /* round out to a 'uintmax_t' alignment */
 576        if (len & (sizeof(uintmax_t) - 1))
 577                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 578        p->next_free += len;
 579        return r;
 580}
 581
 582static void *pool_calloc(size_t count, size_t size)
 583{
 584        size_t len = count * size;
 585        void *r = pool_alloc(len);
 586        memset(r, 0, len);
 587        return r;
 588}
 589
 590static char *pool_strdup(const char *s)
 591{
 592        char *r = pool_alloc(strlen(s) + 1);
 593        strcpy(r, s);
 594        return r;
 595}
 596
 597static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 598{
 599        struct mark_set *s = marks;
 600        while ((idnum >> s->shift) >= 1024) {
 601                s = pool_calloc(1, sizeof(struct mark_set));
 602                s->shift = marks->shift + 10;
 603                s->data.sets[0] = marks;
 604                marks = s;
 605        }
 606        while (s->shift) {
 607                uintmax_t i = idnum >> s->shift;
 608                idnum -= i << s->shift;
 609                if (!s->data.sets[i]) {
 610                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 611                        s->data.sets[i]->shift = s->shift - 10;
 612                }
 613                s = s->data.sets[i];
 614        }
 615        if (!s->data.marked[idnum])
 616                marks_set_count++;
 617        s->data.marked[idnum] = oe;
 618}
 619
 620static struct object_entry *find_mark(uintmax_t idnum)
 621{
 622        uintmax_t orig_idnum = idnum;
 623        struct mark_set *s = marks;
 624        struct object_entry *oe = NULL;
 625        if ((idnum >> s->shift) < 1024) {
 626                while (s && s->shift) {
 627                        uintmax_t i = idnum >> s->shift;
 628                        idnum -= i << s->shift;
 629                        s = s->data.sets[i];
 630                }
 631                if (s)
 632                        oe = s->data.marked[idnum];
 633        }
 634        if (!oe)
 635                die("mark :%" PRIuMAX " not declared", orig_idnum);
 636        return oe;
 637}
 638
 639static struct atom_str *to_atom(const char *s, unsigned short len)
 640{
 641        unsigned int hc = hc_str(s, len) % atom_table_sz;
 642        struct atom_str *c;
 643
 644        for (c = atom_table[hc]; c; c = c->next_atom)
 645                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 646                        return c;
 647
 648        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 649        c->str_len = len;
 650        strncpy(c->str_dat, s, len);
 651        c->str_dat[len] = 0;
 652        c->next_atom = atom_table[hc];
 653        atom_table[hc] = c;
 654        atom_cnt++;
 655        return c;
 656}
 657
 658static struct branch *lookup_branch(const char *name)
 659{
 660        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 661        struct branch *b;
 662
 663        for (b = branch_table[hc]; b; b = b->table_next_branch)
 664                if (!strcmp(name, b->name))
 665                        return b;
 666        return NULL;
 667}
 668
 669static struct branch *new_branch(const char *name)
 670{
 671        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 672        struct branch* b = lookup_branch(name);
 673
 674        if (b)
 675                die("Invalid attempt to create duplicate branch: %s", name);
 676        switch (check_ref_format(name)) {
 677        case 0: break; /* its valid */
 678        case CHECK_REF_FORMAT_ONELEVEL:
 679                break; /* valid, but too few '/', allow anyway */
 680        default:
 681                die("Branch name doesn't conform to GIT standards: %s", name);
 682        }
 683
 684        b = pool_calloc(1, sizeof(struct branch));
 685        b->name = pool_strdup(name);
 686        b->table_next_branch = branch_table[hc];
 687        b->branch_tree.versions[0].mode = S_IFDIR;
 688        b->branch_tree.versions[1].mode = S_IFDIR;
 689        b->active = 0;
 690        b->pack_id = MAX_PACK_ID;
 691        branch_table[hc] = b;
 692        branch_count++;
 693        return b;
 694}
 695
 696static unsigned int hc_entries(unsigned int cnt)
 697{
 698        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 699        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 700}
 701
 702static struct tree_content *new_tree_content(unsigned int cnt)
 703{
 704        struct avail_tree_content *f, *l = NULL;
 705        struct tree_content *t;
 706        unsigned int hc = hc_entries(cnt);
 707
 708        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 709                if (f->entry_capacity >= cnt)
 710                        break;
 711
 712        if (f) {
 713                if (l)
 714                        l->next_avail = f->next_avail;
 715                else
 716                        avail_tree_table[hc] = f->next_avail;
 717        } else {
 718                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 719                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 720                f->entry_capacity = cnt;
 721        }
 722
 723        t = (struct tree_content*)f;
 724        t->entry_count = 0;
 725        t->delta_depth = 0;
 726        return t;
 727}
 728
 729static void release_tree_entry(struct tree_entry *e);
 730static void release_tree_content(struct tree_content *t)
 731{
 732        struct avail_tree_content *f = (struct avail_tree_content*)t;
 733        unsigned int hc = hc_entries(f->entry_capacity);
 734        f->next_avail = avail_tree_table[hc];
 735        avail_tree_table[hc] = f;
 736}
 737
 738static void release_tree_content_recursive(struct tree_content *t)
 739{
 740        unsigned int i;
 741        for (i = 0; i < t->entry_count; i++)
 742                release_tree_entry(t->entries[i]);
 743        release_tree_content(t);
 744}
 745
 746static struct tree_content *grow_tree_content(
 747        struct tree_content *t,
 748        int amt)
 749{
 750        struct tree_content *r = new_tree_content(t->entry_count + amt);
 751        r->entry_count = t->entry_count;
 752        r->delta_depth = t->delta_depth;
 753        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 754        release_tree_content(t);
 755        return r;
 756}
 757
 758static struct tree_entry *new_tree_entry(void)
 759{
 760        struct tree_entry *e;
 761
 762        if (!avail_tree_entry) {
 763                unsigned int n = tree_entry_alloc;
 764                total_allocd += n * sizeof(struct tree_entry);
 765                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 766                while (n-- > 1) {
 767                        *((void**)e) = e + 1;
 768                        e++;
 769                }
 770                *((void**)e) = NULL;
 771        }
 772
 773        e = avail_tree_entry;
 774        avail_tree_entry = *((void**)e);
 775        return e;
 776}
 777
 778static void release_tree_entry(struct tree_entry *e)
 779{
 780        if (e->tree)
 781                release_tree_content_recursive(e->tree);
 782        *((void**)e) = avail_tree_entry;
 783        avail_tree_entry = e;
 784}
 785
 786static struct tree_content *dup_tree_content(struct tree_content *s)
 787{
 788        struct tree_content *d;
 789        struct tree_entry *a, *b;
 790        unsigned int i;
 791
 792        if (!s)
 793                return NULL;
 794        d = new_tree_content(s->entry_count);
 795        for (i = 0; i < s->entry_count; i++) {
 796                a = s->entries[i];
 797                b = new_tree_entry();
 798                memcpy(b, a, sizeof(*a));
 799                if (a->tree && is_null_sha1(b->versions[1].sha1))
 800                        b->tree = dup_tree_content(a->tree);
 801                else
 802                        b->tree = NULL;
 803                d->entries[i] = b;
 804        }
 805        d->entry_count = s->entry_count;
 806        d->delta_depth = s->delta_depth;
 807
 808        return d;
 809}
 810
 811static void start_packfile(void)
 812{
 813        static char tmpfile[PATH_MAX];
 814        struct packed_git *p;
 815        struct pack_header hdr;
 816        int pack_fd;
 817
 818        pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 819                              "pack/tmp_pack_XXXXXX");
 820        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 821        strcpy(p->pack_name, tmpfile);
 822        p->pack_fd = pack_fd;
 823
 824        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 825        hdr.hdr_version = htonl(2);
 826        hdr.hdr_entries = 0;
 827        write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 828
 829        pack_data = p;
 830        pack_size = sizeof(hdr);
 831        object_count = 0;
 832
 833        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 834        all_packs[pack_id] = p;
 835}
 836
 837static int oecmp (const void *a_, const void *b_)
 838{
 839        struct object_entry *a = *((struct object_entry**)a_);
 840        struct object_entry *b = *((struct object_entry**)b_);
 841        return hashcmp(a->sha1, b->sha1);
 842}
 843
 844static char *create_index(void)
 845{
 846        static char tmpfile[PATH_MAX];
 847        SHA_CTX ctx;
 848        struct sha1file *f;
 849        struct object_entry **idx, **c, **last, *e;
 850        struct object_entry_pool *o;
 851        uint32_t array[256];
 852        int i, idx_fd;
 853
 854        /* Build the sorted table of object IDs. */
 855        idx = xmalloc(object_count * sizeof(struct object_entry*));
 856        c = idx;
 857        for (o = blocks; o; o = o->next_pool)
 858                for (e = o->next_free; e-- != o->entries;)
 859                        if (pack_id == e->pack_id)
 860                                *c++ = e;
