f6872fe23db5292d2a15008881303ecb4a9e9a68
   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 write_crash_report(const char *err)
 376{
 377        char *loc = git_path("fast_import_crash_%d", getpid());
 378        FILE *rpt = fopen(loc, "w");
 379        struct branch *b;
 380        unsigned long lu;
 381        struct recent_command *rc;
 382
 383        if (!rpt) {
 384                error("can't write crash report %s: %s", loc, strerror(errno));
 385                return;
 386        }
 387
 388        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 389
 390        fprintf(rpt, "fast-import crash report:\n");
 391        fprintf(rpt, "    fast-import process: %d\n", getpid());
 392        fprintf(rpt, "    parent process     : %d\n", getppid());
 393        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 394        fputc('\n', rpt);
 395
 396        fputs("fatal: ", rpt);
 397        fputs(err, rpt);
 398        fputc('\n', rpt);
 399
 400        fputc('\n', rpt);
 401        fputs("Most Recent Commands Before Crash\n", rpt);
 402        fputs("---------------------------------\n", rpt);
 403        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 404                if (rc->next == &cmd_hist)
 405                        fputs("* ", rpt);
 406                else
 407                        fputs("  ", rpt);
 408                fputs(rc->buf, rpt);
 409                fputc('\n', rpt);
 410        }
 411
 412        fputc('\n', rpt);
 413        fputs("Active Branch LRU\n", rpt);
 414        fputs("-----------------\n", rpt);
 415        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 416                cur_active_branches,
 417                max_active_branches);
 418        fputc('\n', rpt);
 419        fputs("  pos  clock name\n", rpt);
 420        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 421        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 422                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 423                        ++lu, b->last_commit, b->name);
 424
 425        fputc('\n', rpt);
 426        fputs("Inactive Branches\n", rpt);
 427        fputs("-----------------\n", rpt);
 428        for (lu = 0; lu < branch_table_sz; lu++) {
 429                for (b = branch_table[lu]; b; b = b->table_next_branch)
 430                        write_branch_report(rpt, b);
 431        }
 432
 433        fputc('\n', rpt);
 434        fputs("-------------------\n", rpt);
 435        fputs("END OF CRASH REPORT\n", rpt);
 436        fclose(rpt);
 437}
 438
 439static NORETURN void die_nicely(const char *err, va_list params)
 440{
 441        static int zombie;
 442        char message[2 * PATH_MAX];
 443
 444        vsnprintf(message, sizeof(message), err, params);
 445        fputs("fatal: ", stderr);
 446        fputs(message, stderr);
 447        fputc('\n', stderr);
 448
 449        if (!zombie) {
 450                zombie = 1;
 451                write_crash_report(message);
 452        }
 453        exit(128);
 454}
 455
 456static void alloc_objects(unsigned int cnt)
 457{
 458        struct object_entry_pool *b;
 459
 460        b = xmalloc(sizeof(struct object_entry_pool)
 461                + cnt * sizeof(struct object_entry));
 462        b->next_pool = blocks;
 463        b->next_free = b->entries;
 464        b->end = b->entries + cnt;
 465        blocks = b;
 466        alloc_count += cnt;
 467}
 468
 469static struct object_entry *new_object(unsigned char *sha1)
 470{
 471        struct object_entry *e;
 472
 473        if (blocks->next_free == blocks->end)
 474                alloc_objects(object_entry_alloc);
 475
 476        e = blocks->next_free++;
 477        hashcpy(e->sha1, sha1);
 478        return e;
 479}
 480
 481static struct object_entry *find_object(unsigned char *sha1)
 482{
 483        unsigned int h = sha1[0] << 8 | sha1[1];
 484        struct object_entry *e;
 485        for (e = object_table[h]; e; e = e->next)
 486                if (!hashcmp(sha1, e->sha1))
 487                        return e;
 488        return NULL;
 489}
 490
 491static struct object_entry *insert_object(unsigned char *sha1)
 492{
 493        unsigned int h = sha1[0] << 8 | sha1[1];
 494        struct object_entry *e = object_table[h];
 495        struct object_entry *p = NULL;
 496
 497        while (e) {
 498                if (!hashcmp(sha1, e->sha1))
 499                        return e;
 500                p = e;
 501                e = e->next;
 502        }
 503
 504        e = new_object(sha1);
 505        e->next = NULL;
 506        e->offset = 0;
 507        if (p)
 508                p->next = e;
 509        else
 510                object_table[h] = e;
 511        return e;
 512}
 513
 514static unsigned int hc_str(const char *s, size_t len)
 515{
 516        unsigned int r = 0;
 517        while (len-- > 0)
 518                r = r * 31 + *s++;
 519        return r;
 520}
 521
 522static void *pool_alloc(size_t len)
 523{
 524        struct mem_pool *p;
 525        void *r;
 526
 527        for (p = mem_pool; p; p = p->next_pool)
 528                if ((p->end - p->next_free >= len))
 529                        break;
 530
 531        if (!p) {
 532                if (len >= (mem_pool_alloc/2)) {
 533                        total_allocd += len;
 534                        return xmalloc(len);
 535                }
 536                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 537                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 538                p->next_pool = mem_pool;
 539                p->next_free = (char *) p->space;
 540                p->end = p->next_free + mem_pool_alloc;
 541                mem_pool = p;
 542        }
 543
 544        r = p->next_free;
 545        /* round out to a 'uintmax_t' alignment */
 546        if (len & (sizeof(uintmax_t) - 1))
 547                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 548        p->next_free += len;
 549        return r;
 550}
 551
 552static void *pool_calloc(size_t count, size_t size)
 553{
 554        size_t len = count * size;
 555        void *r = pool_alloc(len);
 556        memset(r, 0, len);
 557        return r;
 558}
 559
 560static char *pool_strdup(const char *s)
 561{
 562        char *r = pool_alloc(strlen(s) + 1);
 563        strcpy(r, s);
 564        return r;
 565}
 566
 567static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 568{
 569        struct mark_set *s = marks;
 570        while ((idnum >> s->shift) >= 1024) {
 571                s = pool_calloc(1, sizeof(struct mark_set));
 572                s->shift = marks->shift + 10;
 573                s->data.sets[0] = marks;
 574                marks = s;
 575        }
 576        while (s->shift) {
 577                uintmax_t i = idnum >> s->shift;
 578                idnum -= i << s->shift;
 579                if (!s->data.sets[i]) {
 580                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 581                        s->data.sets[i]->shift = s->shift - 10;
 582                }
 583                s = s->data.sets[i];
 584        }
 585        if (!s->data.marked[idnum])
 586                marks_set_count++;
 587        s->data.marked[idnum] = oe;
 588}
 589
 590static struct object_entry *find_mark(uintmax_t idnum)
 591{
 592        uintmax_t orig_idnum = idnum;
 593        struct mark_set *s = marks;
 594        struct object_entry *oe = NULL;
 595        if ((idnum >> s->shift) < 1024) {
 596                while (s && s->shift) {
 597                        uintmax_t i = idnum >> s->shift;
 598                        idnum -= i << s->shift;
 599                        s = s->data.sets[i];
 600                }
 601                if (s)
 602                        oe = s->data.marked[idnum];
 603        }
 604        if (!oe)
 605                die("mark :%" PRIuMAX " not declared", orig_idnum);
 606        return oe;
 607}
 608
 609static struct atom_str *to_atom(const char *s, unsigned short len)
 610{
 611        unsigned int hc = hc_str(s, len) % atom_table_sz;
 612        struct atom_str *c;
 613
 614        for (c = atom_table[hc]; c; c = c->next_atom)
 615                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 616                        return c;
 617
 618        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 619        c->str_len = len;
 620        strncpy(c->str_dat, s, len);
 621        c->str_dat[len] = 0;
 622        c->next_atom = atom_table[hc];
 623        atom_table[hc] = c;
 624        atom_cnt++;
 625        return c;
 626}
 627
 628static struct branch *lookup_branch(const char *name)
 629{
 630        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 631        struct branch *b;
 632
 633        for (b = branch_table[hc]; b; b = b->table_next_branch)
 634                if (!strcmp(name, b->name))
 635                        return b;
 636        return NULL;
 637}
 638
 639static struct branch *new_branch(const char *name)
 640{
 641        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 642        struct branch* b = lookup_branch(name);
 643
 644        if (b)
 645                die("Invalid attempt to create duplicate branch: %s", name);
 646        switch (check_ref_format(name)) {
 647        case 0: break; /* its valid */
 648        case CHECK_REF_FORMAT_ONELEVEL:
 649                break; /* valid, but too few '/', allow anyway */
 650        default:
 651                die("Branch name doesn't conform to GIT standards: %s", name);
 652        }
 653
 654        b = pool_calloc(1, sizeof(struct branch));
 655        b->name = pool_strdup(name);
