fast-import.con commit Account for tree entry memory costs in fast-import. (8435a9c)
   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 void* pack_base;
 219static size_t pack_mlen;
 220
 221/* Table of objects we've written. */
 222static unsigned int object_entry_alloc = 1000;
 223static struct object_entry_pool *blocks;
 224static struct object_entry *object_table[1 << 16];
 225static struct mark_set *marks;
 226
 227/* Our last blob */
 228static struct last_object last_blob;
 229
 230/* Tree management */
 231static unsigned int tree_entry_alloc = 1000;
 232static void *avail_tree_entry;
 233static unsigned int avail_tree_table_sz = 100;
 234static struct avail_tree_content **avail_tree_table;
 235
 236/* Branch data */
 237static unsigned long max_active_branches = 5;
 238static unsigned long cur_active_branches;
 239static unsigned long branch_table_sz = 1039;
 240static struct branch **branch_table;
 241static struct branch *active_branches;
 242
 243/* Tag data */
 244static struct tag *first_tag;
 245static struct tag *last_tag;
 246
 247/* Input stream parsing */
 248static struct strbuf command_buf;
 249static unsigned long next_mark;
 250
 251
 252static void alloc_objects(int cnt)
 253{
 254        struct object_entry_pool *b;
 255
 256        b = xmalloc(sizeof(struct object_entry_pool)
 257                + cnt * sizeof(struct object_entry));
 258        b->next_pool = blocks;
 259        b->next_free = b->entries;
 260        b->end = b->entries + cnt;
 261        blocks = b;
 262        alloc_count += cnt;
 263}
 264
 265static struct object_entry* new_object(unsigned char *sha1)
 266{
 267        struct object_entry *e;
 268
 269        if (blocks->next_free == blocks->end)
 270                alloc_objects(object_entry_alloc);
 271
 272        e = blocks->next_free++;
 273        memcpy(e->sha1, sha1, sizeof(e->sha1));
 274        return e;
 275}
 276
 277static struct object_entry* find_object(unsigned char *sha1)
 278{
 279        unsigned int h = sha1[0] << 8 | sha1[1];
 280        struct object_entry *e;
 281        for (e = object_table[h]; e; e = e->next)
 282                if (!memcmp(sha1, e->sha1, sizeof(e->sha1)))
 283                        return e;
 284        return NULL;
 285}
 286
 287static struct object_entry* insert_object(unsigned char *sha1)
 288{
 289        unsigned int h = sha1[0] << 8 | sha1[1];
 290        struct object_entry *e = object_table[h];
 291        struct object_entry *p = NULL;
 292
 293        while (e) {
 294                if (!memcmp(sha1, e->sha1, sizeof(e->sha1)))
 295                        return e;
 296                p = e;
 297                e = e->next;
 298        }
 299
 300        e = new_object(sha1);
 301        e->next = NULL;
 302        e->offset = 0;
 303        if (p)
 304                p->next = e;
 305        else
 306                object_table[h] = e;
 307        return e;
 308}
 309
 310static unsigned int hc_str(const char *s, size_t len)
 311{
 312        unsigned int r = 0;
 313        while (len-- > 0)
 314                r = r * 31 + *s++;
 315        return r;
 316}
 317
 318static void* pool_alloc(size_t len)
 319{
 320        struct mem_pool *p;
 321        void *r;
 322
 323        for (p = mem_pool; p; p = p->next_pool)
 324                if ((p->end - p->next_free >= len))
 325                        break;
 326
 327        if (!p) {
 328                if (len >= (mem_pool_alloc/2)) {
 329                        total_allocd += len;
 330                        return xmalloc(len);
 331                }
 332                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 333                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 334                p->next_pool = mem_pool;
 335                p->next_free = p->space;
 336                p->end = p->next_free + mem_pool_alloc;
 337                mem_pool = p;
 338        }
 339
 340        r = p->next_free;
 341        /* round out to a pointer alignment */
 342        if (len & (sizeof(void*) - 1))
 343                len += sizeof(void*) - (len & (sizeof(void*) - 1));
 344        p->next_free += len;
 345        return r;
 346}
 347
 348static void* pool_calloc(size_t count, size_t size)
