fast-import.con commit Map only part of the generated pack file at any point in time. (53dbce7)
   1/*
   2Format of STDIN stream:
   3
   4  stream ::= cmd*;
   5
   6  cmd ::= new_blob
   7        | new_commit
   8        | new_tag
   9        ;
  10
  11  new_blob ::= 'blob' lf
  12        mark?
  13    file_content;
  14  file_content ::= data;
  15
  16  new_commit ::= 'commit' sp ref_str lf
  17    mark?
  18    ('author' sp name '<' email '>' ts tz lf)?
  19    'committer' sp name '<' email '>' ts tz lf
  20    commit_msg
  21    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  22    file_change*
  23    lf;
  24  commit_msg ::= data;
  25
  26  file_change ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf
  27                | 'D' sp path_str lf
  28                ;
  29  mode ::= '644' | '755';
  30
  31  new_tag ::= 'tag' sp tag_str lf
  32    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
  33        'tagger' sp name '<' email '>' ts tz lf
  34    tag_msg;
  35  tag_msg ::= data;
  36
  37     # note: the first idnum in a stream should be 1 and subsequent
  38     # idnums should not have gaps between values as this will cause
  39     # the stream parser to reserve space for the gapped values.  An
  40         # idnum can be updated in the future to a new object by issuing
  41     # a new mark directive with the old idnum.
  42         #
  43  mark ::= 'mark' sp idnum lf;
  44
  45     # note: declen indicates the length of binary_data in bytes.
  46     # declen does not include the lf preceeding or trailing the
  47     # binary data.
  48     #
  49  data ::= 'data' sp declen lf
  50    binary_data
  51        lf;
  52
  53     # note: quoted strings are C-style quoting supporting \c for
  54     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  55         # is the signed byte value in octal.  Note that the only
  56     # characters which must actually be escaped to protect the
  57     # stream formatting is: \, " and LF.  Otherwise these values
  58         # are UTF8.
  59     #
  60  ref_str     ::= ref     | '"' quoted(ref)     '"' ;
  61  sha1exp_str ::= sha1exp | '"' quoted(sha1exp) '"' ;
  62  tag_str     ::= tag     | '"' quoted(tag)     '"' ;
  63  path_str    ::= path    | '"' quoted(path)    '"' ;
  64
  65  declen ::= # unsigned 32 bit value, ascii base10 notation;
  66  binary_data ::= # file content, not interpreted;
  67
  68  sp ::= # ASCII space character;
  69  lf ::= # ASCII newline (LF) character;
  70
  71     # note: a colon (':') must precede the numerical value assigned to
  72         # an idnum.  This is to distinguish it from a ref or tag name as
  73     # GIT does not permit ':' in ref or tag strings.
  74         #
  75  idnum   ::= ':' declen;
  76  path    ::= # GIT style file path, e.g. "a/b/c";
  77  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
  78  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
  79  sha1exp ::= # Any valid GIT SHA1 expression;
  80  hexsha1 ::= # SHA1 in hexadecimal format;
  81
  82     # note: name and email are UTF8 strings, however name must not
  83         # contain '<' or lf and email must not contain any of the
  84     # following: '<', '>', lf.
  85         #
  86  name  ::= # valid GIT author/committer name;
  87  email ::= # valid GIT author/committer email;
  88  ts    ::= # time since the epoch in seconds, ascii base10 notation;
  89  tz    ::= # GIT style timezone;
  90*/
  91
  92#include "builtin.h"
  93#include "cache.h"
  94#include "object.h"
  95#include "blob.h"
  96#include "tree.h"
  97#include "delta.h"
  98#include "pack.h"
  99#include "refs.h"
 100#include "csum-file.h"
 101#include "strbuf.h"
 102#include "quote.h"
 103
 104struct object_entry
 105{
 106        struct object_entry *next;
 107        enum object_type type;
 108        unsigned long offset;
 109        unsigned char sha1[20];
 110};
 111
 112struct object_entry_pool
 113{
 114        struct object_entry_pool *next_pool;
 115        struct object_entry *next_free;
 116        struct object_entry *end;
 117        struct object_entry entries[FLEX_ARRAY]; /* more */
 118};
 119
 120struct mark_set
 121{
 122        int shift;
 123        union {
 124                struct object_entry *marked[1024];
 125                struct mark_set *sets[1024];
 126        } data;
 127};
 128
 129struct last_object
 130{
 131        void *data;
 132        unsigned int len;
 133        unsigned int depth;
 134        unsigned char sha1[20];
 135};
 136
 137struct mem_pool
 138{
 139        struct mem_pool *next_pool;
 140        char *next_free;
 141        char *end;
 142        char space[FLEX_ARRAY]; /* more */
 143};
 144
 145struct atom_str
 146{
 147        struct atom_str *next_atom;
 148        int str_len;
 149        char str_dat[FLEX_ARRAY]; /* more */
 150};
 151
 152struct tree_content;
 153struct tree_entry
 154{
 155        struct tree_content *tree;
 156        struct atom_str* name;
 157        unsigned int mode;
 158        unsigned char sha1[20];
 159};
 160
 161struct tree_content
 162{
 163        unsigned int entry_capacity; /* must match avail_tree_content */
 164        unsigned int entry_count;
 165        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 166};
 167
 168struct avail_tree_content
 169{
 170        unsigned int entry_capacity; /* must match tree_content */
 171        struct avail_tree_content *next_avail;
 172};
 173
 174struct branch
 175{
 176        struct branch *table_next_branch;
 177        struct branch *active_next_branch;
 178        const char *name;
 179        unsigned long last_commit;
 180        struct tree_entry branch_tree;
 181        unsigned char sha1[20];
 182};
 183
 184struct tag
 185{
 186        struct tag *next_tag;
 187        const char *name;
 188        unsigned char sha1[20];
 189};
 190
 191
 192/* Stats and misc. counters */
 193static unsigned long max_depth = 10;
 194static unsigned long alloc_count;
 195static unsigned long branch_count;
 196static unsigned long branch_load_count;
 197static unsigned long remap_count;
 198static unsigned long object_count;
 199static unsigned long duplicate_count;
 200static unsigned long marks_set_count;
 201static unsigned long object_count_by_type[9];
 202static unsigned long duplicate_count_by_type[9];
 203
 204/* Memory pools */
 205static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 206static size_t total_allocd;
 207static struct mem_pool *mem_pool;
 208
 209/* Atom management */
 210static unsigned int atom_table_sz = 4451;
 211static unsigned int atom_cnt;
 212static struct atom_str **atom_table;
 213
 214/* The .pack file being generated */
 215static int pack_fd;
 216static unsigned long pack_size;
 217static unsigned char pack_sha1[20];
 218static unsigned char* pack_base;
 219static unsigned long pack_moff;
 220static unsigned long pack_mlen = 128*1024*1024;
 221static unsigned long page_size;
 222
 223/* Table of objects we've written. */
 224static unsigned int object_entry_alloc = 1000;
 225static struct object_entry_pool *blocks;
 226static struct object_entry *object_table[1 << 16];
 227static struct mark_set *marks;
 228static const char* mark_file;
 229
 230/* Our last blob */
 231static struct last_object last_blob;
 232
 233/* Tree management */
 234static unsigned int tree_entry_alloc = 1000;
 235static void *avail_tree_entry;
 236static unsigned int avail_tree_table_sz = 100;
 237static struct avail_tree_content **avail_tree_table;
 238
 239/* Branch data */
 240static unsigned long max_active_branches = 5;