 656        b->table_next_branch = branch_table[hc];
 657        b->branch_tree.versions[0].mode = S_IFDIR;
 658        b->branch_tree.versions[1].mode = S_IFDIR;
 659        b->active = 0;
 660        b->pack_id = MAX_PACK_ID;
 661        branch_table[hc] = b;
 662        branch_count++;
 663        return b;
 664}
 665
 666static unsigned int hc_entries(unsigned int cnt)
 667{
 668        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 669        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 670}
 671
 672static struct tree_content *new_tree_content(unsigned int cnt)
 673{
 674        struct avail_tree_content *f, *l = NULL;
 675        struct tree_content *t;
 676        unsigned int hc = hc_entries(cnt);
 677
 678        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 679                if (f->entry_capacity >= cnt)
 680                        break;
 681
 682        if (f) {
 683                if (l)
 684                        l->next_avail = f->next_avail;
 685                else
 686                        avail_tree_table[hc] = f->next_avail;
 687        } else {
 688                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 689                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 690                f->entry_capacity = cnt;
 691        }
 692
 693        t = (struct tree_content*)f;
 694        t->entry_count = 0;
 695        t->delta_depth = 0;
 696        return t;
 697}
 698
 699static void release_tree_entry(struct tree_entry *e);
 700static void release_tree_content(struct tree_content *t)
 701{
 702        struct avail_tree_content *f = (struct avail_tree_content*)t;
 703        unsigned int hc = hc_entries(f->entry_capacity);
 704        f->next_avail = avail_tree_table[hc];
 705        avail_tree_table[hc] = f;
 706}
 707
 708static void release_tree_content_recursive(struct tree_content *t)
 709{
 710        unsigned int i;
 711        for (i = 0; i < t->entry_count; i++)
 712                release_tree_entry(t->entries[i]);
 713        release_tree_content(t);
 714}
 715
 716static struct tree_content *grow_tree_content(
 717        struct tree_content *t,
 718        int amt)
 719{
 720        struct tree_content *r = new_tree_content(t->entry_count + amt);
 721        r->entry_count = t->entry_count;
 722        r->delta_depth = t->delta_depth;
 723        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 724        release_tree_content(t);
 725        return r;
 726}
 727
 728static struct tree_entry *new_tree_entry(void)
 729{
 730        struct tree_entry *e;
 731
 732        if (!avail_tree_entry) {
 733                unsigned int n = tree_entry_alloc;
 734                total_allocd += n * sizeof(struct tree_entry);
 735                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 736                while (n-- > 1) {
 737                        *((void**)e) = e + 1;
 738                        e++;
 739                }
 740                *((void**)e) = NULL;
 741        }
 742
 743        e = avail_tree_entry;
 744        avail_tree_entry = *((void**)e);
 745        return e;
 746}
 747
 748static void release_tree_entry(struct tree_entry *e)
 749{
 750        if (e->tree)
 751                release_tree_content_recursive(e->tree);
 752        *((void**)e) = avail_tree_entry;
 753        avail_tree_entry = e;
 754}
 755
 756static struct tree_content *dup_tree_content(struct tree_content *s)
 757{
 758        struct tree_content *d;
 759        struct tree_entry *a, *b;
 760        unsigned int i;
 761
 762        if (!s)
 763                return NULL;
 764        d = new_tree_content(s->entry_count);
 765        for (i = 0; i < s->entry_count; i++) {
 766                a = s->entries[i];
 767                b = new_tree_entry();
 768                memcpy(b, a, sizeof(*a));
 769                if (a->tree && is_null_sha1(b->versions[1].sha1))
 770                        b->tree = dup_tree_content(a->tree);
 771                else
 772                        b->tree = NULL;
 773                d->entries[i] = b;
 774        }
 775        d->entry_count = s->entry_count;
 776        d->delta_depth = s->delta_depth;
 777
 778        return d;
 779}
 780
 781static void start_packfile(void)
 782{
 783        static char tmpfile[PATH_MAX];
 784        struct packed_git *p;
 785        struct pack_header hdr;
 786        int pack_fd;
 787
 788        snprintf(tmpfile, sizeof(tmpfile),
 789                "%s/tmp_pack_XXXXXX", get_object_directory());
 790        pack_fd = xmkstemp(tmpfile);
 791        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 792        strcpy(p->pack_name, tmpfile);
 793        p->pack_fd = pack_fd;
 794
 795        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 796        hdr.hdr_version = htonl(2);
 797        hdr.hdr_entries = 0;
 798        write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 799
 800        pack_data = p;
 801        pack_size = sizeof(hdr);
 802        object_count = 0;
 803
 804        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 805        all_packs[pack_id] = p;
 806}
 807
 808static int oecmp (const void *a_, const void *b_)
 809{
 810        struct object_entry *a = *((struct object_entry**)a_);
 811        struct object_entry *b = *((struct object_entry**)b_);
 812        return hashcmp(a->sha1, b->sha1);
 813}
 814
 815static char *create_index(void)
 816{
 817        static char tmpfile[PATH_MAX];
 818        SHA_CTX ctx;
 819        struct sha1file *f;
 820        struct object_entry **idx, **c, **last, *e;
 821        struct object_entry_pool *o;
 822        uint32_t array[256];
 823        int i, idx_fd;
 824
 825        /* Build the sorted table of object IDs. */
 826        idx = xmalloc(object_count * sizeof(struct object_entry*));
 827        c = idx;
 828        for (o = blocks; o; o = o->next_pool)
 829                for (e = o->next_free; e-- != o->entries;)
 830                        if (pack_id == e->pack_id)
 831                                *c++ = e;
 832        last = idx + object_count;
 833        if (c != last)
 834                die("internal consistency error creating the index");
 835        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 836
 837        /* Generate the fan-out array. */
 838        c = idx;
 839        for (i = 0; i < 256; i++) {
 840                struct object_entry **next = c;;
 841                while (next < last) {
 842                        if ((*next)->sha1[0] != i)
 843                                break;
 844                        next++;
 845                }
 846                array[i] = htonl(next - idx);
 847                c = next;
 848        }
 849
 850        snprintf(tmpfile, sizeof(tmpfile),
 851                "%s/tmp_idx_XXXXXX", get_object_directory());
 852        idx_fd = xmkstemp(tmpfile);
 853        f = sha1fd(idx_fd, tmpfile);
 854        sha1write(f, array, 256 * sizeof(int));
 855        SHA1_Init(&ctx);
 856        for (c = idx; c != last; c++) {
 857                uint32_t offset = htonl((*c)->offset);
 858                sha1write(f, &offset, 4);
 859                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 860                SHA1_Update(&ctx, (*c)->sha1, 20);
 861        }
 862        sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
 863        sha1close(f, NULL, 1);
 864        free(idx);
 865        SHA1_Final(pack_data->sha1, &ctx);
 866        return tmpfile;
 867}
 868
 869static char *keep_pack(char *curr_index_name)
 870{
 871        static char name[PATH_MAX];
 872        static const char *keep_msg = "fast-import";
 873        int keep_fd;
 874
 875        chmod(pack_data->pack_name, 0444);
 876        chmod(curr_index_name, 0444);
 877
 878        snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 879                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 880        keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
 881        if (keep_fd < 0)
 882                die("cannot create keep file");
 883        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 884        if (close(keep_fd))
 885                die("failed to write keep file");
 886
 887        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 888                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 889        if (move_temp_to_file(pack_data->pack_name, name))
 890                die("cannot store pack file");
 891
 892        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 893                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 894        if (move_temp_to_file(curr_index_name, name))
 895                die("cannot store index file");
 896        return name;
 897}
 898
 899static void unkeep_all_packs(void)
 900{
 901        static char name[PATH_MAX];
 902        int k;
 903
 904        for (k = 0; k < pack_id; k++) {
 905                struct packed_git *p = all_packs[k];
 906                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 907                         get_object_directory(), sha1_to_hex(p->sha1));
 908                unlink(name);
 909        }
 910}
 911
 912static void end_packfile(void)
 913{
 914        struct packed_git *old_p = pack_data, *new_p;
 915
 916        if (object_count) {
 917                char *idx_name;
 918                int i;
 919                struct branch *b;
 920                struct tag *t;
 921
 922                close_pack_windows(pack_data);
 923                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 924                                    pack_data->pack_name, object_count);
 925                close(pack_data->pack_fd);