 349{
 350        size_t len = count * size;
 351        void *r = pool_alloc(len);
 352        memset(r, 0, len);
 353        return r;
 354}
 355
 356static char* pool_strdup(const char *s)
 357{
 358        char *r = pool_alloc(strlen(s) + 1);
 359        strcpy(r, s);
 360        return r;
 361}
 362
 363static void insert_mark(unsigned long idnum, struct object_entry *oe)
 364{
 365        struct mark_set *s = marks;
 366        while ((idnum >> s->shift) >= 1024) {
 367                s = pool_calloc(1, sizeof(struct mark_set));
 368                s->shift = marks->shift + 10;
 369                s->data.sets[0] = marks;
 370                marks = s;
 371        }
 372        while (s->shift) {
 373                unsigned long i = idnum >> s->shift;
 374                idnum -= i << s->shift;
 375                if (!s->data.sets[i]) {
 376                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 377                        s->data.sets[i]->shift = s->shift - 10;
 378                }
 379                s = s->data.sets[i];
 380        }
 381        if (!s->data.marked[idnum])
 382                marks_set_count++;
 383        s->data.marked[idnum] = oe;
 384}
 385
 386static struct object_entry* find_mark(unsigned long idnum)
 387{
 388        unsigned long orig_idnum = idnum;
 389        struct mark_set *s = marks;
 390        struct object_entry *oe = NULL;
 391        if ((idnum >> s->shift) < 1024) {
 392                while (s && s->shift) {
 393                        unsigned long i = idnum >> s->shift;
 394                        idnum -= i << s->shift;
 395                        s = s->data.sets[i];
 396                }
 397                if (s)
 398                        oe = s->data.marked[idnum];
 399        }
 400        if (!oe)
 401                die("mark :%lu not declared", orig_idnum);
 402        return oe;
 403}
 404
 405static struct atom_str* to_atom(const char *s, size_t len)
 406{
 407        unsigned int hc = hc_str(s, len) % atom_table_sz;
 408        struct atom_str *c;
 409
 410        for (c = atom_table[hc]; c; c = c->next_atom)
 411                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 412                        return c;
 413
 414        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 415        c->str_len = len;
 416        strncpy(c->str_dat, s, len);
 417        c->str_dat[len] = 0;
 418        c->next_atom = atom_table[hc];
 419        atom_table[hc] = c;
 420        atom_cnt++;
 421        return c;
 422}
 423
 424static struct branch* lookup_branch(const char *name)
 425{
 426        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 427        struct branch *b;
 428
 429        for (b = branch_table[hc]; b; b = b->table_next_branch)
 430                if (!strcmp(name, b->name))
 431                        return b;
 432        return NULL;
 433}
 434
 435static struct branch* new_branch(const char *name)
 436{
 437        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 438        struct branch* b = lookup_branch(name);
 439
 440        if (b)
 441                die("Invalid attempt to create duplicate branch: %s", name);
 442        if (check_ref_format(name))
 443                die("Branch name doesn't conform to GIT standards: %s", name);
 444
 445        b = pool_calloc(1, sizeof(struct branch));
 446        b->name = pool_strdup(name);
 447        b->table_next_branch = branch_table[hc];
 448        branch_table[hc] = b;
 449        branch_count++;
 450        return b;
 451}
 452
 453static unsigned int hc_entries(unsigned int cnt)
 454{
 455        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 456        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 457}
 458
 459static struct tree_content* new_tree_content(unsigned int cnt)
 460{
 461        struct avail_tree_content *f, *l = NULL;
 462        struct tree_content *t;
 463        unsigned int hc = hc_entries(cnt);
 464
 465        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 466                if (f->entry_capacity >= cnt)
 467                        break;
 468
 469        if (f) {
 470                if (l)
 471                        l->next_avail = f->next_avail;