 861        last = idx + object_count;
 862        if (c != last)
 863                die("internal consistency error creating the index");
 864        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 865
 866        /* Generate the fan-out array. */
 867        c = idx;
 868        for (i = 0; i < 256; i++) {
 869                struct object_entry **next = c;;
 870                while (next < last) {
 871                        if ((*next)->sha1[0] != i)
 872                                break;
 873                        next++;
 874                }
 875                array[i] = htonl(next - idx);
 876                c = next;
 877        }
 878
 879        idx_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 880                             "pack/tmp_idx_XXXXXX");
 881        f = sha1fd(idx_fd, tmpfile);
 882        sha1write(f, array, 256 * sizeof(int));
 883        SHA1_Init(&ctx);
 884        for (c = idx; c != last; c++) {
 885                uint32_t offset = htonl((*c)->offset);
 886                sha1write(f, &offset, 4);
 887                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 888                SHA1_Update(&ctx, (*c)->sha1, 20);
 889        }
 890        sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
 891        sha1close(f, NULL, CSUM_FSYNC);
 892        free(idx);
 893        SHA1_Final(pack_data->sha1, &ctx);
 894        return tmpfile;
 895}
 896
 897static char *keep_pack(char *curr_index_name)
 898{
 899        static char name[PATH_MAX];
 900        static const char *keep_msg = "fast-import";
 901        int keep_fd;
 902
 903        chmod(pack_data->pack_name, 0444);
 904        chmod(curr_index_name, 0444);
 905
 906        keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
 907        if (keep_fd < 0)
 908                die("cannot create keep file");
 909        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 910        if (close(keep_fd))
 911                die("failed to write keep file");
 912
 913        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 914                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 915        if (move_temp_to_file(pack_data->pack_name, name))
 916                die("cannot store pack file");
 917
 918        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 919                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 920        if (move_temp_to_file(curr_index_name, name))
 921                die("cannot store index file");
 922        return name;
 923}
 924
 925static void unkeep_all_packs(void)
 926{
 927        static char name[PATH_MAX];
 928        int k;
 929
 930        for (k = 0; k < pack_id; k++) {
 931                struct packed_git *p = all_packs[k];
 932                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 933                         get_object_directory(), sha1_to_hex(p->sha1));
 934                unlink(name);
 935        }
 936}
 937
 938static void end_packfile(void)
 939{
 940        struct packed_git *old_p = pack_data, *new_p;
 941
 942        if (object_count) {
 943                char *idx_name;
 944                int i;
 945                struct branch *b;
 946                struct tag *t;
 947
 948                close_pack_windows(pack_data);
 949                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 950                                    pack_data->pack_name, object_count,
 951                                    NULL, 0);
 952                close(pack_data->pack_fd);
 953                idx_name = keep_pack(create_index());
 954
 955                /* Register the packfile with core git's machinary. */
 956                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 957                if (!new_p)
 958                        die("core git rejected index %s", idx_name);
 959                all_packs[pack_id] = new_p;
 960                install_packed_git(new_p);
 961
 962                /* Print the boundary */
 963                if (pack_edges) {
 964                        fprintf(pack_edges, "%s:", new_p->pack_name);
 965                        for (i = 0; i < branch_table_sz; i++) {
 966                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 967                                        if (b->pack_id == pack_id)
 968                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 969                                }
 970                        }
 971                        for (t = first_tag; t; t = t->next_tag) {
 972                                if (t->pack_id == pack_id)
 973                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 974                        }
 975                        fputc('\n', pack_edges);
 976                        fflush(pack_edges);
 977                }
 978
 979                pack_id++;
 980        }
 981        else {
 982                close(old_p->pack_fd);
 983                unlink(old_p->pack_name);
 984        }
 985        free(old_p);
 986
 987        /* We can't carry a delta across packfiles. */
 988        strbuf_release(&last_blob.data);
 989        last_blob.offset = 0;
 990        last_blob.depth = 0;
 991}
 992
 993static void cycle_packfile(void)
 994{
 995        end_packfile();
 996        start_packfile();
 997}
 998
 999static size_t encode_header(
1000        enum object_type type,
1001        size_t size,
1002        unsigned char *hdr)
1003{
1004        int n = 1;
1005        unsigned char c;
1006
1007        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
1008                die("bad type %d", type);
1009
1010        c = (type << 4) | (size & 15);
1011        size >>= 4;
1012        while (size) {
1013                *hdr++ = c | 0x80;
1014                c = size & 0x7f;
1015                size >>= 7;
1016                n++;
1017        }
1018        *hdr = c;
1019        return n;
1020}
1021
1022static int store_object(
1023        enum object_type type,
1024        struct strbuf *dat,
1025        struct last_object *last,
1026        unsigned char *sha1out,
1027        uintmax_t mark)
1028{
1029        void *out, *delta;
1030        struct object_entry *e;
1031        unsigned char hdr[96];
1032        unsigned char sha1[20];
1033        unsigned long hdrlen, deltalen;
1034        SHA_CTX c;
1035        z_stream s;
1036
1037        hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1038                (unsigned long)dat->len) + 1;
1039        SHA1_Init(&c);
1040        SHA1_Update(&c, hdr, hdrlen);
1041        SHA1_Update(&c, dat->buf, dat->len);
1042        SHA1_Final(sha1, &c);
1043        if (sha1out)
1044                hashcpy(sha1out, sha1);
1045
1046        e = insert_object(sha1);
1047        if (mark)
1048                insert_mark(mark, e);
1049        if (e->offset) {
1050                duplicate_count_by_type[type]++;
1051                return 1;
1052        } else if (find_sha1_pack(sha1, packed_git)) {
1053                e->type = type;
1054                e->pack_id = MAX_PACK_ID;
1055                e->offset = 1; /* just not zero! */
1056                duplicate_count_by_type[type]++;
1057                return 1;
1058        }
1059
1060        if (last && last->data.buf && last->depth < max_depth) {
1061                delta = diff_delta(last->data.buf, last->data.len,
1062                        dat->buf, dat->len,
1063                        &deltalen, 0);
1064                if (delta && deltalen >= dat->len) {
1065                        free(delta);
1066                        delta = NULL;
1067                }
1068        } else
1069                delta = NULL;
1070
1071        memset(&s, 0, sizeof(s));
1072        deflateInit(&s, pack_compression_level);
1073        if (delta) {
1074                s.next_in = delta;
1075                s.avail_in = deltalen;
1076        } else {
1077                s.next_in = (void *)dat->buf;
1078                s.avail_in = dat->len;
1079        }
1080        s.avail_out = deflateBound(&s, s.avail_in);
1081        s.next_out = out = xmalloc(s.avail_out);
1082        while (deflate(&s, Z_FINISH) == Z_OK)
1083                /* nothing */;
1084        deflateEnd(&s);
1085
1086        /* Determine if we should auto-checkpoint. */
1087        if ((pack_size + 60 + s.total_out) > max_packsize
1088                || (pack_size + 60 + s.total_out) < pack_size) {
1089
1090                /* This new object needs to *not* have the current pack_id. */
1091                e->pack_id = pack_id + 1;
1092                cycle_packfile();
1093
1094                /* We cannot carry a delta into the new pack. */
1095                if (delta) {
1096                        free(delta);
1097                        delta = NULL;
1098
1099                        memset(&s, 0, sizeof(s));
1100                        deflateInit(&s, pack_compression_level);
1101                        s.next_in = (void *)dat->buf;
1102                        s.avail_in = dat->len;
1103                        s.avail_out = deflateBound(&s, s.avail_in);
1104                        s.next_out = out = xrealloc(out, s.avail_out);
1105                        while (deflate(&s, Z_FINISH) == Z_OK)
1106                                /* nothing */;
1107                        deflateEnd(&s);
1108                }
1109        }
1110
1111        e->type = type;
1112        e->pack_id = pack_id;
1113        e->offset = pack_size;
1114        object_count++;
1115        object_count_by_type[type]++;
1116
1117        if (delta) {
1118                unsigned long ofs = e->offset - last->offset;
1119                unsigned pos = sizeof(hdr) - 1;
1120
1121                delta_count_by_type[type]++;
1122                e->depth = last->depth + 1;
1123