 926                idx_name = keep_pack(create_index());
 927
 928                /* Register the packfile with core git's machinary. */
 929                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 930                if (!new_p)
 931                        die("core git rejected index %s", idx_name);
 932                all_packs[pack_id] = new_p;
 933                install_packed_git(new_p);
 934
 935                /* Print the boundary */
 936                if (pack_edges) {
 937                        fprintf(pack_edges, "%s:", new_p->pack_name);
 938                        for (i = 0; i < branch_table_sz; i++) {
 939                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 940                                        if (b->pack_id == pack_id)
 941                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 942                                }
 943                        }
 944                        for (t = first_tag; t; t = t->next_tag) {
 945                                if (t->pack_id == pack_id)
 946                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 947                        }
 948                        fputc('\n', pack_edges);
 949                        fflush(pack_edges);
 950                }
 951
 952                pack_id++;
 953        }
 954        else
 955                unlink(old_p->pack_name);
 956        free(old_p);
 957
 958        /* We can't carry a delta across packfiles. */
 959        strbuf_release(&last_blob.data);
 960        last_blob.offset = 0;
 961        last_blob.depth = 0;
 962}
 963
 964static void cycle_packfile(void)
 965{
 966        end_packfile();
 967        start_packfile();
 968}
 969
 970static size_t encode_header(
 971        enum object_type type,
 972        size_t size,
 973        unsigned char *hdr)
 974{
 975        int n = 1;
 976        unsigned char c;
 977
 978        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 979                die("bad type %d", type);
 980
 981        c = (type << 4) | (size & 15);
 982        size >>= 4;
 983        while (size) {
 984                *hdr++ = c | 0x80;
 985                c = size & 0x7f;
 986                size >>= 7;
 987                n++;
 988        }
 989        *hdr = c;
 990        return n;
 991}
 992
 993static int store_object(
 994        enum object_type type,
 995        struct strbuf *dat,
 996        struct last_object *last,
 997        unsigned char *sha1out,
 998        uintmax_t mark)
 999{
1000        void *out, *delta;
1001        struct object_entry *e;
1002        unsigned char hdr[96];
1003        unsigned char sha1[20];
1004        unsigned long hdrlen, deltalen;
1005        SHA_CTX c;
1006        z_stream s;
1007
1008        hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1009                (unsigned long)dat->len) + 1;
1010        SHA1_Init(&c);
1011        SHA1_Update(&c, hdr, hdrlen);
1012        SHA1_Update(&c, dat->buf, dat->len);
1013        SHA1_Final(sha1, &c);
1014        if (sha1out)
1015                hashcpy(sha1out, sha1);
1016
1017        e = insert_object(sha1);
1018        if (mark)
1019                insert_mark(mark, e);
1020        if (e->offset) {
1021                duplicate_count_by_type[type]++;
1022                return 1;
1023        } else if (find_sha1_pack(sha1, packed_git)) {
1024                e->type = type;
1025                e->pack_id = MAX_PACK_ID;
1026                e->offset = 1; /* just not zero! */
1027                duplicate_count_by_type[type]++;
1028                return 1;
1029        }
1030
1031        if (last && last->data.buf && last->depth < max_depth) {
1032                delta = diff_delta(last->data.buf, last->data.len,
1033                        dat->buf, dat->len,
1034                        &deltalen, 0);
1035                if (delta && deltalen >= dat->len) {
1036                        free(delta);
1037                        delta = NULL;
1038                }
1039        } else
1040                delta = NULL;
1041
1042        memset(&s, 0, sizeof(s));
1043        deflateInit(&s, pack_compression_level);
1044        if (delta) {
1045                s.next_in = delta;
1046                s.avail_in = deltalen;
1047        } else {
1048                s.next_in = (void *)dat->buf;
1049                s.avail_in = dat->len;
1050        }
1051        s.avail_out = deflateBound(&s, s.avail_in);
1052        s.next_out = out = xmalloc(s.avail_out);
1053        while (deflate(&s, Z_FINISH) == Z_OK)
1054                /* nothing */;
1055        deflateEnd(&s);
1056
1057        /* Determine if we should auto-checkpoint. */
1058        if ((pack_size + 60 + s.total_out) > max_packsize
1059                || (pack_size + 60 + s.total_out) < pack_size) {
1060
1061                /* This new object needs to *not* have the current pack_id. */
1062                e->pack_id = pack_id + 1;
1063                cycle_packfile();
1064
1065                /* We cannot carry a delta into the new pack. */
1066                if (delta) {
1067                        free(delta);
1068                        delta = NULL;
1069
1070                        memset(&s, 0, sizeof(s));
1071                        deflateInit(&s, pack_compression_level);
1072                        s.next_in = (void *)dat->buf;
1073                        s.avail_in = dat->len;
1074                        s.avail_out = deflateBound(&s, s.avail_in);
1075                        s.next_out = out = xrealloc(out, s.avail_out);
1076                        while (deflate(&s, Z_FINISH) == Z_OK)
1077                                /* nothing */;
1078                        deflateEnd(&s);
1079                }
1080        }
1081
1082        e->type = type;
1083        e->pack_id = pack_id;
1084        e->offset = pack_size;
1085        object_count++;
1086        object_count_by_type[type]++;
1087
1088        if (delta) {
1089                unsigned long ofs = e->offset - last->offset;
1090                unsigned pos = sizeof(hdr) - 1;
1091
1092                delta_count_by_type[type]++;
1093                e->depth = last->depth + 1;
1094
1095                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1096                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1097                pack_size += hdrlen;
1098
1099                hdr[pos] = ofs & 127;
1100                while (ofs >>= 7)
1101                        hdr[--pos] = 128 | (--ofs & 127);
1102                write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1103                pack_size += sizeof(hdr) - pos;
1104        } else {
1105                e->depth = 0;
1106                hdrlen = encode_header(type, dat->len, hdr);
1107                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1108                pack_size += hdrlen;
1109        }
1110
1111        write_or_die(pack_data->pack_fd, out, s.total_out);
1112        pack_size += s.total_out;
1113
1114        free(out);
1115        free(delta);
1116        if (last) {
1117                if (last->no_swap) {
1118                        last->data = *dat;
1119                } else {
1120                        strbuf_swap(&last->data, dat);
1121                }
1122                last->offset = e->offset;
1123                last->depth = e->depth;
1124        }
1125        return 0;
1126}
1127
1128static void *gfi_unpack_entry(
1129        struct object_entry *oe,
1130        unsigned long *sizep)
1131{
1132        enum object_type type;
1133        struct packed_git *p = all_packs[oe->pack_id];
1134        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1135                close_pack_windows(p);
1136                p->pack_size = pack_size + 20;
1137        }
1138        return unpack_entry(p, oe->offset, &type, sizep);
1139}
1140
1141static const char *get_mode(const char *str, uint16_t *modep)
1142{
1143        unsigned char c;
1144        uint16_t mode = 0;
1145
1146        while ((c = *str++) != ' ') {
1147                if (c < '0' || c > '7')
1148                        return NULL;
1149                mode = (mode << 3) + (c - '0');
1150        }
1151        *modep = mode;
1152        return str;
1153}
1154
1155static void load_tree(struct tree_entry *root)
1156{
1157        unsigned char* sha1 = root->versions[1].sha1;
1158        struct object_entry *myoe;
1159        struct tree_content *t;
1160        unsigned long size;
1161        char *buf;
1162        const char *c;
1163
1164        root->tree = t = new_tree_content(8);
1165        if (is_null_sha1(sha1))
1166                return;
1167
1168        myoe = find_object(sha1);
1169        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1170                if (myoe->type != OBJ_TREE)
1171                        die("Not a tree: %s", sha1_to_hex(sha1));
1172                t->delta_depth = myoe->depth;
1173                buf = gfi_unpack_entry(myoe, &size);
1174        } else {
1175                enum object_type type;
1176                buf = read_sha1_file(sha1, &type, &size);
1177                if (!buf || type != OBJ_TREE)
1178                        die("Can't load tree %s", sha1_to_hex(sha1));
1179        }
1180
1181        c = buf;
1182        while (c != (buf + size)) {
1183                struct tree_entry *e = new_tree_entry();
1184
1185                if (t->entry_count == t->entry_capacity)
1186                        root->tree = t = grow_tree_content(t, t->entry_count);
1187                t->entries[t->entry_count++] = e;
1188
1189                e->tree = NULL;
1190                c = get_mode(c, &e->versions[1].mode);
1191                if (!c)