 472                else
 473                        avail_tree_table[hc] = f->next_avail;
 474        } else {
 475                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 476                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 477                f->entry_capacity = cnt;
 478        }
 479
 480        t = (struct tree_content*)f;
 481        t->entry_count = 0;
 482        return t;
 483}
 484
 485static void release_tree_entry(struct tree_entry *e);
 486static void release_tree_content(struct tree_content *t)
 487{
 488        struct avail_tree_content *f = (struct avail_tree_content*)t;
 489        unsigned int hc = hc_entries(f->entry_capacity);
 490        f->next_avail = avail_tree_table[hc];
 491        avail_tree_table[hc] = f;
 492}
 493
 494static void release_tree_content_recursive(struct tree_content *t)
 495{
 496        unsigned int i;
 497        for (i = 0; i < t->entry_count; i++)
 498                release_tree_entry(t->entries[i]);
 499        release_tree_content(t);
 500}
 501
 502static struct tree_content* grow_tree_content(
 503        struct tree_content *t,
 504        int amt)
 505{
 506        struct tree_content *r = new_tree_content(t->entry_count + amt);
 507        r->entry_count = t->entry_count;
 508        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 509        release_tree_content(t);
 510        return r;
 511}
 512
 513static struct tree_entry* new_tree_entry()
 514{
 515        struct tree_entry *e;
 516
 517        if (!avail_tree_entry) {
 518                unsigned int n = tree_entry_alloc;
 519                total_allocd += n * sizeof(struct tree_entry);
 520                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 521                while (n--) {
 522                        *((void**)e) = e + 1;
 523                        e++;
 524                }
 525        }
 526
 527        e = avail_tree_entry;
 528        avail_tree_entry = *((void**)e);
 529        return e;
 530}
 531
 532static void release_tree_entry(struct tree_entry *e)
 533{
 534        if (e->tree)
 535                release_tree_content_recursive(e->tree);
 536        *((void**)e) = avail_tree_entry;
 537        avail_tree_entry = e;
 538}
 539
 540static void yread(int fd, void *buffer, size_t length)
 541{
 542        ssize_t ret = 0;
 543        while (ret < length) {
 544                ssize_t size = xread(fd, (char *) buffer + ret, length - ret);
 545                if (!size)
 546                        die("Read from descriptor %i: end of stream", fd);
 547                if (size < 0)
 548                        die("Read from descriptor %i: %s", fd, strerror(errno));
 549                ret += size;
 550        }
 551}
 552
 553static void ywrite(int fd, void *buffer, size_t length)
 554{
 555        ssize_t ret = 0;
 556        while (ret < length) {
 557                ssize_t size = xwrite(fd, (char *) buffer + ret, length - ret);
 558                if (!size)
 559                        die("Write to descriptor %i: end of file", fd);
 560                if (size < 0)
 561                        die("Write to descriptor %i: %s", fd, strerror(errno));
 562                ret += size;
 563        }
 564}
 565
 566static size_t encode_header(
 567        enum object_type type,
 568        size_t size,
 569        unsigned char *hdr)
 570{
 571        int n = 1;
 572        unsigned char c;
 573
 574        if (type < OBJ_COMMIT || type > OBJ_DELTA)
 575                die("bad type %d", type);
 576
 577        c = (type << 4) | (size & 15);
 578        size >>= 4;
 579        while (size) {
 580                *hdr++ = c | 0x80;
 581                c = size & 0x7f;
 582                size >>= 7;
 583                n++;
 584        }
 585        *hdr = c;
 586        return n;
 587}
 588
 589static int store_object(
 590        enum object_type type,
 591        void *dat,
 592        size_t datlen,
 593        struct last_object *last,
 594        unsigned char *sha1out,
 595        unsigned long mark)
 596{
 597        void *out, *delta;
 598        struct object_entry *e;
 599        unsigned char hdr[96];
 600        unsigned char sha1[20];