1124                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1125                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1126                pack_size += hdrlen;
1127
1128                hdr[pos] = ofs & 127;
1129                while (ofs >>= 7)
1130                        hdr[--pos] = 128 | (--ofs & 127);
1131                write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1132                pack_size += sizeof(hdr) - pos;
1133        } else {
1134                e->depth = 0;
1135                hdrlen = encode_header(type, dat->len, hdr);
1136                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1137                pack_size += hdrlen;
1138        }
1139
1140        write_or_die(pack_data->pack_fd, out, s.total_out);
1141        pack_size += s.total_out;
1142
1143        free(out);
1144        free(delta);
1145        if (last) {
1146                if (last->no_swap) {
1147                        last->data = *dat;
1148                } else {
1149                        strbuf_swap(&last->data, dat);
1150                }
1151                last->offset = e->offset;
1152                last->depth = e->depth;
1153        }
1154        return 0;
1155}
1156
1157/* All calls must be guarded by find_object() or find_mark() to
1158 * ensure the 'struct object_entry' passed was written by this
1159 * process instance.  We unpack the entry by the offset, avoiding
1160 * the need for the corresponding .idx file.  This unpacking rule
1161 * works because we only use OBJ_REF_DELTA within the packfiles
1162 * created by fast-import.
1163 *
1164 * oe must not be NULL.  Such an oe usually comes from giving
1165 * an unknown SHA-1 to find_object() or an undefined mark to
1166 * find_mark().  Callers must test for this condition and use
1167 * the standard read_sha1_file() when it happens.
1168 *
1169 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1170 * find_mark(), where the mark was reloaded from an existing marks
1171 * file and is referencing an object that this fast-import process
1172 * instance did not write out to a packfile.  Callers must test for
1173 * this condition and use read_sha1_file() instead.
1174 */
1175static void *gfi_unpack_entry(
1176        struct object_entry *oe,
1177        unsigned long *sizep)
1178{
1179        enum object_type type;
1180        struct packed_git *p = all_packs[oe->pack_id];
1181        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1182                /* The object is stored in the packfile we are writing to
1183                 * and we have modified it since the last time we scanned
1184                 * back to read a previously written object.  If an old
1185                 * window covered [p->pack_size, p->pack_size + 20) its
1186                 * data is stale and is not valid.  Closing all windows
1187                 * and updating the packfile length ensures we can read
1188                 * the newly written data.
1189                 */
1190                close_pack_windows(p);
1191
1192                /* We have to offer 20 bytes additional on the end of
1193                 * the packfile as the core unpacker code assumes the
1194                 * footer is present at the file end and must promise
1195                 * at least 20 bytes within any window it maps.  But
1196                 * we don't actually create the footer here.
1197                 */
1198                p->pack_size = pack_size + 20;
1199        }
1200        return unpack_entry(p, oe->offset, &type, sizep);
1201}
1202
1203static const char *get_mode(const char *str, uint16_t *modep)
1204{
1205        unsigned char c;
1206        uint16_t mode = 0;
1207
1208        while ((c = *str++) != ' ') {
1209                if (c < '0' || c > '7')
1210                        return NULL;
1211                mode = (mode << 3) + (c - '0');
1212        }
1213        *modep = mode;
1214        return str;
1215}
1216
1217static void load_tree(struct tree_entry *root)
1218{
1219        unsigned char* sha1 = root->versions[1].sha1;
1220        struct object_entry *myoe;
1221        struct tree_content *t;
1222        unsigned long size;
1223        char *buf;
1224        const char *c;
1225
1226        root->tree = t = new_tree_content(8);
1227        if (is_null_sha1(sha1))
1228                return;
1229
1230        myoe = find_object(sha1);
1231        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1232                if (myoe->type != OBJ_TREE)
1233                        die("Not a tree: %s", sha1_to_hex(sha1));
1234                t->delta_depth = myoe->depth;
1235                buf = gfi_unpack_entry(myoe, &size);
1236                if (!buf)
1237                        die("Can't load tree %s", sha1_to_hex(sha1));
1238        } else {
1239                enum object_type type;
1240                buf = read_sha1_file(sha1, &type, &size);
1241                if (!buf || type != OBJ_TREE)
1242                        die("Can't load tree %s", sha1_to_hex(sha1));
1243        }
1244
1245        c = buf;
1246        while (c != (buf + size)) {
1247                struct tree_entry *e = new_tree_entry();
1248
1249                if (t->entry_count == t->entry_capacity)
1250                        root->tree = t = grow_tree_content(t, t->entry_count);
1251                t->entries[t->entry_count++] = e;
1252
1253                e->tree = NULL;
1254                c = get_mode(c, &e->versions[1].mode);
1255                if (!c)
1256                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1257                e->versions[0].mode = e->versions[1].mode;
1258                e->name = to_atom(c, strlen(c));
1259                c += e->name->str_len + 1;
1260                hashcpy(e->versions[0].sha1, (unsigned char*)c);
1261                hashcpy(e->versions[1].sha1, (unsigned char*)c);
1262                c += 20;
1263        }
1264        free(buf);
1265}
1266
1267static int tecmp0 (const void *_a, const void *_b)
1268{
1269        struct tree_entry *a = *((struct tree_entry**)_a);
1270        struct tree_entry *b = *((struct tree_entry**)_b);
1271        return base_name_compare(
1272                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1273                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1274}
1275
1276static int tecmp1 (const void *_a, const void *_b)
1277{
1278        struct tree_entry *a = *((struct tree_entry**)_a);
1279        struct tree_entry *b = *((struct tree_entry**)_b);
1280        return base_name_compare(
1281                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1282                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1283}
1284
1285static void mktree(struct tree_content *t, int v, struct strbuf *b)
1286{
1287        size_t maxlen = 0;
1288        unsigned int i;
1289
1290        if (!v)
1291                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1292        else
1293                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1294
1295        for (i = 0; i < t->entry_count; i++) {
1296                if (t->entries[i]->versions[v].mode)
1297                        maxlen += t->entries[i]->name->str_len + 34;
1298        }
1299
1300        strbuf_reset(b);
1301        strbuf_grow(b, maxlen);
1302        for (i = 0; i < t->entry_count; i++) {
1303                struct tree_entry *e = t->entries[i];
1304                if (!e->versions[v].mode)
1305                        continue;
1306                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1307                                        e->name->str_dat, '\0');
1308                strbuf_add(b, e->versions[v].sha1, 20);
1309        }
1310}
1311
1312static void store_tree(struct tree_entry *root)
1313{
1314        struct tree_content *t = root->tree;
1315        unsigned int i, j, del;
1316        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1317        struct object_entry *le;
1318
1319        if (!is_null_sha1(root->versions[1].sha1))
1320                return;
1321
1322        for (i = 0; i < t->entry_count; i++) {
1323                if (t->entries[i]->tree)
1324                        store_tree(t->entries[i]);
1325        }
1326
1327        le = find_object(root->versions[0].sha1);
1328        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1329                mktree(t, 0, &old_tree);
1330                lo.data = old_tree;
1331                lo.offset = le->offset;
1332                lo.depth = t->delta_depth;
1333        }
1334
1335        mktree(t, 1, &new_tree);
1336        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1337
1338        t->delta_depth = lo.depth;
1339        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1340                struct tree_entry *e = t->entries[i];
1341                if (e->versions[1].mode) {
1342                        e->versions[0].mode = e->versions[1].mode;
1343                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1344                        t->entries[j++] = e;
1345                } else {
1346                        release_tree_entry(e);
1347                        del++;
1348                }
1349        }
1350        t->entry_count -= del;
1351}
1352
1353static int tree_content_set(
1354        struct tree_entry *root,
1355        const char *p,
1356        const unsigned char *sha1,
1357        const uint16_t mode,
1358        struct tree_content *subtree)
1359{
1360        struct tree_content *t = root->tree;
1361        const char *slash1;
1362        unsigned int i, n;
1363        struct tree_entry *e;
1364
1365        slash1 = strchr(p, '/');
1366        if (slash1)
1367                n = slash1 - p;
1368        else
1369                n = strlen(p);
1370        if (!n)
1371                die("Empty path component found in input");
1372        if (!slash1 && !S_ISDIR(mode) && subtree)