1192                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1193                e->versions[0].mode = e->versions[1].mode;
1194                e->name = to_atom(c, strlen(c));
1195                c += e->name->str_len + 1;
1196                hashcpy(e->versions[0].sha1, (unsigned char*)c);
1197                hashcpy(e->versions[1].sha1, (unsigned char*)c);
1198                c += 20;
1199        }
1200        free(buf);
1201}
1202
1203static int tecmp0 (const void *_a, const void *_b)
1204{
1205        struct tree_entry *a = *((struct tree_entry**)_a);
1206        struct tree_entry *b = *((struct tree_entry**)_b);
1207        return base_name_compare(
1208                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1209                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1210}
1211
1212static int tecmp1 (const void *_a, const void *_b)
1213{
1214        struct tree_entry *a = *((struct tree_entry**)_a);
1215        struct tree_entry *b = *((struct tree_entry**)_b);
1216        return base_name_compare(
1217                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1218                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1219}
1220
1221static void mktree(struct tree_content *t, int v, struct strbuf *b)
1222{
1223        size_t maxlen = 0;
1224        unsigned int i;
1225
1226        if (!v)
1227                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1228        else
1229                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1230
1231        for (i = 0; i < t->entry_count; i++) {
1232                if (t->entries[i]->versions[v].mode)
1233                        maxlen += t->entries[i]->name->str_len + 34;
1234        }
1235
1236        strbuf_reset(b);
1237        strbuf_grow(b, maxlen);
1238        for (i = 0; i < t->entry_count; i++) {
1239                struct tree_entry *e = t->entries[i];
1240                if (!e->versions[v].mode)
1241                        continue;
1242                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1243                                        e->name->str_dat, '\0');
1244                strbuf_add(b, e->versions[v].sha1, 20);
1245        }
1246}
1247
1248static void store_tree(struct tree_entry *root)
1249{
1250        struct tree_content *t = root->tree;
1251        unsigned int i, j, del;
1252        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1253        struct object_entry *le;
1254
1255        if (!is_null_sha1(root->versions[1].sha1))
1256                return;
1257
1258        for (i = 0; i < t->entry_count; i++) {
1259                if (t->entries[i]->tree)
1260                        store_tree(t->entries[i]);
1261        }
1262
1263        le = find_object(root->versions[0].sha1);
1264        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1265                mktree(t, 0, &old_tree);
1266                lo.data = old_tree;
1267                lo.offset = le->offset;
1268                lo.depth = t->delta_depth;
1269        }
1270
1271        mktree(t, 1, &new_tree);
1272        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1273
1274        t->delta_depth = lo.depth;
1275        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1276                struct tree_entry *e = t->entries[i];
1277                if (e->versions[1].mode) {
1278                        e->versions[0].mode = e->versions[1].mode;
1279                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1280                        t->entries[j++] = e;
1281                } else {
1282                        release_tree_entry(e);
1283                        del++;
1284                }
1285        }
1286        t->entry_count -= del;
1287}
1288
1289static int tree_content_set(
1290        struct tree_entry *root,
1291        const char *p,
1292        const unsigned char *sha1,
1293        const uint16_t mode,
1294        struct tree_content *subtree)
1295{
1296        struct tree_content *t = root->tree;
1297        const char *slash1;
1298        unsigned int i, n;
1299        struct tree_entry *e;
1300
1301        slash1 = strchr(p, '/');
1302        if (slash1)
1303                n = slash1 - p;
1304        else
1305                n = strlen(p);
1306        if (!n)
1307                die("Empty path component found in input");
1308        if (!slash1 && !S_ISDIR(mode) && subtree)
1309                die("Non-directories cannot have subtrees");
1310
1311        for (i = 0; i < t->entry_count; i++) {
1312                e = t->entries[i];
1313                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1314                        if (!slash1) {
1315                                if (!S_ISDIR(mode)
1316                                                && e->versions[1].mode == mode
1317                                                && !hashcmp(e->versions[1].sha1, sha1))
1318                                        return 0;
1319                                e->versions[1].mode = mode;
1320                                hashcpy(e->versions[1].sha1, sha1);
1321                                if (e->tree)
1322                                        release_tree_content_recursive(e->tree);
1323                                e->tree = subtree;
1324                                hashclr(root->versions[1].sha1);
1325                                return 1;
1326                        }
1327                        if (!S_ISDIR(e->versions[1].mode)) {
1328                                e->tree = new_tree_content(8);
1329                                e->versions[1].mode = S_IFDIR;
1330                        }
1331                        if (!e->tree)
1332                                load_tree(e);
1333                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1334                                hashclr(root->versions[1].sha1);
1335                                return 1;
1336                        }
1337                        return 0;
1338                }
1339        }
1340
1341        if (t->entry_count == t->entry_capacity)
1342                root->tree = t = grow_tree_content(t, t->entry_count);
1343        e = new_tree_entry();
1344        e->name = to_atom(p, n);
1345        e->versions[0].mode = 0;
1346        hashclr(e->versions[0].sha1);
1347        t->entries[t->entry_count++] = e;
1348        if (slash1) {
1349                e->tree = new_tree_content(8);
1350                e->versions[1].mode = S_IFDIR;
1351                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1352        } else {
1353                e->tree = subtree;
1354                e->versions[1].mode = mode;
1355                hashcpy(e->versions[1].sha1, sha1);
1356        }
1357        hashclr(root->versions[1].sha1);
1358        return 1;
1359}
1360
1361static int tree_content_remove(
1362        struct tree_entry *root,
1363        const char *p,
1364        struct tree_entry *backup_leaf)
1365{
1366        struct tree_content *t = root->tree;
1367        const char *slash1;
1368        unsigned int i, n;
1369        struct tree_entry *e;
1370
1371        slash1 = strchr(p, '/');
1372        if (slash1)
1373                n = slash1 - p;
1374        else
1375                n = strlen(p);
1376
1377        for (i = 0; i < t->entry_count; i++) {
1378                e = t->entries[i];
1379                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1380                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1381                                goto del_entry;
1382                        if (!e->tree)
1383                                load_tree(e);
1384                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1385                                for (n = 0; n < e->tree->entry_count; n++) {
1386                                        if (e->tree->entries[n]->versions[1].mode) {
1387                                                hashclr(root->versions[1].sha1);
1388                                                return 1;
1389                                        }
1390                                }
1391                                backup_leaf = NULL;
1392                                goto del_entry;
1393                        }
1394                        return 0;
1395                }
1396        }
1397        return 0;
1398
1399del_entry:
1400        if (backup_leaf)
1401                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1402        else if (e->tree)
1403                release_tree_content_recursive(e->tree);
1404        e->tree = NULL;
1405        e->versions[1].mode = 0;
1406        hashclr(e->versions[1].sha1);
1407        hashclr(root->versions[1].sha1);
1408        return 1;
1409}
1410
1411static int tree_content_get(
1412        struct tree_entry *root,
1413        const char *p,
1414        struct tree_entry *leaf)
1415{
1416        struct tree_content *t = root->tree;
1417        const char *slash1;
1418        unsigned int i, n;
1419        struct tree_entry *e;
1420
1421        slash1 = strchr(p, '/');
1422        if (slash1)
1423                n = slash1 - p;
1424        else
1425                n = strlen(p);
1426
1427        for (i = 0; i < t->entry_count; i++) {
1428                e = t->entries[i];
1429                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1430                        if (!slash1) {
1431                                memcpy(leaf, e, sizeof(*leaf));
1432                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1433                                        leaf->tree = dup_tree_content(e->tree);
1434                                else