 601        unsigned long hdrlen, deltalen;
 602        SHA_CTX c;
 603        z_stream s;
 604
 605        hdrlen = sprintf((char*)hdr,"%s %lu",type_names[type],datlen) + 1;
 606        SHA1_Init(&c);
 607        SHA1_Update(&c, hdr, hdrlen);
 608        SHA1_Update(&c, dat, datlen);
 609        SHA1_Final(sha1, &c);
 610        if (sha1out)
 611                memcpy(sha1out, sha1, sizeof(sha1));
 612
 613        e = insert_object(sha1);
 614        if (mark)
 615                insert_mark(mark, e);
 616        if (e->offset) {
 617                duplicate_count++;
 618                duplicate_count_by_type[type]++;
 619                return 1;
 620        }
 621        e->type = type;
 622        e->offset = pack_size;
 623        object_count++;
 624        object_count_by_type[type]++;
 625
 626        if (last && last->data && last->depth < max_depth)
 627                delta = diff_delta(last->data, last->len,
 628                        dat, datlen,
 629                        &deltalen, 0);
 630        else
 631                delta = 0;
 632
 633        memset(&s, 0, sizeof(s));
 634        deflateInit(&s, zlib_compression_level);
 635
 636        if (delta) {
 637                last->depth++;
 638                s.next_in = delta;
 639                s.avail_in = deltalen;
 640                hdrlen = encode_header(OBJ_DELTA, deltalen, hdr);
 641                ywrite(pack_fd, hdr, hdrlen);
 642                ywrite(pack_fd, last->sha1, sizeof(sha1));
 643                pack_size += hdrlen + sizeof(sha1);
 644        } else {
 645                if (last)
 646                        last->depth = 0;
 647                s.next_in = dat;
 648                s.avail_in = datlen;
 649                hdrlen = encode_header(type, datlen, hdr);
 650                ywrite(pack_fd, hdr, hdrlen);
 651                pack_size += hdrlen;
 652        }
 653
 654        s.avail_out = deflateBound(&s, s.avail_in);
 655        s.next_out = out = xmalloc(s.avail_out);
 656        while (deflate(&s, Z_FINISH) == Z_OK)
 657                /* nothing */;
 658        deflateEnd(&s);
 659
 660        ywrite(pack_fd, out, s.total_out);
 661        pack_size += s.total_out;
 662
 663        free(out);
 664        if (delta)
 665                free(delta);
 666        if (last) {
 667                if (last->data)
 668                        free(last->data);
 669                last->data = dat;
 670                last->len = datlen;
 671                memcpy(last->sha1, sha1, sizeof(sha1));
 672        }
 673        return 0;
 674}
 675
 676static void* map_pack(unsigned long offset)
 677{
 678        if (offset >= pack_size)
 679                die("object offset outside of pack file");
 680        if (offset >= pack_mlen) {
 681                if (pack_base)
 682                        munmap(pack_base, pack_mlen);
 683                /* round out how much we map to 16 MB units */
 684                pack_mlen = pack_size;
 685                if (pack_mlen & ((1 << 24) - 1))
 686                        pack_mlen = ((pack_mlen >> 24) + 1) << 24;
 687                pack_base = mmap(NULL,pack_mlen,PROT_READ,MAP_SHARED,pack_fd,0);
 688                if (pack_base == MAP_FAILED)
 689                        die("Failed to map generated pack: %s", strerror(errno));
 690                remap_count++;
 691        }
 692        return (char*)pack_base + offset;
 693}
 694
 695static unsigned long unpack_object_header(unsigned long offset,
 696        enum object_type *type,
 697        unsigned long *sizep)
 698{
 699        unsigned shift;
 700        unsigned char c;
 701        unsigned long size;
 702
 703        c = *(unsigned char*)map_pack(offset++);
 704        *type = (c >> 4) & 7;
 705        size = c & 15;
 706        shift = 4;
 707        while (c & 0x80) {
 708                c = *(unsigned char*)map_pack(offset++);
 709                size += (c & 0x7f) << shift;
 710                shift += 7;
 711        }
 712        *sizep = size;
 713        return offset;
 714}
 715
 716static void *unpack_non_delta_entry(unsigned long o, unsigned long sz)
 717{
 718        z_stream stream;
 719        unsigned char *result;
 720
 721        result = xmalloc(sz + 1);