 241static unsigned long cur_active_branches;
 242static unsigned long branch_table_sz = 1039;
 243static struct branch **branch_table;
 244static struct branch *active_branches;
 245
 246/* Tag data */
 247static struct tag *first_tag;
 248static struct tag *last_tag;
 249
 250/* Input stream parsing */
 251static struct strbuf command_buf;
 252static unsigned long next_mark;
 253static FILE* branch_log;
 254
 255
 256static void alloc_objects(int cnt)
 257{
 258        struct object_entry_pool *b;
 259
 260        b = xmalloc(sizeof(struct object_entry_pool)
 261                + cnt * sizeof(struct object_entry));
 262        b->next_pool = blocks;
 263        b->next_free = b->entries;
 264        b->end = b->entries + cnt;
 265        blocks = b;
 266        alloc_count += cnt;
 267}
 268
 269static struct object_entry* new_object(unsigned char *sha1)
 270{
 271        struct object_entry *e;
 272
 273        if (blocks->next_free == blocks->end)
 274                alloc_objects(object_entry_alloc);
 275
 276        e = blocks->next_free++;
 277        memcpy(e->sha1, sha1, sizeof(e->sha1));
 278        return e;
 279}
 280
 281static struct object_entry* find_object(unsigned char *sha1)
 282{
 283        unsigned int h = sha1[0] << 8 | sha1[1];
 284        struct object_entry *e;
 285        for (e = object_table[h]; e; e = e->next)
 286                if (!memcmp(sha1, e->sha1, sizeof(e->sha1)))
 287                        return e;
 288        return NULL;
 289}
 290
 291static struct object_entry* insert_object(unsigned char *sha1)
 292{
 293        unsigned int h = sha1[0] << 8 | sha1[1];
 294        struct object_entry *e = object_table[h];
 295        struct object_entry *p = NULL;
 296
 297        while (e) {
 298                if (!memcmp(sha1, e->sha1, sizeof(e->sha1)))
 299                        return e;
 300                p = e;
 301                e = e->next;
 302        }
 303
 304        e = new_object(sha1);
 305        e->next = NULL;
 306        e->offset = 0;
 307        if (p)
 308                p->next = e;
 309        else
 310                object_table[h] = e;
 311        return e;
 312}
 313
 314static unsigned int hc_str(const char *s, size_t len)
 315{
 316        unsigned int r = 0;
 317        while (len-- > 0)
 318                r = r * 31 + *s++;
 319        return r;
 320}
 321
 322static void* pool_alloc(size_t len)
 323{
 324        struct mem_pool *p;
 325        void *r;
 326
 327        for (p = mem_pool; p; p = p->next_pool)
 328                if ((p->end - p->next_free >= len))
 329                        break;
 330
 331        if (!p) {
 332                if (len >= (mem_pool_alloc/2)) {
 333                        total_allocd += len;
 334                        return xmalloc(len);
 335                }
 336                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 337                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 338                p->next_pool = mem_pool;
 339                p->next_free = p->space;
 340                p->end = p->next_free + mem_pool_alloc;
 341                mem_pool = p;
 342        }
 343
 344        r = p->next_free;
 345        /* round out to a pointer alignment */
 346        if (len & (sizeof(void*) - 1))
 347                len += sizeof(void*) - (len & (sizeof(void*) - 1));
 348        p->next_free += len;
 349        return r;
 350}
 351
 352static void* pool_calloc(size_t count, size_t size)
 353{
 354        size_t len = count * size;
 355        void *r = pool_alloc(len);
 356        memset(r, 0, len);
 357        return r;
 358}
 359
 360static char* pool_strdup(const char *s)
 361{
 362        char *r = pool_alloc(strlen(s) + 1);
 363        strcpy(r, s);
 364        return r;
 365}
 366
 367static void insert_mark(unsigned long idnum, struct object_entry *oe)
 368{
 369        struct mark_set *s = marks;
 370        while ((idnum >> s->shift) >= 1024) {
 371                s = pool_calloc(1, sizeof(struct mark_set));
 372                s->shift = marks->shift + 10;
 373                s->data.sets[0] = marks;
 374                marks = s;
 375        }
 376        while (s->shift) {
 377                unsigned long i = idnum >> s->shift;
 378                idnum -= i << s->shift;
 379                if (!s->data.sets[i]) {
 380                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 381                        s->data.sets[i]->shift = s->shift - 10;
 382                }
 383                s = s->data.sets[i];
 384        }
 385        if (!s->data.marked[idnum])
 386                marks_set_count++;
 387        s->data.marked[idnum] = oe;
 388}
 389
 390static struct object_entry* find_mark(unsigned long idnum)
 391{
 392        unsigned long orig_idnum = idnum;
 393        struct mark_set *s = marks;
 394        struct object_entry *oe = NULL;
 395        if ((idnum >> s->shift) < 1024) {
 396                while (s && s->shift) {
 397                        unsigned long i = idnum >> s->shift;
 398                        idnum -= i << s->shift;
 399                        s = s->data.sets[i];
 400                }
 401                if (s)
 402                        oe = s->data.marked[idnum];
 403        }
 404        if (!oe)
 405                die("mark :%lu not declared", orig_idnum);
 406        return oe;
 407}
 408
 409static struct atom_str* to_atom(const char *s, size_t len)
 410{
 411        unsigned int hc = hc_str(s, len) % atom_table_sz;
 412        struct atom_str *c;
 413
 414        for (c = atom_table[hc]; c; c = c->next_atom)
 415                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 416                        return c;
 417
 418        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 419        c->str_len = len;
 420        strncpy(c->str_dat, s, len);
 421        c->str_dat[len] = 0;
 422        c->next_atom = atom_table[hc];
 423        atom_table[hc] = c;
 424        atom_cnt++;
 425        return c;
 426}
 427
 428static struct branch* lookup_branch(const char *name)
 429{
 430        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 431        struct branch *b;
 432
 433        for (b = branch_table[hc]; b; b = b->table_next_branch)
 434                if (!strcmp(name, b->name))
 435                        return b;
 436        return NULL;
 437}
 438
 439static struct branch* new_branch(const char *name)
 440{
 441        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 442        struct branch* b = lookup_branch(name);
 443
 444        if (b)
 445                die("Invalid attempt to create duplicate branch: %s", name);
 446        if (check_ref_format(name))
 447                die("Branch name doesn't conform to GIT standards: %s", name);
 448
 449        b = pool_calloc(1, sizeof(struct branch));
 450        b->name = pool_strdup(name);
 451        b->table_next_branch = branch_table[hc];
 452        branch_table[hc] = b;
 453        branch_count++;
 454        return b;
 455}
 456
 457static unsigned int hc_entries(unsigned int cnt)
 458{
 459        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 460        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 461}
 462
 463static struct tree_content* new_tree_content(unsigned int cnt)
 464{
 465        struct avail_tree_content *f, *l = NULL;
 466        struct tree_content *t;
 467        unsigned int hc = hc_entries(cnt);
 468
 469        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 470                if (f->entry_capacity >= cnt)
 471                        break;