1435                                        leaf->tree = NULL;
1436                                return 1;
1437                        }
1438                        if (!S_ISDIR(e->versions[1].mode))
1439                                return 0;
1440                        if (!e->tree)
1441                                load_tree(e);
1442                        return tree_content_get(e, slash1 + 1, leaf);
1443                }
1444        }
1445        return 0;
1446}
1447
1448static int update_branch(struct branch *b)
1449{
1450        static const char *msg = "fast-import";
1451        struct ref_lock *lock;
1452        unsigned char old_sha1[20];
1453
1454        if (read_ref(b->name, old_sha1))
1455                hashclr(old_sha1);
1456        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1457        if (!lock)
1458                return error("Unable to lock %s", b->name);
1459        if (!force_update && !is_null_sha1(old_sha1)) {
1460                struct commit *old_cmit, *new_cmit;
1461
1462                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1463                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1464                if (!old_cmit || !new_cmit) {
1465                        unlock_ref(lock);
1466                        return error("Branch %s is missing commits.", b->name);
1467                }
1468
1469                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1470                        unlock_ref(lock);
1471                        warning("Not updating %s"
1472                                " (new tip %s does not contain %s)",
1473                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1474                        return -1;
1475                }
1476        }
1477        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1478                return error("Unable to update %s", b->name);
1479        return 0;
1480}
1481
1482static void dump_branches(void)
1483{
1484        unsigned int i;
1485        struct branch *b;
1486
1487        for (i = 0; i < branch_table_sz; i++) {
1488                for (b = branch_table[i]; b; b = b->table_next_branch)
1489                        failure |= update_branch(b);
1490        }
1491}
1492
1493static void dump_tags(void)
1494{
1495        static const char *msg = "fast-import";
1496        struct tag *t;
1497        struct ref_lock *lock;
1498        char ref_name[PATH_MAX];
1499
1500        for (t = first_tag; t; t = t->next_tag) {
1501                sprintf(ref_name, "tags/%s", t->name);
1502                lock = lock_ref_sha1(ref_name, NULL);
1503                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1504                        failure |= error("Unable to update %s", ref_name);
1505        }
1506}
1507
1508static void dump_marks_helper(FILE *f,
1509        uintmax_t base,
1510        struct mark_set *m)
1511{
1512        uintmax_t k;
1513        if (m->shift) {
1514                for (k = 0; k < 1024; k++) {
1515                        if (m->data.sets[k])
1516                                dump_marks_helper(f, (base + k) << m->shift,
1517                                        m->data.sets[k]);
1518                }
1519        } else {
1520                for (k = 0; k < 1024; k++) {
1521                        if (m->data.marked[k])
1522                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1523                                        sha1_to_hex(m->data.marked[k]->sha1));
1524                }
1525        }
1526}
1527
1528static void dump_marks(void)
1529{
1530        static struct lock_file mark_lock;
1531        int mark_fd;
1532        FILE *f;
1533
1534        if (!mark_file)
1535                return;
1536
1537        mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1538        if (mark_fd < 0) {
1539                failure |= error("Unable to write marks file %s: %s",
1540                        mark_file, strerror(errno));
1541                return;
1542        }
1543
1544        f = fdopen(mark_fd, "w");
1545        if (!f) {
1546                int saved_errno = errno;
1547                rollback_lock_file(&mark_lock);
1548                failure |= error("Unable to write marks file %s: %s",
1549                        mark_file, strerror(saved_errno));
1550                return;
1551        }
1552
1553        /*
1554         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1555         * Assign -1 to the lock file descriptor so that commit_lock_file()
1556         * won't try to close() it.
1557         */
1558        mark_lock.fd = -1;
1559
1560        dump_marks_helper(f, 0, marks);
1561        if (ferror(f) || fclose(f)) {
1562                int saved_errno = errno;
1563                rollback_lock_file(&mark_lock);
1564                failure |= error("Unable to write marks file %s: %s",
1565                        mark_file, strerror(saved_errno));
1566                return;
1567        }
1568
1569        if (commit_lock_file(&mark_lock)) {
1570                int saved_errno = errno;
1571                rollback_lock_file(&mark_lock);
1572                failure |= error("Unable to commit marks file %s: %s",
1573                        mark_file, strerror(saved_errno));
1574                return;
1575        }
1576}
1577
1578static int read_next_command(void)
1579{
1580        static int stdin_eof = 0;
1581
1582        if (stdin_eof) {
1583                unread_command_buf = 0;
1584                return EOF;
1585        }
1586
1587        do {
1588                if (unread_command_buf) {
1589                        unread_command_buf = 0;
1590                } else {
1591                        struct recent_command *rc;
1592
1593                        strbuf_detach(&command_buf, NULL);
1594                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1595                        if (stdin_eof)
1596                                return EOF;
1597
1598                        rc = rc_free;
1599                        if (rc)
1600                                rc_free = rc->next;
1601                        else {
1602                                rc = cmd_hist.next;
1603                                cmd_hist.next = rc->next;
1604                                cmd_hist.next->prev = &cmd_hist;
1605                                free(rc->buf);
1606                        }
1607
1608                        rc->buf = command_buf.buf;
1609                        rc->prev = cmd_tail;
1610                        rc->next = cmd_hist.prev;
1611                        rc->prev->next = rc;
1612                        cmd_tail = rc;
1613                }
1614        } while (command_buf.buf[0] == '#');
1615
1616        return 0;
1617}
1618
1619static void skip_optional_lf(void)
1620{
1621        int term_char = fgetc(stdin);
1622        if (term_char != '\n' && term_char != EOF)
1623                ungetc(term_char, stdin);
1624}
1625
1626static void cmd_mark(void)
1627{
1628        if (!prefixcmp(command_buf.buf, "mark :")) {
1629                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1630                read_next_command();
1631        }
1632        else
1633                next_mark = 0;
1634}
1635
1636static void cmd_data(struct strbuf *sb)
1637{
1638        strbuf_reset(sb);
1639
1640        if (prefixcmp(command_buf.buf, "data "))
1641                die("Expected 'data n' command, found: %s", command_buf.buf);
1642
1643        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1644                char *term = xstrdup(command_buf.buf + 5 + 2);
1645                size_t term_len = command_buf.len - 5 - 2;
1646
1647                strbuf_detach(&command_buf, NULL);
1648                for (;;) {
1649                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1650                                die("EOF in data (terminator '%s' not found)", term);
1651                        if (term_len == command_buf.len
1652                                && !strcmp(term, command_buf.buf))
1653                                break;
1654                        strbuf_addbuf(sb, &command_buf);
1655                        strbuf_addch(sb, '\n');
1656                }
1657                free(term);
1658        }
1659        else {
1660                size_t n = 0, length;
1661
1662                length = strtoul(command_buf.buf + 5, NULL, 10);
1663
1664                while (n < length) {
1665                        size_t s = strbuf_fread(sb, length - n, stdin);
1666                        if (!s && feof(stdin))
1667                                die("EOF in data (%lu bytes remaining)",
1668                                        (unsigned long)(length - n));
1669                        n += s;
1670                }
1671        }
1672
1673        skip_optional_lf();
1674}
1675
1676static int validate_raw_date(const char *src, char *result, int maxlen)
1677{
1678        const char *orig_src = src;
1679        char *endp, sign;
1680
1681        strtoul(src, &endp, 10);
1682        if (endp == src || *endp != ' ')
1683                return -1;
1684
1685        src = endp + 1;
1686        if (*src != '-' && *src != '+')
1687                return -1;
1688        sign = *src;
1689
1690        strtoul(src + 1, &endp, 10);
1691        if (endp == src || *endp || (endp - orig_src) >= maxlen)
1692                return -1;
1693
1694        strcpy(result, orig_src);
1695        return 0;
1696}
1697
1698static char *parse_ident(const char *buf)
1699{
1700        const char *gt;
1701        size_t name_len;
1702        char *ident;
1703
1704        gt = strrchr(buf, '>');
1705        if (!gt)