1373                die("Non-directories cannot have subtrees");
1374
1375        for (i = 0; i < t->entry_count; i++) {
1376                e = t->entries[i];
1377                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1378                        if (!slash1) {
1379                                if (!S_ISDIR(mode)
1380                                                && e->versions[1].mode == mode
1381                                                && !hashcmp(e->versions[1].sha1, sha1))
1382                                        return 0;
1383                                e->versions[1].mode = mode;
1384                                hashcpy(e->versions[1].sha1, sha1);
1385                                if (e->tree)
1386                                        release_tree_content_recursive(e->tree);
1387                                e->tree = subtree;
1388                                hashclr(root->versions[1].sha1);
1389                                return 1;
1390                        }
1391                        if (!S_ISDIR(e->versions[1].mode)) {
1392                                e->tree = new_tree_content(8);
1393                                e->versions[1].mode = S_IFDIR;
1394                        }
1395                        if (!e->tree)
1396                                load_tree(e);
1397                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1398                                hashclr(root->versions[1].sha1);
1399                                return 1;
1400                        }
1401                        return 0;
1402                }
1403        }
1404
1405        if (t->entry_count == t->entry_capacity)
1406                root->tree = t = grow_tree_content(t, t->entry_count);
1407        e = new_tree_entry();
1408        e->name = to_atom(p, n);
1409        e->versions[0].mode = 0;
1410        hashclr(e->versions[0].sha1);
1411        t->entries[t->entry_count++] = e;
1412        if (slash1) {
1413                e->tree = new_tree_content(8);
1414                e->versions[1].mode = S_IFDIR;
1415                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1416        } else {
1417                e->tree = subtree;
1418                e->versions[1].mode = mode;
1419                hashcpy(e->versions[1].sha1, sha1);
1420        }
1421        hashclr(root->versions[1].sha1);
1422        return 1;
1423}
1424
1425static int tree_content_remove(
1426        struct tree_entry *root,
1427        const char *p,
1428        struct tree_entry *backup_leaf)
1429{
1430        struct tree_content *t = root->tree;
1431        const char *slash1;
1432        unsigned int i, n;
1433        struct tree_entry *e;
1434
1435        slash1 = strchr(p, '/');
1436        if (slash1)
1437                n = slash1 - p;
1438        else
1439                n = strlen(p);
1440
1441        for (i = 0; i < t->entry_count; i++) {
1442                e = t->entries[i];
1443                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1444                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1445                                goto del_entry;
1446                        if (!e->tree)
1447                                load_tree(e);
1448                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1449                                for (n = 0; n < e->tree->entry_count; n++) {
1450                                        if (e->tree->entries[n]->versions[1].mode) {
1451                                                hashclr(root->versions[1].sha1);
1452                                                return 1;
1453                                        }
1454                                }
1455                                backup_leaf = NULL;
1456                                goto del_entry;
1457                        }
1458                        return 0;
1459                }
1460        }
1461        return 0;
1462
1463del_entry:
1464        if (backup_leaf)
1465                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1466        else if (e->tree)
1467                release_tree_content_recursive(e->tree);
1468        e->tree = NULL;
1469        e->versions[1].mode = 0;
1470        hashclr(e->versions[1].sha1);
1471        hashclr(root->versions[1].sha1);
1472        return 1;
1473}
1474
1475static int tree_content_get(
1476        struct tree_entry *root,
1477        const char *p,
1478        struct tree_entry *leaf)
1479{
1480        struct tree_content *t = root->tree;
1481        const char *slash1;
1482        unsigned int i, n;
1483        struct tree_entry *e;
1484
1485        slash1 = strchr(p, '/');
1486        if (slash1)
1487                n = slash1 - p;
1488        else
1489                n = strlen(p);
1490
1491        for (i = 0; i < t->entry_count; i++) {
1492                e = t->entries[i];
1493                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1494                        if (!slash1) {
1495                                memcpy(leaf, e, sizeof(*leaf));
1496                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1497                                        leaf->tree = dup_tree_content(e->tree);
1498                                else
1499                                        leaf->tree = NULL;
1500                                return 1;
1501                        }
1502                        if (!S_ISDIR(e->versions[1].mode))
1503                                return 0;
1504                        if (!e->tree)
1505                                load_tree(e);
1506                        return tree_content_get(e, slash1 + 1, leaf);
1507                }
1508        }
1509        return 0;
1510}
1511
1512static int update_branch(struct branch *b)
1513{
1514        static const char *msg = "fast-import";
1515        struct ref_lock *lock;
1516        unsigned char old_sha1[20];
1517
1518        if (is_null_sha1(b->sha1))
1519                return 0;
1520        if (read_ref(b->name, old_sha1))
1521                hashclr(old_sha1);
1522        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1523        if (!lock)
1524                return error("Unable to lock %s", b->name);
1525        if (!force_update && !is_null_sha1(old_sha1)) {
1526                struct commit *old_cmit, *new_cmit;
1527
1528                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1529                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1530                if (!old_cmit || !new_cmit) {
1531                        unlock_ref(lock);
1532                        return error("Branch %s is missing commits.", b->name);
1533                }
1534
1535                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1536                        unlock_ref(lock);
1537                        warning("Not updating %s"
1538                                " (new tip %s does not contain %s)",
1539                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1540                        return -1;
1541                }
1542        }
1543        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1544                return error("Unable to update %s", b->name);
1545        return 0;
1546}
1547
1548static void dump_branches(void)
1549{
1550        unsigned int i;
1551        struct branch *b;
1552
1553        for (i = 0; i < branch_table_sz; i++) {
1554                for (b = branch_table[i]; b; b = b->table_next_branch)
1555                        failure |= update_branch(b);
1556        }
1557}
1558
1559static void dump_tags(void)
1560{
1561        static const char *msg = "fast-import";
1562        struct tag *t;
1563        struct ref_lock *lock;
1564        char ref_name[PATH_MAX];
1565
1566        for (t = first_tag; t; t = t->next_tag) {
1567                sprintf(ref_name, "tags/%s", t->name);
1568                lock = lock_ref_sha1(ref_name, NULL);
1569                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1570                        failure |= error("Unable to update %s", ref_name);
1571        }
1572}
1573
1574static void dump_marks_helper(FILE *f,
1575        uintmax_t base,
1576        struct mark_set *m)
1577{
1578        uintmax_t k;
1579        if (m->shift) {
1580                for (k = 0; k < 1024; k++) {
1581                        if (m->data.sets[k])
1582                                dump_marks_helper(f, (base + k) << m->shift,
1583                                        m->data.sets[k]);
1584                }
1585        } else {
1586                for (k = 0; k < 1024; k++) {
1587                        if (m->data.marked[k])
1588                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1589                                        sha1_to_hex(m->data.marked[k]->sha1));
1590                }
1591        }
1592}
1593
1594static void dump_marks(void)
1595{
1596        static struct lock_file mark_lock;
1597        int mark_fd;
1598        FILE *f;
1599
1600        if (!mark_file)
1601                return;
1602
1603        mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1604        if (mark_fd < 0) {
1605                failure |= error("Unable to write marks file %s: %s",
1606                        mark_file, strerror(errno));
1607                return;
1608        }
1609
1610        f = fdopen(mark_fd, "w");
1611        if (!f) {
1612                int saved_errno = errno;
1613                rollback_lock_file(&mark_lock);
1614                failure |= error("Unable to write marks file %s: %s",
1615                        mark_file, strerror(saved_errno));
1616                return;
1617        }
1618
1619        /*
1620         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1621         * Assign -1 to the lock file descriptor so that commit_lock_file()
1622         * won't try to close() it.