 472
 473        if (f) {
 474                if (l)
 475                        l->next_avail = f->next_avail;
 476                else
 477                        avail_tree_table[hc] = f->next_avail;
 478        } else {
 479                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 480                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 481                f->entry_capacity = cnt;
 482        }
 483
 484        t = (struct tree_content*)f;
 485        t->entry_count = 0;
 486        return t;
 487}
 488
 489static void release_tree_entry(struct tree_entry *e);
 490static void release_tree_content(struct tree_content *t)
 491{
 492        struct avail_tree_content *f = (struct avail_tree_content*)t;
 493        unsigned int hc = hc_entries(f->entry_capacity);
 494        f->next_avail = avail_tree_table[hc];
 495        avail_tree_table[hc] = f;
 496}
 497
 498static void release_tree_content_recursive(struct tree_content *t)
 499{
 500        unsigned int i;
 501        for (i = 0; i < t->entry_count; i++)
 502                release_tree_entry(t->entries[i]);
 503        release_tree_content(t);
 504}
 505
 506static struct tree_content* grow_tree_content(
 507        struct tree_content *t,
 508        int amt)
 509{
 510        struct tree_content *r = new_tree_content(t->entry_count + amt);
 511        r->entry_count = t->entry_count;
 512        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 513        release_tree_content(t);
 514        return r;
 515}
 516
 517static struct tree_entry* new_tree_entry()
 518{
 519        struct tree_entry *e;
 520
 521        if (!avail_tree_entry) {
 522                unsigned int n = tree_entry_alloc;
 523                total_allocd += n * sizeof(struct tree_entry);
 524                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 525                while (n-- > 1) {
 526                        *((void**)e) = e + 1;
 527                        e++;
 528                }
 529                *((void**)e) = NULL;
 530        }
 531
 532        e = avail_tree_entry;
 533        avail_tree_entry = *((void**)e);
 534        return e;
 535}
 536
 537static void release_tree_entry(struct tree_entry *e)
 538{
 539        if (e->tree)
 540                release_tree_content_recursive(e->tree);
 541        *((void**)e) = avail_tree_entry;
 542        avail_tree_entry = e;
 543}
 544
 545static void yread(int fd, void *buffer, size_t length)
 546{
 547        ssize_t ret = 0;
 548        while (ret < length) {
 549                ssize_t size = xread(fd, (char *) buffer + ret, length - ret);
 550                if (!size)
 551                        die("Read from descriptor %i: end of stream", fd);
 552                if (size < 0)
 553                        die("Read from descriptor %i: %s", fd, strerror(errno));
 554                ret += size;
 555        }
 556}
 557
 558static void ywrite(int fd, void *buffer, size_t length)
 559{
 560        ssize_t ret = 0;
 561        while (ret < length) {
 562                ssize_t size = xwrite(fd, (char *) buffer + ret, length - ret);
 563                if (!size)
 564                        die("Write to descriptor %i: end of file", fd);
 565                if (size < 0)
 566                        die("Write to descriptor %i: %s", fd, strerror(errno));
 567                ret += size;
 568        }
 569}
 570
 571static size_t encode_header(
 572        enum object_type type,
 573        size_t size,
 574        unsigned char *hdr)
 575{
 576        int n = 1;
 577        unsigned char c;
 578
 579        if (type < OBJ_COMMIT || type > OBJ_DELTA)
 580                die("bad type %d", type);
 581
 582        c = (type << 4) | (size & 15);
 583        size >>= 4;
 584        while (size) {
 585                *hdr++ = c | 0x80;
 586                c = size & 0x7f;
 587                size >>= 7;
 588                n++;
 589        }
 590        *hdr = c;
 591        return n;
 592}
 593
 594static int store_object(
 595        enum object_type type,
 596        void *dat,
 597        size_t datlen,
 598        struct last_object *last,
 599        unsigned char *sha1out,
 600        unsigned long mark)
 601{
 602        void *out, *delta;
 603        struct object_entry *e;
 604        unsigned char hdr[96];
 605        unsigned char sha1[20];
 606        unsigned long hdrlen, deltalen;
 607        SHA_CTX c;
 608        z_stream s;
 609
 610        hdrlen = sprintf((char*)hdr,"%s %lu",type_names[type],datlen) + 1;
 611        SHA1_Init(&c);
 612        SHA1_Update(&c, hdr, hdrlen);
 613        SHA1_Update(&c, dat, datlen);
 614        SHA1_Final(sha1, &c);
 615        if (sha1out)
 616                memcpy(sha1out, sha1, sizeof(sha1));
 617
 618        e = insert_object(sha1);
 619        if (mark)
 620                insert_mark(mark, e);
 621        if (e->offset) {
 622                duplicate_count++;
 623                duplicate_count_by_type[type]++;
 624                return 1;
 625        }
 626        e->type = type;
 627        e->offset = pack_size;
 628        object_count++;
 629        object_count_by_type[type]++;
 630
 631        if (last && last->data && last->depth < max_depth)
 632                delta = diff_delta(last->data, last->len,
 633                        dat, datlen,
 634                        &deltalen, 0);
 635        else
 636                delta = 0;
 637
 638        memset(&s, 0, sizeof(s));
 639        deflateInit(&s, zlib_compression_level);
 640
 641        if (delta) {
 642                last->depth++;
 643                s.next_in = delta;
 644                s.avail_in = deltalen;
 645                hdrlen = encode_header(OBJ_DELTA, deltalen, hdr);
 646                ywrite(pack_fd, hdr, hdrlen);
 647                ywrite(pack_fd, last->sha1, sizeof(sha1));
 648                pack_size += hdrlen + sizeof(sha1);
 649        } else {
 650                if (last)
 651                        last->depth = 0;
 652                s.next_in = dat;
 653                s.avail_in = datlen;
 654                hdrlen = encode_header(type, datlen, hdr);
 655                ywrite(pack_fd, hdr, hdrlen);
 656                pack_size += hdrlen;
 657        }
 658
 659        s.avail_out = deflateBound(&s, s.avail_in);
 660        s.next_out = out = xmalloc(s.avail_out);
 661        while (deflate(&s, Z_FINISH) == Z_OK)
 662                /* nothing */;
 663        deflateEnd(&s);
 664
 665        ywrite(pack_fd, out, s.total_out);
 666        pack_size += s.total_out;
 667
 668        free(out);
 669        if (delta)
 670                free(delta);
 671        if (last) {
 672                if (last->data)
 673                        free(last->data);
 674                last->data = dat;
 675                last->len = datlen;
 676                memcpy(last->sha1, sha1, sizeof(sha1));
 677        }
 678        return 0;
 679}
 680
 681static unsigned char* map_pack(unsigned long offset, unsigned int *left)
 682{
 683        if (offset >= pack_size)
 684                die("object offset outside of pack file");
 685        if (!pack_base
 686                        || offset < pack_moff
 687                        || (offset + 20) >= (pack_moff + pack_mlen)) {
 688                if (pack_base)
 689                        munmap(pack_base, pack_mlen);
 690                pack_moff = (offset / page_size) * page_size;
 691                pack_base = mmap(NULL,pack_mlen,PROT_READ,MAP_SHARED,
 692                        pack_fd,pack_moff);