 722        result[sz] = 0;
 723
 724        memset(&stream, 0, sizeof(stream));
 725        stream.next_in = map_pack(o);
 726        stream.avail_in = pack_mlen - o;
 727        stream.next_out = result;
 728        stream.avail_out = sz;
 729
 730        inflateInit(&stream);
 731        for (;;) {
 732                int st = inflate(&stream, Z_FINISH);
 733                if (st == Z_STREAM_END)
 734                        break;
 735                if (st == Z_OK) {
 736                        o = stream.next_in - (unsigned char*)pack_base;
 737                        stream.next_in = map_pack(o);
 738                        stream.avail_in = pack_mlen - o;
 739                        continue;
 740                }
 741                die("Error from zlib during inflate.");
 742        }
 743        inflateEnd(&stream);
 744        if (stream.total_out != sz)
 745                die("Error after inflate: sizes mismatch");
 746        return result;
 747}
 748
 749static void *unpack_entry(unsigned long offset, unsigned long *sizep);
 750
 751static void *unpack_delta_entry(unsigned long offset,
 752        unsigned long delta_size,
 753        unsigned long *sizep)
 754{
 755        struct object_entry *base_oe;
 756        unsigned char *base_sha1;
 757        void *delta_data, *base, *result;
 758        unsigned long base_size, result_size;
 759
 760        base_sha1 = (unsigned char*)map_pack(offset + 20) - 20;
 761        base_oe = find_object(base_sha1);
 762        if (!base_oe)
 763                die("I'm broken; I can't find a base I know must be here.");
 764        base = unpack_entry(base_oe->offset, &base_size);
 765        delta_data = unpack_non_delta_entry(offset + 20, delta_size);
 766        result = patch_delta(base, base_size,
 767                             delta_data, delta_size,
 768                             &result_size);
 769        if (!result)
 770                die("failed to apply delta");
 771        free(delta_data);
 772        free(base);
 773        *sizep = result_size;
 774        return result;
 775}
 776
 777static void *unpack_entry(unsigned long offset, unsigned long *sizep)
 778{
 779        unsigned long size;
 780        enum object_type kind;
 781
 782        offset = unpack_object_header(offset, &kind, &size);
 783        switch (kind) {
 784        case OBJ_DELTA:
 785                return unpack_delta_entry(offset, size, sizep);
 786        case OBJ_COMMIT:
 787        case OBJ_TREE:
 788        case OBJ_BLOB:
 789        case OBJ_TAG:
 790                *sizep = size;
 791                return unpack_non_delta_entry(offset, size);
 792        default:
 793                die("I created an object I can't read!");
 794        }
 795}
 796
 797static const char *get_mode(const char *str, unsigned int *modep)
 798{
 799        unsigned char c;
 800        unsigned int mode = 0;
 801
 802        while ((c = *str++) != ' ') {
 803                if (c < '0' || c > '7')
 804                        return NULL;
 805                mode = (mode << 3) + (c - '0');
 806        }
 807        *modep = mode;
 808        return str;
 809}
 810
 811static void load_tree(struct tree_entry *root)
 812{
 813        struct object_entry *myoe;
 814        struct tree_content *t;
 815        unsigned long size;
 816        char *buf;
 817        const char *c;
 818
 819        root->tree = t = new_tree_content(8);
 820        if (!memcmp(root->sha1, null_sha1, 20))
 821                return;
 822
 823        myoe = find_object(root->sha1);
 824        if (myoe) {
 825                if (myoe->type != OBJ_TREE)
 826                        die("Not a tree: %s", sha1_to_hex(root->sha1));
 827                buf = unpack_entry(myoe->offset, &size);
 828        } else {
 829                char type[20];
 830                buf = read_sha1_file(root->sha1, type, &size);
 831                if (!buf || strcmp(type, tree_type))
 832                        die("Can't load tree %s", sha1_to_hex(root->sha1));
 833        }
 834
 835        c = buf;
 836        while (c != (buf + size)) {
 837                struct tree_entry *e = new_tree_entry();
 838
 839                if (t->entry_count == t->entry_capacity)