1623         */
1624        mark_lock.fd = -1;
1625
1626        dump_marks_helper(f, 0, marks);
1627        if (ferror(f) || fclose(f)) {
1628                int saved_errno = errno;
1629                rollback_lock_file(&mark_lock);
1630                failure |= error("Unable to write marks file %s: %s",
1631                        mark_file, strerror(saved_errno));
1632                return;
1633        }
1634
1635        if (commit_lock_file(&mark_lock)) {
1636                int saved_errno = errno;
1637                rollback_lock_file(&mark_lock);
1638                failure |= error("Unable to commit marks file %s: %s",
1639                        mark_file, strerror(saved_errno));
1640                return;
1641        }
1642}
1643
1644static int read_next_command(void)
1645{
1646        static int stdin_eof = 0;
1647
1648        if (stdin_eof) {
1649                unread_command_buf = 0;
1650                return EOF;
1651        }
1652
1653        do {
1654                if (unread_command_buf) {
1655                        unread_command_buf = 0;
1656                } else {
1657                        struct recent_command *rc;
1658
1659                        strbuf_detach(&command_buf, NULL);
1660                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1661                        if (stdin_eof)
1662                                return EOF;
1663
1664                        rc = rc_free;
1665                        if (rc)
1666                                rc_free = rc->next;
1667                        else {
1668                                rc = cmd_hist.next;
1669                                cmd_hist.next = rc->next;
1670                                cmd_hist.next->prev = &cmd_hist;
1671                                free(rc->buf);
1672                        }
1673
1674                        rc->buf = command_buf.buf;
1675                        rc->prev = cmd_tail;
1676                        rc->next = cmd_hist.prev;
1677                        rc->prev->next = rc;
1678                        cmd_tail = rc;
1679                }
1680        } while (command_buf.buf[0] == '#');
1681
1682        return 0;
1683}
1684
1685static void skip_optional_lf(void)
1686{
1687        int term_char = fgetc(stdin);
1688        if (term_char != '\n' && term_char != EOF)
1689                ungetc(term_char, stdin);
1690}
1691
1692static void parse_mark(void)
1693{
1694        if (!prefixcmp(command_buf.buf, "mark :")) {
1695                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1696                read_next_command();
1697        }
1698        else
1699                next_mark = 0;
1700}
1701
1702static void parse_data(struct strbuf *sb)
1703{
1704        strbuf_reset(sb);
1705
1706        if (prefixcmp(command_buf.buf, "data "))
1707                die("Expected 'data n' command, found: %s", command_buf.buf);
1708
1709        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1710                char *term = xstrdup(command_buf.buf + 5 + 2);
1711                size_t term_len = command_buf.len - 5 - 2;
1712
1713                strbuf_detach(&command_buf, NULL);
1714                for (;;) {
1715                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1716                                die("EOF in data (terminator '%s' not found)", term);
1717                        if (term_len == command_buf.len
1718                                && !strcmp(term, command_buf.buf))
1719                                break;
1720                        strbuf_addbuf(sb, &command_buf);
1721                        strbuf_addch(sb, '\n');
1722                }
1723                free(term);
1724        }
1725        else {
1726                size_t n = 0, length;
1727
1728                length = strtoul(command_buf.buf + 5, NULL, 10);
1729
1730                while (n < length) {
1731                        size_t s = strbuf_fread(sb, length - n, stdin);
1732                        if (!s && feof(stdin))
1733                                die("EOF in data (%lu bytes remaining)",
1734                                        (unsigned long)(length - n));
1735                        n += s;
1736                }
1737        }
1738
1739        skip_optional_lf();
1740}
1741
1742static int validate_raw_date(const char *src, char *result, int maxlen)
1743{
1744        const char *orig_src = src;
1745        char *endp, sign;
1746
1747        strtoul(src, &endp, 10);
1748        if (endp == src || *endp != ' ')
1749                return -1;
1750
1751        src = endp + 1;
1752        if (*src != '-' && *src != '+')
1753                return -1;
1754        sign = *src;
1755
1756        strtoul(src + 1, &endp, 10);
1757        if (endp == src || *endp || (endp - orig_src) >= maxlen)
1758                return -1;
1759
1760        strcpy(result, orig_src);
1761        return 0;
1762}
1763
1764static char *parse_ident(const char *buf)
1765{
1766        const char *gt;
1767        size_t name_len;
1768        char *ident;
1769
1770        gt = strrchr(buf, '>');
1771        if (!gt)
1772                die("Missing > in ident string: %s", buf);
1773        gt++;
1774        if (*gt != ' ')
1775                die("Missing space after > in ident string: %s", buf);
1776        gt++;
1777        name_len = gt - buf;
1778        ident = xmalloc(name_len + 24);
1779        strncpy(ident, buf, name_len);
1780
1781        switch (whenspec) {
1782        case WHENSPEC_RAW:
1783                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1784                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1785                break;
1786        case WHENSPEC_RFC2822:
1787                if (parse_date(gt, ident + name_len, 24) < 0)
1788                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1789                break;
1790        case WHENSPEC_NOW:
1791                if (strcmp("now", gt))
1792                        die("Date in ident must be 'now': %s", buf);
1793                datestamp(ident + name_len, 24);
1794                break;
1795        }
1796
1797        return ident;
1798}
1799
1800static void parse_new_blob(void)
1801{
1802        static struct strbuf buf = STRBUF_INIT;
1803
1804        read_next_command();
1805        parse_mark();
1806        parse_data(&buf);
1807        store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1808}
1809
1810static void unload_one_branch(void)
1811{
1812        while (cur_active_branches
1813                && cur_active_branches >= max_active_branches) {
1814                uintmax_t min_commit = ULONG_MAX;
1815                struct branch *e, *l = NULL, *p = NULL;
1816
1817                for (e = active_branches; e; e = e->active_next_branch) {
1818                        if (e->last_commit < min_commit) {
1819                                p = l;
1820                                min_commit = e->last_commit;
1821                        }
1822                        l = e;
1823                }
1824
1825                if (p) {
1826                        e = p->active_next_branch;
1827                        p->active_next_branch = e->active_next_branch;
1828                } else {
1829                        e = active_branches;
1830                        active_branches = e->active_next_branch;
1831                }
1832                e->active = 0;
1833                e->active_next_branch = NULL;
1834                if (e->branch_tree.tree) {
1835                        release_tree_content_recursive(e->branch_tree.tree);
1836                        e->branch_tree.tree = NULL;
1837                }
1838                cur_active_branches--;
1839        }
1840}
1841
1842static void load_branch(struct branch *b)
1843{
1844        load_tree(&b->branch_tree);
1845        if (!b->active) {
1846                b->active = 1;
1847                b->active_next_branch = active_branches;
1848                active_branches = b;
1849                cur_active_branches++;
1850                branch_load_count++;
1851        }
1852}
1853
1854static void file_change_m(struct branch *b)
1855{
1856        const char *p = command_buf.buf + 2;
1857        static struct strbuf uq = STRBUF_INIT;
1858        const char *endp;
1859        struct object_entry *oe = oe;
1860        unsigned char sha1[20];
1861        uint16_t mode, inline_data = 0;
1862
1863        p = get_mode(p, &mode);
1864        if (!p)
1865                die("Corrupt mode: %s", command_buf.buf);
1866        switch (mode) {
1867        case S_IFREG | 0644:
1868        case S_IFREG | 0755:
1869        case S_IFLNK:
1870        case S_IFGITLINK:
1871        case 0644:
1872        case 0755:
1873                /* ok */