 693                if (pack_base == MAP_FAILED)
 694                        die("Failed to map generated pack: %s", strerror(errno));
 695                remap_count++;
 696        }
 697        offset -= pack_moff;
 698        if (left)
 699                *left = pack_mlen - offset;
 700        return pack_base + offset;
 701}
 702
 703static unsigned long unpack_object_header(unsigned long offset,
 704        enum object_type *type,
 705        unsigned long *sizep)
 706{
 707        unsigned shift;
 708        unsigned char c;
 709        unsigned long size;
 710
 711        c = *map_pack(offset++, NULL);
 712        *type = (c >> 4) & 7;
 713        size = c & 15;
 714        shift = 4;
 715        while (c & 0x80) {
 716                c = *map_pack(offset++, NULL);
 717                size += (c & 0x7f) << shift;
 718                shift += 7;
 719        }
 720        *sizep = size;
 721        return offset;
 722}
 723
 724static void *unpack_non_delta_entry(unsigned long o, unsigned long sz)
 725{
 726        z_stream stream;
 727        unsigned char *result;
 728
 729        result = xmalloc(sz + 1);
 730        result[sz] = 0;
 731
 732        memset(&stream, 0, sizeof(stream));
 733        stream.next_in = map_pack(o, &stream.avail_in);
 734        stream.next_out = result;
 735        stream.avail_out = sz;
 736
 737        inflateInit(&stream);
 738        for (;;) {
 739                int st = inflate(&stream, Z_FINISH);
 740                if (st == Z_STREAM_END)
 741                        break;
 742                if (st == Z_OK || st == Z_BUF_ERROR) {
 743                        o = stream.next_in - pack_base + pack_moff;
 744                        stream.next_in = map_pack(o, &stream.avail_in);
 745                        continue;
 746                }
 747                die("Error %i from zlib during inflate.", st);
 748        }
 749        inflateEnd(&stream);
 750        if (stream.total_out != sz)
 751                die("Error after inflate: sizes mismatch");
 752        return result;
 753}
 754
 755static void *unpack_entry(unsigned long offset, unsigned long *sizep);
 756
 757static void *unpack_delta_entry(unsigned long offset,
 758        unsigned long delta_size,
 759        unsigned long *sizep)
 760{
 761        struct object_entry *base_oe;
 762        unsigned char *base_sha1;
 763        void *delta_data, *base, *result;
 764        unsigned long base_size, result_size;
 765
 766        base_sha1 = map_pack(offset, NULL);
 767        base_oe = find_object(base_sha1);
 768        if (!base_oe)
 769                die("I'm broken; I can't find a base I know must be here.");
 770        base = unpack_entry(base_oe->offset, &base_size);
 771        delta_data = unpack_non_delta_entry(offset + 20, delta_size);
 772        result = patch_delta(base, base_size,
 773                             delta_data, delta_size,
 774                             &result_size);
 775        if (!result)
 776                die("failed to apply delta");
 777        free(delta_data);
 778        free(base);
 779        *sizep = result_size;
 780        return result;
 781}
 782
 783static void *unpack_entry(unsigned long offset, unsigned long *sizep)
 784{
 785        unsigned long size;
 786        enum object_type kind;
 787
 788        offset = unpack_object_header(offset, &kind, &size);
 789        switch (kind) {
 790        case OBJ_DELTA:
 791                return unpack_delta_entry(offset, size, sizep);
 792        case OBJ_COMMIT:
 793        case OBJ_TREE:
 794        case OBJ_BLOB:
 795        case OBJ_TAG:
 796                *sizep = size;
 797                return unpack_non_delta_entry(offset, size);
 798        default:
 799                die("I created an object I can't read!");
 800        }
 801}
 802
 803static const char *get_mode(const char *str, unsigned int *modep)
 804{
 805        unsigned char c;
 806        unsigned int mode = 0;
 807
 808        while ((c = *str++) != ' ') {
 809                if (c < '0' || c > '7')
 810                        return NULL;
 811                mode = (mode << 3) + (c - '0');
 812        }
 813        *modep = mode;
 814        return str;
 815}
 816
 817static void load_tree(struct tree_entry *root)
 818{
 819        struct object_entry *myoe;
 820        struct tree_content *t;
 821        unsigned long size;
 822        char *buf;
 823        const char *c;
 824
 825        root->tree = t = new_tree_content(8);
 826        if (!memcmp(root->sha1, null_sha1, 20))
 827                return;
 828
 829        myoe = find_object(root->sha1);
 830        if (myoe) {
 831                if (myoe->type != OBJ_TREE)
 832                        die("Not a tree: %s", sha1_to_hex(root->sha1));
 833                buf = unpack_entry(myoe->offset, &size);
 834        } else {
 835                char type[20];
 836                buf = read_sha1_file(root->sha1, type, &size);
 837                if (!buf || strcmp(type, tree_type))
 838                        die("Can't load tree %s", sha1_to_hex(root->sha1));
 839        }
 840
 841        c = buf;
 842        while (c != (buf + size)) {
 843                struct tree_entry *e = new_tree_entry();
 844
 845                if (t->entry_count == t->entry_capacity)
 846                        root->tree = t = grow_tree_content(t, 8);
 847                t->entries[t->entry_count++] = e;
 848
 849                e->tree = NULL;
 850                c = get_mode(c, &e->mode);
 851                if (!c)
 852                        die("Corrupt mode in %s", sha1_to_hex(root->sha1));
 853                e->name = to_atom(c, strlen(c));
 854                c += e->name->str_len + 1;
 855                memcpy(e->sha1, c, sizeof(e->sha1));
 856                c += 20;
 857        }
 858        free(buf);
 859}
 860
 861static int tecmp (const void *_a, const void *_b)
 862{
 863        struct tree_entry *a = *((struct tree_entry**)_a);
 864        struct tree_entry *b = *((struct tree_entry**)_b);
 865        return base_name_compare(
 866                a->name->str_dat, a->name->str_len, a->mode,
 867                b->name->str_dat, b->name->str_len, b->mode);
 868}
 869
 870static void store_tree(struct tree_entry *root)
 871{
 872        struct tree_content *t = root->tree;
 873        unsigned int i;
 874        size_t maxlen;
 875        char *buf, *c;
 876
 877        if (memcmp(root->sha1, null_sha1, 20))
 878                return;
 879
 880        maxlen = 0;
 881        for (i = 0; i < t->entry_count; i++) {
 882                maxlen += t->entries[i]->name->str_len + 34;
 883                if (t->entries[i]->tree)
 884                        store_tree(t->entries[i]);
 885        }
 886
 887        qsort(t->entries, t->entry_count, sizeof(t->entries[0]), tecmp);
 888        buf = c = xmalloc(maxlen);
 889        for (i = 0; i < t->entry_count; i++) {
 890                struct tree_entry *e = t->entries[i];
 891                c += sprintf(c, "%o", e->mode);
 892                *c++ = ' ';
 893                strcpy(c, e->name->str_dat);
 894                c += e->name->str_len + 1;
 895                memcpy(c, e->sha1, 20);
 896                c += 20;
 897        }
 898        store_object(OBJ_TREE, buf, c - buf, NULL, root->sha1, 0);
 899        free(buf);
 900}
 901
 902static int tree_content_set(
 903        struct tree_entry *root,
 904        const char *p,
 905        const unsigned char *sha1,
 906        const unsigned int mode)
 907{
 908        struct tree_content *t = root->tree;
 909        const char *slash1;