 840                        root->tree = t = grow_tree_content(t, 8);
 841                t->entries[t->entry_count++] = e;
 842
 843                e->tree = NULL;
 844                c = get_mode(c, &e->mode);
 845                if (!c)
 846                        die("Corrupt mode in %s", sha1_to_hex(root->sha1));
 847                e->name = to_atom(c, strlen(c));
 848                c += e->name->str_len + 1;
 849                memcpy(e->sha1, c, sizeof(e->sha1));
 850                c += 20;
 851        }
 852        free(buf);
 853}
 854
 855static int tecmp (const void *_a, const void *_b)
 856{
 857        struct tree_entry *a = *((struct tree_entry**)_a);
 858        struct tree_entry *b = *((struct tree_entry**)_b);
 859        return base_name_compare(
 860                a->name->str_dat, a->name->str_len, a->mode,
 861                b->name->str_dat, b->name->str_len, b->mode);
 862}
 863
 864static void store_tree(struct tree_entry *root)
 865{
 866        struct tree_content *t = root->tree;
 867        unsigned int i;
 868        size_t maxlen;
 869        char *buf, *c;
 870
 871        if (memcmp(root->sha1, null_sha1, 20))
 872                return;
 873
 874        maxlen = 0;
 875        for (i = 0; i < t->entry_count; i++) {
 876                maxlen += t->entries[i]->name->str_len + 34;
 877                if (t->entries[i]->tree)
 878                        store_tree(t->entries[i]);
 879        }
 880
 881        qsort(t->entries, t->entry_count, sizeof(t->entries[0]), tecmp);
 882        buf = c = xmalloc(maxlen);
 883        for (i = 0; i < t->entry_count; i++) {
 884                struct tree_entry *e = t->entries[i];
 885                c += sprintf(c, "%o", e->mode);
 886                *c++ = ' ';
 887                strcpy(c, e->name->str_dat);
 888                c += e->name->str_len + 1;
 889                memcpy(c, e->sha1, 20);
 890                c += 20;
 891        }
 892        store_object(OBJ_TREE, buf, c - buf, NULL, root->sha1, 0);
 893        free(buf);
 894}
 895
 896static int tree_content_set(
 897        struct tree_entry *root,
 898        const char *p,
 899        const unsigned char *sha1,
 900        const unsigned int mode)
 901{
 902        struct tree_content *t = root->tree;
 903        const char *slash1;
 904        unsigned int i, n;
 905        struct tree_entry *e;
 906
 907        slash1 = strchr(p, '/');
 908        if (slash1)
 909                n = slash1 - p;
 910        else
 911                n = strlen(p);
 912
 913        for (i = 0; i < t->entry_count; i++) {
 914                e = t->entries[i];
 915                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
 916                        if (!slash1) {
 917                                if (e->mode == mode && !memcmp(e->sha1, sha1, 20))
 918                                        return 0;
 919                                e->mode = mode;
 920                                memcpy(e->sha1, sha1, 20);
 921                                if (e->tree) {
 922                                        release_tree_content_recursive(e->tree);
 923                                        e->tree = NULL;
 924                                }
 925                                memcpy(root->sha1, null_sha1, 20);
 926                                return 1;
 927                        }
 928                        if (!S_ISDIR(e->mode)) {
 929                                e->tree = new_tree_content(8);
 930                                e->mode = S_IFDIR;
 931                        }
 932                        if (!e->tree)
 933                                load_tree(e);
 934                        if (tree_content_set(e, slash1 + 1, sha1, mode)) {
 935                                memcpy(root->sha1, null_sha1, 20);
 936                                return 1;
 937                        }
 938                        return 0;
 939                }
 940        }
 941
 942        if (t->entry_count == t->entry_capacity)
 943                root->tree = t = grow_tree_content(t, 8);
 944        e = new_tree_entry();
 945        e->name = to_atom(p, n);
 946        t->entries[t->entry_count++] = e;
 947        if (slash1) {