1874                break;
1875        default:
1876                die("Corrupt mode: %s", command_buf.buf);
1877        }
1878
1879        if (*p == ':') {
1880                char *x;
1881                oe = find_mark(strtoumax(p + 1, &x, 10));
1882                hashcpy(sha1, oe->sha1);
1883                p = x;
1884        } else if (!prefixcmp(p, "inline")) {
1885                inline_data = 1;
1886                p += 6;
1887        } else {
1888                if (get_sha1_hex(p, sha1))
1889                        die("Invalid SHA1: %s", command_buf.buf);
1890                oe = find_object(sha1);
1891                p += 40;
1892        }
1893        if (*p++ != ' ')
1894                die("Missing space after SHA1: %s", command_buf.buf);
1895
1896        strbuf_reset(&uq);
1897        if (!unquote_c_style(&uq, p, &endp)) {
1898                if (*endp)
1899                        die("Garbage after path in: %s", command_buf.buf);
1900                p = uq.buf;
1901        }
1902
1903        if (S_ISGITLINK(mode)) {
1904                if (inline_data)
1905                        die("Git links cannot be specified 'inline': %s",
1906                                command_buf.buf);
1907                else if (oe) {
1908                        if (oe->type != OBJ_COMMIT)
1909                                die("Not a commit (actually a %s): %s",
1910                                        typename(oe->type), command_buf.buf);
1911                }
1912                /*
1913                 * Accept the sha1 without checking; it expected to be in
1914                 * another repository.
1915                 */
1916        } else if (inline_data) {
1917                static struct strbuf buf = STRBUF_INIT;
1918
1919                if (p != uq.buf) {
1920                        strbuf_addstr(&uq, p);
1921                        p = uq.buf;
1922                }
1923                read_next_command();
1924                parse_data(&buf);
1925                store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1926        } else if (oe) {
1927                if (oe->type != OBJ_BLOB)
1928                        die("Not a blob (actually a %s): %s",
1929                                typename(oe->type), command_buf.buf);
1930        } else {
1931                enum object_type type = sha1_object_info(sha1, NULL);
1932                if (type < 0)
1933                        die("Blob not found: %s", command_buf.buf);
1934                if (type != OBJ_BLOB)
1935                        die("Not a blob (actually a %s): %s",
1936                            typename(type), command_buf.buf);
1937        }
1938
1939        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1940}
1941
1942static void file_change_d(struct branch *b)
1943{
1944        const char *p = command_buf.buf + 2;
1945        static struct strbuf uq = STRBUF_INIT;
1946        const char *endp;
1947
1948        strbuf_reset(&uq);
1949        if (!unquote_c_style(&uq, p, &endp)) {
1950                if (*endp)
1951                        die("Garbage after path in: %s", command_buf.buf);
1952                p = uq.buf;
1953        }
1954        tree_content_remove(&b->branch_tree, p, NULL);
1955}
1956
1957static void file_change_cr(struct branch *b, int rename)
1958{
1959        const char *s, *d;
1960        static struct strbuf s_uq = STRBUF_INIT;
1961        static struct strbuf d_uq = STRBUF_INIT;
1962        const char *endp;
1963        struct tree_entry leaf;
1964
1965        s = command_buf.buf + 2;
1966        strbuf_reset(&s_uq);
1967        if (!unquote_c_style(&s_uq, s, &endp)) {
1968                if (*endp != ' ')
1969                        die("Missing space after source: %s", command_buf.buf);
1970        } else {
1971                endp = strchr(s, ' ');
1972                if (!endp)
1973                        die("Missing space after source: %s", command_buf.buf);
1974                strbuf_add(&s_uq, s, endp - s);
1975        }
1976        s = s_uq.buf;
1977
1978        endp++;
1979        if (!*endp)
1980                die("Missing dest: %s", command_buf.buf);
1981
1982        d = endp;
1983        strbuf_reset(&d_uq);
1984        if (!unquote_c_style(&d_uq, d, &endp)) {
1985                if (*endp)
1986                        die("Garbage after dest in: %s", command_buf.buf);
1987                d = d_uq.buf;
1988        }
1989
1990        memset(&leaf, 0, sizeof(leaf));
1991        if (rename)
1992                tree_content_remove(&b->branch_tree, s, &leaf);
1993        else
1994                tree_content_get(&b->branch_tree, s, &leaf);
1995        if (!leaf.versions[1].mode)
1996                die("Path %s not in branch", s);
1997        tree_content_set(&b->branch_tree, d,
1998                leaf.versions[1].sha1,
1999                leaf.versions[1].mode,
2000                leaf.tree);
2001}
2002
2003static void file_change_deleteall(struct branch *b)
2004{
2005        release_tree_content_recursive(b->branch_tree.tree);
2006        hashclr(b->branch_tree.versions[0].sha1);
2007        hashclr(b->branch_tree.versions[1].sha1);
2008        load_tree(&b->branch_tree);
2009}
2010
2011static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2012{
2013        if (!buf || size < 46)
2014                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2015        if (memcmp("tree ", buf, 5)
2016                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2017                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2018        hashcpy(b->branch_tree.versions[0].sha1,
2019                b->branch_tree.versions[1].sha1);
2020}
2021
2022static void parse_from_existing(struct branch *b)
2023{
2024        if (is_null_sha1(b->sha1)) {
2025                hashclr(b->branch_tree.versions[0].sha1);
2026                hashclr(b->branch_tree.versions[1].sha1);
2027        } else {
2028                unsigned long size;
2029                char *buf;
2030
2031                buf = read_object_with_reference(b->sha1,
2032                        commit_type, &size, b->sha1);
2033                parse_from_commit(b, buf, size);
2034                free(buf);
2035        }
2036}
2037
2038static int parse_from(struct branch *b)
2039{
2040        const char *from;
2041        struct branch *s;
2042
2043        if (prefixcmp(command_buf.buf, "from "))
2044                return 0;
2045
2046        if (b->branch_tree.tree) {
2047                release_tree_content_recursive(b->branch_tree.tree);
2048                b->branch_tree.tree = NULL;
2049        }
2050
2051        from = strchr(command_buf.buf, ' ') + 1;
2052        s = lookup_branch(from);
2053        if (b == s)
2054                die("Can't create a branch from itself: %s", b->name);
2055        else if (s) {
2056                unsigned char *t = s->branch_tree.versions[1].sha1;
2057                hashcpy(b->sha1, s->sha1);
2058                hashcpy(b->branch_tree.versions[0].sha1, t);
2059                hashcpy(b->branch_tree.versions[1].sha1, t);
2060        } else if (*from == ':') {
2061                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2062                struct object_entry *oe = find_mark(idnum);
2063                if (oe->type != OBJ_COMMIT)
2064                        die("Mark :%" PRIuMAX " not a commit", idnum);
2065                hashcpy(b->sha1, oe->sha1);
2066                if (oe->pack_id != MAX_PACK_ID) {
2067                        unsigned long size;
2068                        char *buf = gfi_unpack_entry(oe, &size);
2069                        parse_from_commit(b, buf, size);
2070                        free(buf);
2071                } else
2072                        parse_from_existing(b);
2073        } else if (!get_sha1(from, b->sha1))
2074                parse_from_existing(b);
2075        else
2076                die("Invalid ref name or SHA1 expression: %s", from);
2077
2078        read_next_command();
2079        return 1;
2080}
2081
2082static struct hash_list *parse_merge(unsigned int *count)
2083{
2084        struct hash_list *list = NULL, *n, *e = e;
2085        const char *from;
2086        struct branch *s;
2087
2088        *count = 0;
2089        while (!prefixcmp(command_buf.buf, "merge ")) {
2090                from = strchr(command_buf.buf, ' ') + 1;
2091                n = xmalloc(sizeof(*n));
2092                s = lookup_branch(from);
2093                if (s)
2094                        hashcpy(n->sha1, s->sha1);
2095                else if (*from == ':') {
2096                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2097                        struct object_entry *oe = find_mark(idnum);
2098                        if (oe->type != OBJ_COMMIT)