 910        unsigned int i, n;
 911        struct tree_entry *e;
 912
 913        slash1 = strchr(p, '/');
 914        if (slash1)
 915                n = slash1 - p;
 916        else
 917                n = strlen(p);
 918
 919        for (i = 0; i < t->entry_count; i++) {
 920                e = t->entries[i];
 921                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
 922                        if (!slash1) {
 923                                if (e->mode == mode && !memcmp(e->sha1, sha1, 20))
 924                                        return 0;
 925                                e->mode = mode;
 926                                memcpy(e->sha1, sha1, 20);
 927                                if (e->tree) {
 928                                        release_tree_content_recursive(e->tree);
 929                                        e->tree = NULL;
 930                                }
 931                                memcpy(root->sha1, null_sha1, 20);
 932                                return 1;
 933                        }
 934                        if (!S_ISDIR(e->mode)) {
 935                                e->tree = new_tree_content(8);
 936                                e->mode = S_IFDIR;
 937                        }
 938                        if (!e->tree)
 939                                load_tree(e);
 940                        if (tree_content_set(e, slash1 + 1, sha1, mode)) {
 941                                memcpy(root->sha1, null_sha1, 20);
 942                                return 1;
 943                        }
 944                        return 0;
 945                }
 946        }
 947
 948        if (t->entry_count == t->entry_capacity)
 949                root->tree = t = grow_tree_content(t, 8);
 950        e = new_tree_entry();
 951        e->name = to_atom(p, n);
 952        t->entries[t->entry_count++] = e;
 953        if (slash1) {
 954                e->tree = new_tree_content(8);
 955                e->mode = S_IFDIR;
 956                tree_content_set(e, slash1 + 1, sha1, mode);
 957        } else {
 958                e->tree = NULL;
 959                e->mode = mode;
 960                memcpy(e->sha1, sha1, 20);
 961        }
 962        memcpy(root->sha1, null_sha1, 20);
 963        return 1;
 964}
 965
 966static int tree_content_remove(struct tree_entry *root, const char *p)
 967{
 968        struct tree_content *t = root->tree;
 969        const char *slash1;
 970        unsigned int i, n;
 971        struct tree_entry *e;
 972
 973        slash1 = strchr(p, '/');
 974        if (slash1)
 975                n = slash1 - p;
 976        else
 977                n = strlen(p);
 978
 979        for (i = 0; i < t->entry_count; i++) {
 980                e = t->entries[i];
 981                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
 982                        if (!slash1 || !S_ISDIR(e->mode))
 983                                goto del_entry;
 984                        if (!e->tree)
 985                                load_tree(e);
 986                        if (tree_content_remove(e, slash1 + 1)) {
 987                                if (!e->tree->entry_count)
 988                                        goto del_entry;
 989                                memcpy(root->sha1, null_sha1, 20);
 990                                return 1;
 991                        }
 992                        return 0;
 993                }
 994        }
 995        return 0;
 996
 997del_entry:
 998        for (i++; i < t->entry_count; i++)
 999                t->entries[i-1] = t->entries[i];
1000        t->entry_count--;
1001        release_tree_entry(e);
1002        memcpy(root->sha1, null_sha1, 20);
1003        return 1;
1004}
1005
1006static void init_pack_header()
1007{
1008        struct pack_header hdr;
1009
1010        hdr.hdr_signature = htonl(PACK_SIGNATURE);
1011        hdr.hdr_version = htonl(2);
1012        hdr.hdr_entries = 0;
1013
1014        ywrite(pack_fd, &hdr, sizeof(hdr));
1015        pack_size = sizeof(hdr);
1016}
1017
1018static void fixup_header_footer()
1019{
1020        SHA_CTX c;
1021        char hdr[8];
1022        unsigned long cnt;
1023        char *buf;
1024        size_t n;
1025
1026        if (lseek(pack_fd, 0, SEEK_SET) != 0)
1027                die("Failed seeking to start: %s", strerror(errno));
1028
1029        SHA1_Init(&c);
1030        yread(pack_fd, hdr, 8);
1031        SHA1_Update(&c, hdr, 8);
1032
1033        cnt = htonl(object_count);
1034        SHA1_Update(&c, &cnt, 4);
1035        ywrite(pack_fd, &cnt, 4);
1036
1037        buf = xmalloc(128 * 1024);
1038        for (;;) {
1039                n = xread(pack_fd, buf, 128 * 1024);
1040                if (n <= 0)
1041                        break;
1042                SHA1_Update(&c, buf, n);
1043        }
1044        free(buf);
1045
1046        SHA1_Final(pack_sha1, &c);
1047        ywrite(pack_fd, pack_sha1, sizeof(pack_sha1));
1048}
1049
1050static int oecmp (const void *_a, const void *_b)
1051{
1052        struct object_entry *a = *((struct object_entry**)_a);
1053        struct object_entry *b = *((struct object_entry**)_b);
1054        return memcmp(a->sha1, b->sha1, sizeof(a->sha1));
1055}
1056
1057static void write_index(const char *idx_name)
1058{
1059        struct sha1file *f;
1060        struct object_entry **idx, **c, **last;
1061        struct object_entry *e;
1062        struct object_entry_pool *o;
1063        unsigned int array[256];
1064        int i;
1065
1066        /* Build the sorted table of object IDs. */
1067        idx = xmalloc(object_count * sizeof(struct object_entry*));
1068        c = idx;
1069        for (o = blocks; o; o = o->next_pool)
1070                for (e = o->entries; e != o->next_free; e++)
1071                        *c++ = e;
1072        last = idx + object_count;
1073        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
1074
1075        /* Generate the fan-out array. */
1076        c = idx;
1077        for (i = 0; i < 256; i++) {
1078                struct object_entry **next = c;;
1079                while (next < last) {
1080                        if ((*next)->sha1[0] != i)
1081                                break;
1082                        next++;
1083                }
1084                array[i] = htonl(next - idx);
1085                c = next;
1086        }
1087
1088        f = sha1create("%s", idx_name);
1089        sha1write(f, array, 256 * sizeof(int));
1090        for (c = idx; c != last; c++) {
1091                unsigned int offset = htonl((*c)->offset);
1092                sha1write(f, &offset, 4);
1093                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
1094        }
1095        sha1write(f, pack_sha1, sizeof(pack_sha1));
1096        sha1close(f, NULL, 1);
1097        free(idx);
1098}
1099
1100static void dump_branches()
1101{
1102        static const char *msg = "fast-import";
1103        unsigned int i;
1104        struct branch *b;
1105        struct ref_lock *lock;
1106
1107        for (i = 0; i < branch_table_sz; i++) {
1108                for (b = branch_table[i]; b; b = b->table_next_branch) {
1109                        lock = lock_any_ref_for_update(b->name, NULL, 0);
1110                        if (!lock || write_ref_sha1(lock, b->sha1, msg) < 0)
1111                                die("Can't write %s", b->name);
1112                }
1113        }
1114}
1115
1116static void dump_tags()
1117{
1118        static const char *msg = "fast-import";
1119        struct tag *t;