1706                die("Missing > in ident string: %s", buf);
1707        gt++;
1708        if (*gt != ' ')
1709                die("Missing space after > in ident string: %s", buf);
1710        gt++;
1711        name_len = gt - buf;
1712        ident = xmalloc(name_len + 24);
1713        strncpy(ident, buf, name_len);
1714
1715        switch (whenspec) {
1716        case WHENSPEC_RAW:
1717                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1718                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1719                break;
1720        case WHENSPEC_RFC2822:
1721                if (parse_date(gt, ident + name_len, 24) < 0)
1722                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1723                break;
1724        case WHENSPEC_NOW:
1725                if (strcmp("now", gt))
1726                        die("Date in ident must be 'now': %s", buf);
1727                datestamp(ident + name_len, 24);
1728                break;
1729        }
1730
1731        return ident;
1732}
1733
1734static void cmd_new_blob(void)
1735{
1736        static struct strbuf buf = STRBUF_INIT;
1737
1738        read_next_command();
1739        cmd_mark();
1740        cmd_data(&buf);
1741        store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1742}
1743
1744static void unload_one_branch(void)
1745{
1746        while (cur_active_branches
1747                && cur_active_branches >= max_active_branches) {
1748                uintmax_t min_commit = ULONG_MAX;
1749                struct branch *e, *l = NULL, *p = NULL;
1750
1751                for (e = active_branches; e; e = e->active_next_branch) {
1752                        if (e->last_commit < min_commit) {
1753                                p = l;
1754                                min_commit = e->last_commit;
1755                        }
1756                        l = e;
1757                }
1758
1759                if (p) {
1760                        e = p->active_next_branch;
1761                        p->active_next_branch = e->active_next_branch;
1762                } else {
1763                        e = active_branches;
1764                        active_branches = e->active_next_branch;
1765                }
1766                e->active = 0;
1767                e->active_next_branch = NULL;
1768                if (e->branch_tree.tree) {
1769                        release_tree_content_recursive(e->branch_tree.tree);
1770                        e->branch_tree.tree = NULL;
1771                }
1772                cur_active_branches--;
1773        }
1774}
1775
1776static void load_branch(struct branch *b)
1777{
1778        load_tree(&b->branch_tree);
1779        if (!b->active) {
1780                b->active = 1;
1781                b->active_next_branch = active_branches;
1782                active_branches = b;
1783                cur_active_branches++;
1784                branch_load_count++;
1785        }
1786}
1787
1788static void file_change_m(struct branch *b)
1789{
1790        const char *p = command_buf.buf + 2;
1791        static struct strbuf uq = STRBUF_INIT;
1792        const char *endp;
1793        struct object_entry *oe = oe;
1794        unsigned char sha1[20];
1795        uint16_t mode, inline_data = 0;
1796
1797        p = get_mode(p, &mode);
1798        if (!p)
1799                die("Corrupt mode: %s", command_buf.buf);
1800        switch (mode) {
1801        case S_IFREG | 0644:
1802        case S_IFREG | 0755:
1803        case S_IFLNK:
1804        case 0644:
1805        case 0755:
1806                /* ok */
1807                break;
1808        default:
1809                die("Corrupt mode: %s", command_buf.buf);
1810        }
1811
1812        if (*p == ':') {
1813                char *x;
1814                oe = find_mark(strtoumax(p + 1, &x, 10));
1815                hashcpy(sha1, oe->sha1);
1816                p = x;
1817        } else if (!prefixcmp(p, "inline")) {
1818                inline_data = 1;
1819                p += 6;
1820        } else {
1821                if (get_sha1_hex(p, sha1))
1822                        die("Invalid SHA1: %s", command_buf.buf);
1823                oe = find_object(sha1);
1824                p += 40;
1825        }
1826        if (*p++ != ' ')
1827                die("Missing space after SHA1: %s", command_buf.buf);
1828
1829        strbuf_reset(&uq);
1830        if (!unquote_c_style(&uq, p, &endp)) {
1831                if (*endp)
1832                        die("Garbage after path in: %s", command_buf.buf);
1833                p = uq.buf;
1834        }
1835
1836        if (inline_data) {
1837                static struct strbuf buf = STRBUF_INIT;
1838
1839                if (p != uq.buf) {
1840                        strbuf_addstr(&uq, p);
1841                        p = uq.buf;
1842                }
1843                read_next_command();
1844                cmd_data(&buf);
1845                store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1846        } else if (oe) {
1847                if (oe->type != OBJ_BLOB)
1848                        die("Not a blob (actually a %s): %s",
1849                                typename(oe->type), command_buf.buf);
1850        } else {
1851                enum object_type type = sha1_object_info(sha1, NULL);
1852                if (type < 0)
1853                        die("Blob not found: %s", command_buf.buf);
1854                if (type != OBJ_BLOB)
1855                        die("Not a blob (actually a %s): %s",
1856                            typename(type), command_buf.buf);
1857        }
1858
1859        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1860}
1861
1862static void file_change_d(struct branch *b)
1863{
1864        const char *p = command_buf.buf + 2;
1865        static struct strbuf uq = STRBUF_INIT;
1866        const char *endp;
1867
1868        strbuf_reset(&uq);
1869        if (!unquote_c_style(&uq, p, &endp)) {
1870                if (*endp)
1871                        die("Garbage after path in: %s", command_buf.buf);
1872                p = uq.buf;
1873        }
1874        tree_content_remove(&b->branch_tree, p, NULL);
1875}
1876
1877static void file_change_cr(struct branch *b, int rename)
1878{
1879        const char *s, *d;
1880        static struct strbuf s_uq = STRBUF_INIT;
1881        static struct strbuf d_uq = STRBUF_INIT;
1882        const char *endp;
1883        struct tree_entry leaf;
1884
1885        s = command_buf.buf + 2;
1886        strbuf_reset(&s_uq);
1887        if (!unquote_c_style(&s_uq, s, &endp)) {
1888                if (*endp != ' ')
1889                        die("Missing space after source: %s", command_buf.buf);
1890        } else {
1891                endp = strchr(s, ' ');
1892                if (!endp)
1893                        die("Missing space after source: %s", command_buf.buf);
1894                strbuf_add(&s_uq, s, endp - s);
1895        }
1896        s = s_uq.buf;
1897
1898        endp++;
1899        if (!*endp)
1900                die("Missing dest: %s", command_buf.buf);
1901
1902        d = endp;
1903        strbuf_reset(&d_uq);
1904        if (!unquote_c_style(&d_uq, d, &endp)) {
1905                if (*endp)
1906                        die("Garbage after dest in: %s", command_buf.buf);
1907                d = d_uq.buf;
1908        }
1909
1910        memset(&leaf, 0, sizeof(leaf));
1911        if (rename)
1912                tree_content_remove(&b->branch_tree, s, &leaf);
1913        else
1914                tree_content_get(&b->branch_tree, s, &leaf);
1915        if (!leaf.versions[1].mode)
1916                die("Path %s not in branch", s);
1917        tree_content_set(&b->branch_tree, d,
1918                leaf.versions[1].sha1,
1919                leaf.versions[1].mode,
1920                leaf.tree);
1921}
1922
1923static void file_change_deleteall(struct branch *b)
1924{
1925        release_tree_content_recursive(b->branch_tree.tree);
1926        hashclr(b->branch_tree.versions[0].sha1);
1927        hashclr(b->branch_tree.versions[1].sha1);
1928        load_tree(&b->branch_tree);
1929}
1930
1931static void cmd_from_commit(struct branch *b, char *buf, unsigned long size)
1932{
1933        if (!buf || size < 46)
1934                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
1935        if (memcmp("tree ", buf, 5)
1936                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1937                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1938        hashcpy(b->branch_tree.versions[0].sha1,
1939                b->branch_tree.versions[1].sha1);
1940}
1941
1942static void cmd_from_existing(struct branch *b)
1943{
1944        if (is_null_sha1(b->sha1)) {
1945                hashclr(b->branch_tree.versions[0].sha1);
1946                hashclr(b->branch_tree.versions[1].sha1);
1947        } else {
1948                unsigned long size;
1949                char *buf;
1950
1951                buf = read_object_with_reference(b->sha1,
1952                        commit_type, &size, b->sha1);
1953                cmd_from_commit(b, buf, size);
1954                free(buf);
1955        }
1956}
1957
1958static int cmd_from(struct branch *b)
1959{
1960        const char *from;
1961        struct branch *s;
1962
1963        if (prefixcmp(command_buf.buf, "from "))
1964                return 0;
1965
1966        if (b->branch_tree.tree) {
1967                release_tree_content_recursive(b->branch_tree.tree);
1968                b->branch_tree.tree = NULL;
1969        }
1970
1971        from = strchr(command_buf.buf, ' ') + 1;
1972        s = lookup_branch(from);
1973        if (b == s)