2099                                die("Mark :%" PRIuMAX " not a commit", idnum);
2100                        hashcpy(n->sha1, oe->sha1);
2101                } else if (!get_sha1(from, n->sha1)) {
2102                        unsigned long size;
2103                        char *buf = read_object_with_reference(n->sha1,
2104                                commit_type, &size, n->sha1);
2105                        if (!buf || size < 46)
2106                                die("Not a valid commit: %s", from);
2107                        free(buf);
2108                } else
2109                        die("Invalid ref name or SHA1 expression: %s", from);
2110
2111                n->next = NULL;
2112                if (list)
2113                        e->next = n;
2114                else
2115                        list = n;
2116                e = n;
2117                (*count)++;
2118                read_next_command();
2119        }
2120        return list;
2121}
2122
2123static void parse_new_commit(void)
2124{
2125        static struct strbuf msg = STRBUF_INIT;
2126        struct branch *b;
2127        char *sp;
2128        char *author = NULL;
2129        char *committer = NULL;
2130        struct hash_list *merge_list = NULL;
2131        unsigned int merge_count;
2132
2133        /* Obtain the branch name from the rest of our command */
2134        sp = strchr(command_buf.buf, ' ') + 1;
2135        b = lookup_branch(sp);
2136        if (!b)
2137                b = new_branch(sp);
2138
2139        read_next_command();
2140        parse_mark();
2141        if (!prefixcmp(command_buf.buf, "author ")) {
2142                author = parse_ident(command_buf.buf + 7);
2143                read_next_command();
2144        }
2145        if (!prefixcmp(command_buf.buf, "committer ")) {
2146                committer = parse_ident(command_buf.buf + 10);
2147                read_next_command();
2148        }
2149        if (!committer)
2150                die("Expected committer but didn't get one");
2151        parse_data(&msg);
2152        read_next_command();
2153        parse_from(b);
2154        merge_list = parse_merge(&merge_count);
2155
2156        /* ensure the branch is active/loaded */
2157        if (!b->branch_tree.tree || !max_active_branches) {
2158                unload_one_branch();
2159                load_branch(b);
2160        }
2161
2162        /* file_change* */
2163        while (command_buf.len > 0) {
2164                if (!prefixcmp(command_buf.buf, "M "))
2165                        file_change_m(b);
2166                else if (!prefixcmp(command_buf.buf, "D "))
2167                        file_change_d(b);
2168                else if (!prefixcmp(command_buf.buf, "R "))
2169                        file_change_cr(b, 1);
2170                else if (!prefixcmp(command_buf.buf, "C "))
2171                        file_change_cr(b, 0);
2172                else if (!strcmp("deleteall", command_buf.buf))
2173                        file_change_deleteall(b);
2174                else {
2175                        unread_command_buf = 1;
2176                        break;
2177                }
2178                if (read_next_command() == EOF)
2179                        break;
2180        }
2181
2182        /* build the tree and the commit */
2183        store_tree(&b->branch_tree);
2184        hashcpy(b->branch_tree.versions[0].sha1,
2185                b->branch_tree.versions[1].sha1);
2186
2187        strbuf_reset(&new_data);
2188        strbuf_addf(&new_data, "tree %s\n",
2189                sha1_to_hex(b->branch_tree.versions[1].sha1));
2190        if (!is_null_sha1(b->sha1))
2191                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2192        while (merge_list) {
2193                struct hash_list *next = merge_list->next;
2194                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2195                free(merge_list);
2196                merge_list = next;
2197        }
2198        strbuf_addf(&new_data,
2199                "author %s\n"
2200                "committer %s\n"
2201                "\n",
2202                author ? author : committer, committer);
2203        strbuf_addbuf(&new_data, &msg);
2204        free(author);
2205        free(committer);
2206
2207        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2208                b->pack_id = pack_id;
2209        b->last_commit = object_count_by_type[OBJ_COMMIT];
2210}
2211
2212static void parse_new_tag(void)
2213{
2214        static struct strbuf msg = STRBUF_INIT;
2215        char *sp;
2216        const char *from;
2217        char *tagger;
2218        struct branch *s;
2219        struct tag *t;
2220        uintmax_t from_mark = 0;
2221        unsigned char sha1[20];
2222
2223        /* Obtain the new tag name from the rest of our command */
2224        sp = strchr(command_buf.buf, ' ') + 1;
2225        t = pool_alloc(sizeof(struct tag));
2226        t->next_tag = NULL;
2227        t->name = pool_strdup(sp);
2228        if (last_tag)
2229                last_tag->next_tag = t;
2230        else
2231                first_tag = t;
2232        last_tag = t;
2233        read_next_command();
2234
2235        /* from ... */
2236        if (prefixcmp(command_buf.buf, "from "))
2237                die("Expected from command, got %s", command_buf.buf);
2238        from = strchr(command_buf.buf, ' ') + 1;
2239        s = lookup_branch(from);
2240        if (s) {
2241                hashcpy(sha1, s->sha1);
2242        } else if (*from == ':') {
2243                struct object_entry *oe;
2244                from_mark = strtoumax(from + 1, NULL, 10);
2245                oe = find_mark(from_mark);
2246                if (oe->type != OBJ_COMMIT)
2247                        die("Mark :%" PRIuMAX " not a commit", from_mark);
2248                hashcpy(sha1, oe->sha1);
2249        } else if (!get_sha1(from, sha1)) {
2250                unsigned long size;
2251                char *buf;
2252
2253                buf = read_object_with_reference(sha1,
2254                        commit_type, &size, sha1);
2255                if (!buf || size < 46)
2256                        die("Not a valid commit: %s", from);
2257                free(buf);
2258        } else
2259                die("Invalid ref name or SHA1 expression: %s", from);
2260        read_next_command();
2261
2262        /* tagger ... */
2263        if (!prefixcmp(command_buf.buf, "tagger ")) {
2264                tagger = parse_ident(command_buf.buf + 7);
2265                read_next_command();
2266        } else
2267                tagger = NULL;
2268
2269        /* tag payload/message */
2270        parse_data(&msg);
2271
2272        /* build the tag object */
2273        strbuf_reset(&new_data);
2274
2275        strbuf_addf(&new_data,
2276                    "object %s\n"
2277                    "type %s\n"
2278                    "tag %s\n",
2279                    sha1_to_hex(sha1), commit_type, t->name);
2280        if (tagger)
2281                strbuf_addf(&new_data,
2282                            "tagger %s\n", tagger);
2283        strbuf_addch(&new_data, '\n');
2284        strbuf_addbuf(&new_data, &msg);
2285        free(tagger);
2286
2287        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2288                t->pack_id = MAX_PACK_ID;
2289        else
2290                t->pack_id = pack_id;
2291}
2292
2293static void parse_reset_branch(void)
2294{
2295        struct branch *b;
2296        char *sp;
2297
2298        /* Obtain the branch name from the rest of our command */
2299        sp = strchr(command_buf.buf, ' ') + 1;
2300        b = lookup_branch(sp);
2301        if (b) {
2302                hashclr(b->sha1);
2303                hashclr(b->branch_tree.versions[0].sha1);
2304                hashclr(b->branch_tree.versions[1].sha1);
2305                if (b->branch_tree.tree) {
2306                        release_tree_content_recursive(b->branch_tree.tree);
2307                        b->branch_tree.tree = NULL;
2308                }
2309        }
2310        else
2311                b = new_branch(sp);
2312        read_next_command();
2313        parse_from(b);
2314        if (command_buf.len > 0)
2315                unread_command_buf = 1;
2316}
2317
2318static void parse_checkpoint(void)
2319{
2320        if (object_count) {
2321                cycle_packfile();
2322                dump_branches();
2323                dump_tags();
2324                dump_marks();
2325        }
2326        skip_optional_lf();
2327}
2328
2329static void parse_progress(void)
2330{
2331        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2332        fputc('\n', stdout);
2333        fflush(stdout);
2334        skip_optional_lf();