1120        struct ref_lock *lock;
1121        char path[PATH_MAX];
1122
1123        for (t = first_tag; t; t = t->next_tag) {
1124                sprintf(path, "refs/tags/%s", t->name);
1125                lock = lock_any_ref_for_update(path, NULL, 0);
1126                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1127                        die("Can't write %s", path);
1128        }
1129}
1130
1131static void dump_marks_helper(FILE *f,
1132        unsigned long base,
1133        struct mark_set *m)
1134{
1135        int k;
1136        if (m->shift) {
1137                for (k = 0; k < 1024; k++) {
1138                        if (m->data.sets[k])
1139                                dump_marks_helper(f, (base + k) << m->shift,
1140                                        m->data.sets[k]);
1141                }
1142        } else {
1143                for (k = 0; k < 1024; k++) {
1144                        if (m->data.marked[k])
1145                                fprintf(f, ":%lu %s\n", base + k,
1146                                        sha1_to_hex(m->data.marked[k]->sha1));
1147                }
1148        }
1149}
1150
1151static void dump_marks()
1152{
1153        if (mark_file)
1154        {
1155                FILE *f = fopen(mark_file, "w");
1156                dump_marks_helper(f, 0, marks);
1157                fclose(f);
1158        }
1159}
1160
1161static void read_next_command()
1162{
1163        read_line(&command_buf, stdin, '\n');
1164}
1165
1166static void cmd_mark()
1167{
1168        if (!strncmp("mark :", command_buf.buf, 6)) {
1169                next_mark = strtoul(command_buf.buf + 6, NULL, 10);
1170                read_next_command();
1171        }
1172        else
1173                next_mark = 0;
1174}
1175
1176static void* cmd_data (size_t *size)
1177{
1178        size_t n = 0;
1179        void *buffer;
1180        size_t length;
1181
1182        if (strncmp("data ", command_buf.buf, 5))
1183                die("Expected 'data n' command, found: %s", command_buf.buf);
1184
1185        length = strtoul(command_buf.buf + 5, NULL, 10);
1186        buffer = xmalloc(length);
1187
1188        while (n < length) {
1189                size_t s = fread((char*)buffer + n, 1, length - n, stdin);
1190                if (!s && feof(stdin))
1191                        die("EOF in data (%lu bytes remaining)", length - n);
1192                n += s;
1193        }
1194
1195        if (fgetc(stdin) != '\n')
1196                die("An lf did not trail the binary data as expected.");
1197
1198        *size = length;
1199        return buffer;
1200}
1201
1202static void cmd_new_blob()
1203{
1204        size_t l;
1205        void *d;
1206
1207        read_next_command();
1208        cmd_mark();
1209        d = cmd_data(&l);
1210
1211        if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
1212                free(d);
1213}
1214
1215static void unload_one_branch()
1216{
1217        while (cur_active_branches
1218                && cur_active_branches >= max_active_branches) {
1219                unsigned long min_commit = ULONG_MAX;
1220                struct branch *e, *l = NULL, *p = NULL;
1221
1222                for (e = active_branches; e; e = e->active_next_branch) {
1223                        if (e->last_commit < min_commit) {
1224                                p = l;
1225                                min_commit = e->last_commit;
1226                        }
1227                        l = e;
1228                }
1229
1230                if (p) {
1231                        e = p->active_next_branch;
1232                        p->active_next_branch = e->active_next_branch;
1233                } else {
1234                        e = active_branches;
1235                        active_branches = e->active_next_branch;
1236                }
1237                e->active_next_branch = NULL;
1238                if (e->branch_tree.tree) {
1239                        release_tree_content_recursive(e->branch_tree.tree);
1240                        e->branch_tree.tree = NULL;
1241                }
1242                cur_active_branches--;
1243        }
1244}
1245
1246static void load_branch(struct branch *b)
1247{
1248        load_tree(&b->branch_tree);
1249        b->active_next_branch = active_branches;
1250        active_branches = b;
1251        cur_active_branches++;
1252        branch_load_count++;
1253}
1254
1255static void file_change_m(struct branch *b)
1256{
1257        const char *p = command_buf.buf + 2;
1258        char *p_uq;
1259        const char *endp;
1260        struct object_entry *oe;
1261        unsigned char sha1[20];
1262        unsigned int mode;
1263        char type[20];
1264
1265        p = get_mode(p, &mode);
1266        if (!p)
1267                die("Corrupt mode: %s", command_buf.buf);
1268        switch (mode) {
1269        case S_IFREG | 0644:
1270        case S_IFREG | 0755:
1271        case S_IFLNK:
1272        case 0644:
1273        case 0755:
1274                /* ok */
1275                break;
1276        default:
1277                die("Corrupt mode: %s", command_buf.buf);
1278        }
1279
1280        if (*p == ':') {
1281                char *x;
1282                oe = find_mark(strtoul(p + 1, &x, 10));
1283                p = x;
1284        } else {
1285                if (get_sha1_hex(p, sha1))
1286                        die("Invalid SHA1: %s", command_buf.buf);
1287                oe = find_object(sha1);
1288                p += 40;
1289        }
1290        if (*p++ != ' ')
1291                die("Missing space after SHA1: %s", command_buf.buf);
1292
1293        p_uq = unquote_c_style(p, &endp);
1294        if (p_uq) {
1295                if (*endp)
1296                        die("Garbage after path in: %s", command_buf.buf);
1297                p = p_uq;
1298        }
1299
1300        if (oe) {
1301                if (oe->type != OBJ_BLOB)
1302                        die("Not a blob (actually a %s): %s",
1303                                command_buf.buf, type_names[oe->type]);
1304        } else {
1305                if (sha1_object_info(sha1, type, NULL))
1306                        die("Blob not found: %s", command_buf.buf);
1307                if (strcmp(blob_type, type))
1308                        die("Not a blob (actually a %s): %s",
1309                                command_buf.buf, type);
1310        }
1311
1312        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
1313
1314        if (p_uq)
1315                free(p_uq);
1316}
1317
1318static void file_change_d(struct branch *b)
1319{
1320        const char *p = command_buf.buf + 2;
1321        char *p_uq;
1322        const char *endp;
1323
1324        p_uq = unquote_c_style(p, &endp);
1325        if (p_uq) {
1326                if (*endp)
1327                        die("Garbage after path in: %s", command_buf.buf);
1328                p = p_uq;
1329        }
1330        tree_content_remove(&b->branch_tree, p);
1331        if (p_uq)
1332                free(p_uq);
1333}
1334
1335static void cmd_from(struct branch *b)
1336{
1337        const char *from, *endp;
1338        char *str_uq;
1339        struct branch *s;
1340
1341        if (strncmp("from ", command_buf.buf, 5))
1342                return;
1343
1344        if (b->last_commit)
1345                die("Can't reinitailize branch %s", b->name);
1346
1347        from = strchr(command_buf.buf, ' ') + 1;
1348        str_uq = unquote_c_style(from, &endp);
1349        if (str_uq) {
1350                if (*endp)