1974                die("Can't create a branch from itself: %s", b->name);
1975        else if (s) {
1976                unsigned char *t = s->branch_tree.versions[1].sha1;
1977                hashcpy(b->sha1, s->sha1);
1978                hashcpy(b->branch_tree.versions[0].sha1, t);
1979                hashcpy(b->branch_tree.versions[1].sha1, t);
1980        } else if (*from == ':') {
1981                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1982                struct object_entry *oe = find_mark(idnum);
1983                if (oe->type != OBJ_COMMIT)
1984                        die("Mark :%" PRIuMAX " not a commit", idnum);
1985                hashcpy(b->sha1, oe->sha1);
1986                if (oe->pack_id != MAX_PACK_ID) {
1987                        unsigned long size;
1988                        char *buf = gfi_unpack_entry(oe, &size);
1989                        cmd_from_commit(b, buf, size);
1990                        free(buf);
1991                } else
1992                        cmd_from_existing(b);
1993        } else if (!get_sha1(from, b->sha1))
1994                cmd_from_existing(b);
1995        else
1996                die("Invalid ref name or SHA1 expression: %s", from);
1997
1998        read_next_command();
1999        return 1;
2000}
2001
2002static struct hash_list *cmd_merge(unsigned int *count)
2003{
2004        struct hash_list *list = NULL, *n, *e = e;
2005        const char *from;
2006        struct branch *s;
2007
2008        *count = 0;
2009        while (!prefixcmp(command_buf.buf, "merge ")) {
2010                from = strchr(command_buf.buf, ' ') + 1;
2011                n = xmalloc(sizeof(*n));
2012                s = lookup_branch(from);
2013                if (s)
2014                        hashcpy(n->sha1, s->sha1);
2015                else if (*from == ':') {
2016                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2017                        struct object_entry *oe = find_mark(idnum);
2018                        if (oe->type != OBJ_COMMIT)
2019                                die("Mark :%" PRIuMAX " not a commit", idnum);
2020                        hashcpy(n->sha1, oe->sha1);
2021                } else if (!get_sha1(from, n->sha1)) {
2022                        unsigned long size;
2023                        char *buf = read_object_with_reference(n->sha1,
2024                                commit_type, &size, n->sha1);
2025                        if (!buf || size < 46)
2026                                die("Not a valid commit: %s", from);
2027                        free(buf);
2028                } else
2029                        die("Invalid ref name or SHA1 expression: %s", from);
2030
2031                n->next = NULL;
2032                if (list)
2033                        e->next = n;
2034                else
2035                        list = n;
2036                e = n;
2037                (*count)++;
2038                read_next_command();
2039        }
2040        return list;
2041}
2042
2043static void cmd_new_commit(void)
2044{
2045        static struct strbuf msg = STRBUF_INIT;
2046        struct branch *b;
2047        char *sp;
2048        char *author = NULL;
2049        char *committer = NULL;
2050        struct hash_list *merge_list = NULL;
2051        unsigned int merge_count;
2052
2053        /* Obtain the branch name from the rest of our command */
2054        sp = strchr(command_buf.buf, ' ') + 1;
2055        b = lookup_branch(sp);
2056        if (!b)
2057                b = new_branch(sp);
2058
2059        read_next_command();
2060        cmd_mark();
2061        if (!prefixcmp(command_buf.buf, "author ")) {
2062                author = parse_ident(command_buf.buf + 7);
2063                read_next_command();
2064        }
2065        if (!prefixcmp(command_buf.buf, "committer ")) {
2066                committer = parse_ident(command_buf.buf + 10);
2067                read_next_command();
2068        }
2069        if (!committer)
2070                die("Expected committer but didn't get one");
2071        cmd_data(&msg);
2072        read_next_command();
2073        cmd_from(b);
2074        merge_list = cmd_merge(&merge_count);
2075
2076        /* ensure the branch is active/loaded */
2077        if (!b->branch_tree.tree || !max_active_branches) {
2078                unload_one_branch();
2079                load_branch(b);
2080        }
2081
2082        /* file_change* */
2083        while (command_buf.len > 0) {
2084                if (!prefixcmp(command_buf.buf, "M "))
2085                        file_change_m(b);
2086                else if (!prefixcmp(command_buf.buf, "D "))
2087                        file_change_d(b);
2088                else if (!prefixcmp(command_buf.buf, "R "))
2089                        file_change_cr(b, 1);
2090                else if (!prefixcmp(command_buf.buf, "C "))
2091                        file_change_cr(b, 0);
2092                else if (!strcmp("deleteall", command_buf.buf))
2093                        file_change_deleteall(b);
2094                else {
2095                        unread_command_buf = 1;
2096                        break;
2097                }
2098                if (read_next_command() == EOF)
2099                        break;
2100        }
2101
2102        /* build the tree and the commit */
2103        store_tree(&b->branch_tree);
2104        hashcpy(b->branch_tree.versions[0].sha1,
2105                b->branch_tree.versions[1].sha1);
2106
2107        strbuf_reset(&new_data);
2108        strbuf_addf(&new_data, "tree %s\n",
2109                sha1_to_hex(b->branch_tree.versions[1].sha1));
2110        if (!is_null_sha1(b->sha1))
2111                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2112        while (merge_list) {
2113                struct hash_list *next = merge_list->next;
2114                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2115                free(merge_list);
2116                merge_list = next;
2117        }
2118        strbuf_addf(&new_data,
2119                "author %s\n"
2120                "committer %s\n"
2121                "\n",
2122                author ? author : committer, committer);
2123        strbuf_addbuf(&new_data, &msg);
2124        free(author);
2125        free(committer);
2126
2127        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2128                b->pack_id = pack_id;
2129        b->last_commit = object_count_by_type[OBJ_COMMIT];
2130}
2131
2132static void cmd_new_tag(void)
2133{
2134        static struct strbuf msg = STRBUF_INIT;
2135        char *sp;
2136        const char *from;
2137        char *tagger;
2138        struct branch *s;
2139        struct tag *t;
2140        uintmax_t from_mark = 0;
2141        unsigned char sha1[20];
2142
2143        /* Obtain the new tag name from the rest of our command */
2144        sp = strchr(command_buf.buf, ' ') + 1;
2145        t = pool_alloc(sizeof(struct tag));
2146        t->next_tag = NULL;
2147        t->name = pool_strdup(sp);
2148        if (last_tag)
2149                last_tag->next_tag = t;
2150        else
2151                first_tag = t;
2152        last_tag = t;
2153        read_next_command();
2154
2155        /* from ... */
2156        if (prefixcmp(command_buf.buf, "from "))
2157                die("Expected from command, got %s", command_buf.buf);
2158        from = strchr(command_buf.buf, ' ') + 1;
2159        s = lookup_branch(from);
2160        if (s) {
2161                hashcpy(sha1, s->sha1);
2162        } else if (*from == ':') {
2163                struct object_entry *oe;
2164                from_mark = strtoumax(from + 1, NULL, 10);
2165                oe = find_mark(from_mark);
2166                if (oe->type != OBJ_COMMIT)
2167                        die("Mark :%" PRIuMAX " not a commit", from_mark);
2168                hashcpy(sha1, oe->sha1);
2169        } else if (!get_sha1(from, sha1)) {
2170                unsigned long size;
2171                char *buf;
2172
2173                buf = read_object_with_reference(sha1,
2174                        commit_type, &size, sha1);
2175                if (!buf || size < 46)
2176                        die("Not a valid commit: %s", from);
2177                free(buf);
2178        } else
2179                die("Invalid ref name or SHA1 expression: %s", from);
2180        read_next_command();
2181
2182        /* tagger ... */
2183        if (prefixcmp(command_buf.buf, "tagger "))
2184                die("Expected tagger command, got %s", command_buf.buf);
2185        tagger = parse_ident(command_buf.buf + 7);
2186
2187        /* tag payload/message */
2188        read_next_command();
2189        cmd_data(&msg);
2190
2191        /* build the tag object */
2192        strbuf_reset(&new_data);
2193        strbuf_addf(&new_data,
2194                "object %s\n"
2195                "type %s\n"
2196                "tag %s\n"
2197                "tagger %s\n"
2198                "\n",
2199                sha1_to_hex(sha1), commit_type, t->name, tagger);
2200        strbuf_addbuf(&new_data, &msg);
2201        free(tagger);
2202
2203        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2204                t->pack_id = MAX_PACK_ID;
2205        else
2206                t->pack_id = pack_id;
2207}
2208
2209static void cmd_reset_branch(void)
2210{
2211        struct branch *b;
2212        char *sp;
2213
2214        /* Obtain the branch name from the rest of our command */
2215        sp = strchr(command_buf.buf, ' ') + 1;
2216        b = lookup_branch(sp);
2217        if (b) {
2218                hashclr(b->sha1);