2335}
2336
2337static void import_marks(const char *input_file)
2338{
2339        char line[512];
2340        FILE *f = fopen(input_file, "r");
2341        if (!f)
2342                die("cannot read %s: %s", input_file, strerror(errno));
2343        while (fgets(line, sizeof(line), f)) {
2344                uintmax_t mark;
2345                char *end;
2346                unsigned char sha1[20];
2347                struct object_entry *e;
2348
2349                end = strchr(line, '\n');
2350                if (line[0] != ':' || !end)
2351                        die("corrupt mark line: %s", line);
2352                *end = 0;
2353                mark = strtoumax(line + 1, &end, 10);
2354                if (!mark || end == line + 1
2355                        || *end != ' ' || get_sha1(end + 1, sha1))
2356                        die("corrupt mark line: %s", line);
2357                e = find_object(sha1);
2358                if (!e) {
2359                        enum object_type type = sha1_object_info(sha1, NULL);
2360                        if (type < 0)
2361                                die("object not found: %s", sha1_to_hex(sha1));
2362                        e = insert_object(sha1);
2363                        e->type = type;
2364                        e->pack_id = MAX_PACK_ID;
2365                        e->offset = 1; /* just not zero! */
2366                }
2367                insert_mark(mark, e);
2368        }
2369        fclose(f);
2370}
2371
2372static int git_pack_config(const char *k, const char *v, void *cb)
2373{
2374        if (!strcmp(k, "pack.depth")) {
2375                max_depth = git_config_int(k, v);
2376                if (max_depth > MAX_DEPTH)
2377                        max_depth = MAX_DEPTH;
2378                return 0;
2379        }
2380        if (!strcmp(k, "pack.compression")) {
2381                int level = git_config_int(k, v);
2382                if (level == -1)
2383                        level = Z_DEFAULT_COMPRESSION;
2384                else if (level < 0 || level > Z_BEST_COMPRESSION)
2385                        die("bad pack compression level %d", level);
2386                pack_compression_level = level;
2387                pack_compression_seen = 1;
2388                return 0;
2389        }
2390        return git_default_config(k, v, cb);
2391}
2392
2393static const char fast_import_usage[] =
2394"git fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2395
2396int main(int argc, const char **argv)
2397{
2398        unsigned int i, show_stats = 1;
2399
2400        setup_git_directory();
2401        git_config(git_pack_config, NULL);
2402        if (!pack_compression_seen && core_compression_seen)
2403                pack_compression_level = core_compression_level;
2404
2405        alloc_objects(object_entry_alloc);
2406        strbuf_init(&command_buf, 0);
2407        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2408        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2409        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2410        marks = pool_calloc(1, sizeof(struct mark_set));
2411
2412        for (i = 1; i < argc; i++) {
2413                const char *a = argv[i];
2414
2415                if (*a != '-' || !strcmp(a, "--"))
2416                        break;
2417                else if (!prefixcmp(a, "--date-format=")) {
2418                        const char *fmt = a + 14;
2419                        if (!strcmp(fmt, "raw"))
2420                                whenspec = WHENSPEC_RAW;
2421                        else if (!strcmp(fmt, "rfc2822"))
2422                                whenspec = WHENSPEC_RFC2822;
2423                        else if (!strcmp(fmt, "now"))
2424                                whenspec = WHENSPEC_NOW;
2425                        else
2426                                die("unknown --date-format argument %s", fmt);
2427                }
2428                else if (!prefixcmp(a, "--max-pack-size="))
2429                        max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2430                else if (!prefixcmp(a, "--depth=")) {
2431                        max_depth = strtoul(a + 8, NULL, 0);
2432                        if (max_depth > MAX_DEPTH)
2433                                die("--depth cannot exceed %u", MAX_DEPTH);
2434                }
2435                else if (!prefixcmp(a, "--active-branches="))
2436                        max_active_branches = strtoul(a + 18, NULL, 0);
2437                else if (!prefixcmp(a, "--import-marks="))
2438                        import_marks(a + 15);
2439                else if (!prefixcmp(a, "--export-marks="))
2440                        mark_file = a + 15;
2441                else if (!prefixcmp(a, "--export-pack-edges=")) {
2442                        if (pack_edges)
2443                                fclose(pack_edges);
2444                        pack_edges = fopen(a + 20, "a");
2445                        if (!pack_edges)
2446                                die("Cannot open %s: %s", a + 20, strerror(errno));
2447                } else if (!strcmp(a, "--force"))
2448                        force_update = 1;
2449                else if (!strcmp(a, "--quiet"))
2450                        show_stats = 0;
2451                else if (!strcmp(a, "--stats"))
2452                        show_stats = 1;
2453                else
2454                        die("unknown option %s", a);
2455        }
2456        if (i != argc)
2457                usage(fast_import_usage);
2458
2459        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2460        for (i = 0; i < (cmd_save - 1); i++)
2461                rc_free[i].next = &rc_free[i + 1];
2462        rc_free[cmd_save - 1].next = NULL;
2463
2464        prepare_packed_git();
2465        start_packfile();
2466        set_die_routine(die_nicely);
2467        while (read_next_command() != EOF) {
2468                if (!strcmp("blob", command_buf.buf))
2469                        parse_new_blob();
2470                else if (!prefixcmp(command_buf.buf, "commit "))
2471                        parse_new_commit();
2472                else if (!prefixcmp(command_buf.buf, "tag "))
2473                        parse_new_tag();
2474                else if (!prefixcmp(command_buf.buf, "reset "))
2475                        parse_reset_branch();
2476                else if (!strcmp("checkpoint", command_buf.buf))
2477                        parse_checkpoint();
2478                else if (!prefixcmp(command_buf.buf, "progress "))
2479                        parse_progress();
2480                else
2481                        die("Unsupported command: %s", command_buf.buf);
2482        }
2483        end_packfile();
2484
2485        dump_branches();
2486        dump_tags();
2487        unkeep_all_packs();
2488        dump_marks();
2489
2490        if (pack_edges)
2491                fclose(pack_edges);
2492
2493        if (show_stats) {
2494                uintmax_t total_count = 0, duplicate_count = 0;
2495                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2496                        total_count += object_count_by_type[i];
2497                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2498                        duplicate_count += duplicate_count_by_type[i];
2499
2500                fprintf(stderr, "%s statistics:\n", argv[0]);
2501                fprintf(stderr, "---------------------------------------------------------------------\n");
2502                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2503                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2504                fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
2505                fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
2506                fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
2507                fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
2508                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2509                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2510                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2511                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2512                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2513                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2514                fprintf(stderr, "---------------------------------------------------------------------\n");
2515                pack_report();
2516                fprintf(stderr, "---------------------------------------------------------------------\n");
2517                fprintf(stderr, "\n");
2518        }
2519
2520        return failure ? 1 : 0;
2521}