1351                        die("Garbage after string in: %s", command_buf.buf);
1352                from = str_uq;
1353        }
1354
1355        s = lookup_branch(from);
1356        if (b == s)
1357                die("Can't create a branch from itself: %s", b->name);
1358        else if (s) {
1359                memcpy(b->sha1, s->sha1, 20);
1360                memcpy(b->branch_tree.sha1, s->branch_tree.sha1, 20);
1361        } else if (*from == ':') {
1362                unsigned long idnum = strtoul(from + 1, NULL, 10);
1363                struct object_entry *oe = find_mark(idnum);
1364                unsigned long size;
1365                char *buf;
1366                if (oe->type != OBJ_COMMIT)
1367                        die("Mark :%lu not a commit", idnum);
1368                memcpy(b->sha1, oe->sha1, 20);
1369                buf = unpack_entry(oe->offset, &size);
1370                if (!buf || size < 46)
1371                        die("Not a valid commit: %s", from);
1372                if (memcmp("tree ", buf, 5)
1373                        || get_sha1_hex(buf + 5, b->branch_tree.sha1))
1374                        die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1375                free(buf);
1376        } else if (!get_sha1(from, b->sha1)) {
1377                if (!memcmp(b->sha1, null_sha1, 20))
1378                        memcpy(b->branch_tree.sha1, null_sha1, 20);
1379                else {
1380                        unsigned long size;
1381                        char *buf;
1382
1383                        buf = read_object_with_reference(b->sha1,
1384                                type_names[OBJ_COMMIT], &size, b->sha1);
1385                        if (!buf || size < 46)
1386                                die("Not a valid commit: %s", from);
1387                        if (memcmp("tree ", buf, 5)
1388                                || get_sha1_hex(buf + 5, b->branch_tree.sha1))
1389                                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1390                        free(buf);
1391                }
1392        } else
1393                die("Invalid ref name or SHA1 expression: %s", from);
1394
1395        read_next_command();
1396}
1397
1398static void cmd_new_commit()
1399{
1400        struct branch *b;
1401        void *msg;
1402        size_t msglen;
1403        char *str_uq;
1404        const char *endp;
1405        char *sp;
1406        char *author = NULL;
1407        char *committer = NULL;
1408        char *body;
1409
1410        /* Obtain the branch name from the rest of our command */
1411        sp = strchr(command_buf.buf, ' ') + 1;
1412        str_uq = unquote_c_style(sp, &endp);
1413        if (str_uq) {
1414                if (*endp)
1415                        die("Garbage after ref in: %s", command_buf.buf);
1416                sp = str_uq;
1417        }
1418        b = lookup_branch(sp);
1419        if (!b)
1420                b = new_branch(sp);
1421        if (str_uq)
1422                free(str_uq);
1423
1424        read_next_command();
1425        cmd_mark();
1426        if (!strncmp("author ", command_buf.buf, 7)) {
1427                author = strdup(command_buf.buf);
1428                read_next_command();
1429        }
1430        if (!strncmp("committer ", command_buf.buf, 10)) {
1431                committer = strdup(command_buf.buf);
1432                read_next_command();
1433        }
1434        if (!committer)
1435                die("Expected committer but didn't get one");
1436        msg = cmd_data(&msglen);
1437        read_next_command();
1438        cmd_from(b);
1439
1440        /* ensure the branch is active/loaded */
1441        if (!b->branch_tree.tree || !max_active_branches) {
1442                unload_one_branch();
1443                load_branch(b);
1444        }
1445
1446        /* file_change* */
1447        for (;;) {
1448                if (1 == command_buf.len)
1449                        break;
1450                else if (!strncmp("M ", command_buf.buf, 2))
1451                        file_change_m(b);
1452                else if (!strncmp("D ", command_buf.buf, 2))
1453                        file_change_d(b);
1454                else
1455                        die("Unsupported file_change: %s", command_buf.buf);
1456                read_next_command();
1457        }
1458
1459        /* build the tree and the commit */
1460        store_tree(&b->branch_tree);
1461        body = xmalloc(97 + msglen
1462                + (author
1463                        ? strlen(author) + strlen(committer)
1464                        : 2 * strlen(committer)));
1465        sp = body;
1466        sp += sprintf(sp, "tree %s\n", sha1_to_hex(b->branch_tree.sha1));
1467        if (memcmp(b->sha1, null_sha1, 20))
1468                sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
1469        if (author)
1470                sp += sprintf(sp, "%s\n", author);
1471        else
1472                sp += sprintf(sp, "author %s\n", committer + 10);
1473        sp += sprintf(sp, "%s\n\n", committer);
1474        memcpy(sp, msg, msglen);
1475        sp += msglen;
1476        if (author)
1477                free(author);
1478        free(committer);
1479        free(msg);
1480
1481        store_object(OBJ_COMMIT, body, sp - body, NULL, b->sha1, next_mark);
1482        free(body);
1483        b->last_commit = object_count_by_type[OBJ_COMMIT];
1484
1485        if (branch_log) {
1486                int need_dq = quote_c_style(b->name, NULL, NULL, 0);
1487                fprintf(branch_log, "commit ");
1488                if (need_dq) {
1489                        fputc('"', branch_log);
1490                        quote_c_style(b->name, NULL, branch_log, 0);
1491                        fputc('"', branch_log);
1492                } else
1493                        fprintf(branch_log, "%s", b->name);
1494                fprintf(branch_log," :%lu %s\n",next_mark,sha1_to_hex(b->sha1));
1495        }
1496}
1497
1498static void cmd_new_tag()
1499{
1500        char *str_uq;
1501        const char *endp;
1502        char *sp;
1503        const char *from;
1504        char *tagger;
1505        struct branch *s;
1506        void *msg;
1507        size_t msglen;
1508        char *body;
1509        struct tag *t;
1510        unsigned long from_mark = 0;
1511        unsigned char sha1[20];
1512
1513        /* Obtain the new tag name from the rest of our command */
1514        sp = strchr(command_buf.buf, ' ') + 1;
1515        str_uq = unquote_c_style(sp, &endp);
1516        if (str_uq) {
1517                if (*endp)
1518                        die("Garbage after tag name in: %s", command_buf.buf);
1519                sp = str_uq;
1520        }
1521        t = pool_alloc(sizeof(struct tag));
1522        t->next_tag = NULL;
1523        t->name = pool_strdup(sp);
1524        if (last_tag)
1525                last_tag->next_tag = t;
1526        else
1527                first_tag = t;
1528        last_tag = t;
1529        if (str_uq)
1530                free(str_uq);
1531        read_next_command();
1532
1533        /* from ... */
1534        if (strncmp("from ", command_buf.buf, 5))
1535                die("Expected from command, got %s", command_buf.buf);
1536
1537        from = strchr(command_buf.buf, ' ') + 1;
1538        str_uq = unquote_c_style(from, &endp);
1539        if (str_uq) {
1540                if (*endp)
1541                        die("Garbage after string in: %s", command_buf.buf);
1542                from = str_uq;
1543        }
1544
1545        s = lookup_branch(from);