2219                hashclr(b->branch_tree.versions[0].sha1);
2220                hashclr(b->branch_tree.versions[1].sha1);
2221                if (b->branch_tree.tree) {
2222                        release_tree_content_recursive(b->branch_tree.tree);
2223                        b->branch_tree.tree = NULL;
2224                }
2225        }
2226        else
2227                b = new_branch(sp);
2228        read_next_command();
2229        if (!cmd_from(b) && command_buf.len > 0)
2230                unread_command_buf = 1;
2231}
2232
2233static void cmd_checkpoint(void)
2234{
2235        if (object_count) {
2236                cycle_packfile();
2237                dump_branches();
2238                dump_tags();
2239                dump_marks();
2240        }
2241        skip_optional_lf();
2242}
2243
2244static void cmd_progress(void)
2245{
2246        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2247        fputc('\n', stdout);
2248        fflush(stdout);
2249        skip_optional_lf();
2250}
2251
2252static void import_marks(const char *input_file)
2253{
2254        char line[512];
2255        FILE *f = fopen(input_file, "r");
2256        if (!f)
2257                die("cannot read %s: %s", input_file, strerror(errno));
2258        while (fgets(line, sizeof(line), f)) {
2259                uintmax_t mark;
2260                char *end;
2261                unsigned char sha1[20];
2262                struct object_entry *e;
2263
2264                end = strchr(line, '\n');
2265                if (line[0] != ':' || !end)
2266                        die("corrupt mark line: %s", line);
2267                *end = 0;
2268                mark = strtoumax(line + 1, &end, 10);
2269                if (!mark || end == line + 1
2270                        || *end != ' ' || get_sha1(end + 1, sha1))
2271                        die("corrupt mark line: %s", line);
2272                e = find_object(sha1);
2273                if (!e) {
2274                        enum object_type type = sha1_object_info(sha1, NULL);
2275                        if (type < 0)
2276                                die("object not found: %s", sha1_to_hex(sha1));
2277                        e = insert_object(sha1);
2278                        e->type = type;
2279                        e->pack_id = MAX_PACK_ID;
2280                        e->offset = 1; /* just not zero! */
2281                }
2282                insert_mark(mark, e);
2283        }
2284        fclose(f);
2285}
2286
2287static int git_pack_config(const char *k, const char *v)
2288{
2289        if (!strcmp(k, "pack.depth")) {
2290                max_depth = git_config_int(k, v);
2291                if (max_depth > MAX_DEPTH)
2292                        max_depth = MAX_DEPTH;
2293                return 0;
2294        }
2295        if (!strcmp(k, "pack.compression")) {
2296                int level = git_config_int(k, v);
2297                if (level == -1)
2298                        level = Z_DEFAULT_COMPRESSION;
2299                else if (level < 0 || level > Z_BEST_COMPRESSION)
2300                        die("bad pack compression level %d", level);
2301                pack_compression_level = level;
2302                pack_compression_seen = 1;
2303                return 0;
2304        }
2305        return git_default_config(k, v);
2306}
2307
2308static const char fast_import_usage[] =
2309"git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2310
2311int main(int argc, const char **argv)
2312{
2313        unsigned int i, show_stats = 1;
2314
2315        git_config(git_pack_config);
2316        if (!pack_compression_seen && core_compression_seen)
2317                pack_compression_level = core_compression_level;
2318
2319        alloc_objects(object_entry_alloc);
2320        strbuf_init(&command_buf, 0);
2321        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2322        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2323        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2324        marks = pool_calloc(1, sizeof(struct mark_set));
2325
2326        for (i = 1; i < argc; i++) {
2327                const char *a = argv[i];
2328
2329                if (*a != '-' || !strcmp(a, "--"))
2330                        break;
2331                else if (!prefixcmp(a, "--date-format=")) {
2332                        const char *fmt = a + 14;
2333                        if (!strcmp(fmt, "raw"))
2334                                whenspec = WHENSPEC_RAW;
2335                        else if (!strcmp(fmt, "rfc2822"))
2336                                whenspec = WHENSPEC_RFC2822;
2337                        else if (!strcmp(fmt, "now"))
2338                                whenspec = WHENSPEC_NOW;
2339                        else
2340                                die("unknown --date-format argument %s", fmt);
2341                }
2342                else if (!prefixcmp(a, "--max-pack-size="))
2343                        max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2344                else if (!prefixcmp(a, "--depth=")) {
2345                        max_depth = strtoul(a + 8, NULL, 0);
2346                        if (max_depth > MAX_DEPTH)
2347                                die("--depth cannot exceed %u", MAX_DEPTH);
2348                }
2349                else if (!prefixcmp(a, "--active-branches="))
2350                        max_active_branches = strtoul(a + 18, NULL, 0);
2351                else if (!prefixcmp(a, "--import-marks="))
2352                        import_marks(a + 15);
2353                else if (!prefixcmp(a, "--export-marks="))
2354                        mark_file = a + 15;
2355                else if (!prefixcmp(a, "--export-pack-edges=")) {
2356                        if (pack_edges)
2357                                fclose(pack_edges);
2358                        pack_edges = fopen(a + 20, "a");
2359                        if (!pack_edges)
2360                                die("Cannot open %s: %s", a + 20, strerror(errno));
2361                } else if (!strcmp(a, "--force"))
2362                        force_update = 1;
2363                else if (!strcmp(a, "--quiet"))
2364                        show_stats = 0;
2365                else if (!strcmp(a, "--stats"))
2366                        show_stats = 1;
2367                else
2368                        die("unknown option %s", a);
2369        }
2370        if (i != argc)
2371                usage(fast_import_usage);
2372
2373        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2374        for (i = 0; i < (cmd_save - 1); i++)
2375                rc_free[i].next = &rc_free[i + 1];
2376        rc_free[cmd_save - 1].next = NULL;
2377
2378        prepare_packed_git();
2379        start_packfile();
2380        set_die_routine(die_nicely);
2381        while (read_next_command() != EOF) {
2382                if (!strcmp("blob", command_buf.buf))
2383                        cmd_new_blob();
2384                else if (!prefixcmp(command_buf.buf, "commit "))
2385                        cmd_new_commit();
2386                else if (!prefixcmp(command_buf.buf, "tag "))
2387                        cmd_new_tag();
2388                else if (!prefixcmp(command_buf.buf, "reset "))
2389                        cmd_reset_branch();
2390                else if (!strcmp("checkpoint", command_buf.buf))
2391                        cmd_checkpoint();
2392                else if (!prefixcmp(command_buf.buf, "progress "))
2393                        cmd_progress();
2394                else
2395                        die("Unsupported command: %s", command_buf.buf);
2396        }
2397        end_packfile();
2398
2399        dump_branches();
2400        dump_tags();
2401        unkeep_all_packs();
2402        dump_marks();
2403
2404        if (pack_edges)
2405                fclose(pack_edges);
2406
2407        if (show_stats) {
2408                uintmax_t total_count = 0, duplicate_count = 0;
2409                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2410                        total_count += object_count_by_type[i];
2411                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2412                        duplicate_count += duplicate_count_by_type[i];
2413
2414                fprintf(stderr, "%s statistics:\n", argv[0]);
2415                fprintf(stderr, "---------------------------------------------------------------------\n");
2416                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2417                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2418                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]);
2419                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]);
2420                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]);
2421                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]);
2422                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2423                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2424                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2425                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2426                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2427                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2428                fprintf(stderr, "---------------------------------------------------------------------\n");
2429                pack_report();
2430                fprintf(stderr, "---------------------------------------------------------------------\n");
2431                fprintf(stderr, "\n");
2432        }
2433
2434        return failure ? 1 : 0;
2435}