1546        if (s) {
1547                memcpy(sha1, s->sha1, 20);
1548        } else if (*from == ':') {
1549                from_mark = strtoul(from + 1, NULL, 10);
1550                struct object_entry *oe = find_mark(from_mark);
1551                if (oe->type != OBJ_COMMIT)
1552                        die("Mark :%lu not a commit", from_mark);
1553                memcpy(sha1, oe->sha1, 20);
1554        } else if (!get_sha1(from, sha1)) {
1555                unsigned long size;
1556                char *buf;
1557
1558                buf = read_object_with_reference(sha1,
1559                        type_names[OBJ_COMMIT], &size, sha1);
1560                if (!buf || size < 46)
1561                        die("Not a valid commit: %s", from);
1562                free(buf);
1563        } else
1564                die("Invalid ref name or SHA1 expression: %s", from);
1565
1566        if (str_uq)
1567                free(str_uq);
1568        read_next_command();
1569
1570        /* tagger ... */
1571        if (strncmp("tagger ", command_buf.buf, 7))
1572                die("Expected tagger command, got %s", command_buf.buf);
1573        tagger = strdup(command_buf.buf);
1574
1575        /* tag payload/message */
1576        read_next_command();
1577        msg = cmd_data(&msglen);
1578
1579        /* build the tag object */
1580        body = xmalloc(67 + strlen(t->name) + strlen(tagger) + msglen);
1581        sp = body;
1582        sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
1583        sp += sprintf(sp, "type %s\n", type_names[OBJ_COMMIT]);
1584        sp += sprintf(sp, "tag %s\n", t->name);
1585        sp += sprintf(sp, "%s\n\n", tagger);
1586        memcpy(sp, msg, msglen);
1587        sp += msglen;
1588        free(tagger);
1589        free(msg);
1590
1591        store_object(OBJ_TAG, body, sp - body, NULL, t->sha1, 0);
1592        free(body);
1593
1594        if (branch_log) {
1595                int need_dq = quote_c_style(t->name, NULL, NULL, 0);
1596                fprintf(branch_log, "tag ");
1597                if (need_dq) {
1598                        fputc('"', branch_log);
1599                        quote_c_style(t->name, NULL, branch_log, 0);
1600                        fputc('"', branch_log);
1601                } else
1602                        fprintf(branch_log, "%s", t->name);
1603                fprintf(branch_log," :%lu %s\n",from_mark,sha1_to_hex(t->sha1));
1604        }
1605}
1606
1607static const char fast_import_usage[] =
1608"git-fast-import [--objects=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file] [--branch-log=log] temp.pack";
1609
1610int main(int argc, const char **argv)
1611{
1612        const char *base_name;
1613        int i;
1614        unsigned long est_obj_cnt = 1000;
1615        char *pack_name;
1616        char *idx_name;
1617        struct stat sb;
1618
1619        setup_ident();
1620        git_config(git_default_config);
1621        page_size = getpagesize();
1622
1623        for (i = 1; i < argc; i++) {
1624                const char *a = argv[i];
1625
1626                if (*a != '-' || !strcmp(a, "--"))
1627                        break;
1628                else if (!strncmp(a, "--objects=", 10))
1629                        est_obj_cnt = strtoul(a + 10, NULL, 0);
1630                else if (!strncmp(a, "--depth=", 8))
1631                        max_depth = strtoul(a + 8, NULL, 0);
1632                else if (!strncmp(a, "--active-branches=", 18))
1633                        max_active_branches = strtoul(a + 18, NULL, 0);
1634                else if (!strncmp(a, "--export-marks=", 15))
1635                        mark_file = a + 15;
1636                else if (!strncmp(a, "--branch-log=", 13)) {
1637                        branch_log = fopen(a + 13, "w");
1638                        if (!branch_log)
1639                                die("Can't create %s: %s", a + 13, strerror(errno));
1640                }
1641                else
1642                        die("unknown option %s", a);
1643        }
1644        if ((i+1) != argc)
1645                usage(fast_import_usage);
1646        base_name = argv[i];
1647
1648        pack_name = xmalloc(strlen(base_name) + 6);
1649        sprintf(pack_name, "%s.pack", base_name);
1650        idx_name = xmalloc(strlen(base_name) + 5);
1651        sprintf(idx_name, "%s.idx", base_name);
1652
1653        pack_fd = open(pack_name, O_RDWR|O_CREAT|O_EXCL, 0666);
1654        if (pack_fd < 0)
1655                die("Can't create %s: %s", pack_name, strerror(errno));
1656
1657        init_pack_header();
1658        alloc_objects(est_obj_cnt);
1659        strbuf_init(&command_buf);
1660
1661        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
1662        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
1663        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
1664        marks = pool_calloc(1, sizeof(struct mark_set));
1665
1666        for (;;) {
1667                read_next_command();
1668                if (command_buf.eof)
1669                        break;
1670                else if (!strcmp("blob", command_buf.buf))
1671                        cmd_new_blob();
1672                else if (!strncmp("commit ", command_buf.buf, 7))
1673                        cmd_new_commit();
1674                else if (!strncmp("tag ", command_buf.buf, 4))
1675                        cmd_new_tag();
1676                else
1677                        die("Unsupported command: %s", command_buf.buf);
1678        }
1679
1680        fixup_header_footer();
1681        close(pack_fd);
1682        write_index(idx_name);
1683        dump_branches();
1684        dump_tags();
1685        dump_marks();
1686        fclose(branch_log);
1687
1688        fprintf(stderr, "%s statistics:\n", argv[0]);
1689        fprintf(stderr, "---------------------------------------------------\n");
1690        fprintf(stderr, "Alloc'd objects: %10lu (%10lu overflow  )\n", alloc_count, alloc_count - est_obj_cnt);
1691        fprintf(stderr, "Total objects:   %10lu (%10lu duplicates)\n", object_count, duplicate_count);
1692        fprintf(stderr, "      blobs  :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB]);
1693        fprintf(stderr, "      trees  :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE]);
1694        fprintf(stderr, "      commits:   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT]);
1695        fprintf(stderr, "      tags   :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG]);
1696        fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
1697        fprintf(stderr, "      marks:     %10u (%10lu unique    )\n", (1 << marks->shift) * 1024, marks_set_count);
1698        fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
1699        fprintf(stderr, "Memory total:    %10lu KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
1700        fprintf(stderr, "       pools:    %10lu KiB\n", total_allocd/1024);
1701        fprintf(stderr, "     objects:    %10lu KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
1702        fprintf(stderr, "Pack remaps:     %10lu\n", remap_count);
1703        fprintf(stderr, "---------------------------------------------------\n");
1704
1705        stat(pack_name, &sb);
1706        fprintf(stderr, "Pack size:       %10lu KiB\n", (unsigned long)(sb.st_size/1024));
1707        stat(idx_name, &sb);
1708        fprintf(stderr, "Index size:      %10lu KiB\n", (unsigned long)(sb.st_size/1024));
1709
1710        fprintf(stderr, "\n");
1711
1712        return 0;
1713}