 948                e->tree = new_tree_content(8);
 949                e->mode = S_IFDIR;
 950                tree_content_set(e, slash1 + 1, sha1, mode);
 951        } else {
 952                e->tree = NULL;
 953                e->mode = mode;
 954                memcpy(e->sha1, sha1, 20);
 955        }
 956        memcpy(root->sha1, null_sha1, 20);
 957        return 1;
 958}
 959
 960static int tree_content_remove(struct tree_entry *root, const char *p)
 961{
 962        struct tree_content *t = root->tree;
 963        const char *slash1;
 964        unsigned int i, n;
 965        struct tree_entry *e;
 966
 967        slash1 = strchr(p, '/');
 968        if (slash1)
 969                n = slash1 - p;
 970        else
 971                n = strlen(p);
 972
 973        for (i = 0; i < t->entry_count; i++) {
 974                e = t->entries[i];
 975                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
 976                        if (!slash1 || !S_ISDIR(e->mode))
 977                                goto del_entry;
 978                        if (!e->tree)
 979                                load_tree(e);
 980                        if (tree_content_remove(e, slash1 + 1)) {
 981                                if (!e->tree->entry_count)
 982                                        goto del_entry;
 983                                memcpy(root->sha1, null_sha1, 20);
 984                                return 1;
 985                        }
 986                        return 0;
 987                }
 988        }
 989        return 0;
 990
 991del_entry:
 992        for (i++; i < t->entry_count; i++)
 993                t->entries[i-1] = t->entries[i];
 994        t->entry_count--;
 995        release_tree_entry(e);
 996        memcpy(root->sha1, null_sha1, 20);
 997        return 1;
 998}
 999
1000static void init_pack_header()
1001{
1002        struct pack_header hdr;
1003
1004        hdr.hdr_signature = htonl(PACK_SIGNATURE);
1005        hdr.hdr_version = htonl(2);
1006        hdr.hdr_entries = 0;
1007
1008        ywrite(pack_fd, &hdr, sizeof(hdr));
1009        pack_size = sizeof(hdr);
1010}
1011
1012static void fixup_header_footer()
1013{
1014        SHA_CTX c;
1015        char hdr[8];
1016        unsigned long cnt;
1017        char *buf;
1018        size_t n;
1019
1020        if (lseek(pack_fd, 0, SEEK_SET) != 0)
1021                die("Failed seeking to start: %s", strerror(errno));
1022
1023        SHA1_Init(&c);
1024        yread(pack_fd, hdr, 8);
1025        SHA1_Update(&c, hdr, 8);
1026
1027        cnt = htonl(object_count);
1028        SHA1_Update(&c, &cnt, 4);
1029        ywrite(pack_fd, &cnt, 4);
1030
1031        buf = xmalloc(128 * 1024);
1032        for (;;) {
1033                n = xread(pack_fd, buf, 128 * 1024);
1034                if (n <= 0)
1035                        break;
1036                SHA1_Update(&c, buf, n);
1037        }
1038        free(buf);
1039
1040        SHA1_Final(pack_sha1, &c);
1041        ywrite(pack_fd, pack_sha1, sizeof(pack_sha1));
1042}
1043
1044static int oecmp (const void *_a, const void *_b)
1045{
1046        struct object_entry *a = *((struct object_entry**)_a);
1047        struct object_entry *b = *((struct object_entry**)_b);
1048        return memcmp(a->sha1, b->sha1, sizeof(a->sha1));
1049}
1050
1051static void write_index(const char *idx_name)
1052{
1053        struct sha1file *f;
1054        struct object_entry **idx, **c, **last;
1055        struct object_entry *e;
1056        struct object_entry_pool *o;
1057        unsigned int array[256];
1058        int i;
1059
1060        /* Build the sorted table of object IDs. */
1061        idx = xmalloc(object_count * sizeof(struct object_entry*));
1062        c = idx;
1063        for (o = blocks; o; o = o->next_pool)
1064                for (e = o->entries; e != o->next_free; e++)
1065                        *c++ = e;
1066        last = idx + object_count;
1067        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
1068
1069        /* Generate the fan-out array. */
1070        c = idx;
1071        for (i = 0; i < 256; i++) {
1072                struct object_entry **next = c;;
1073                while (next < last) {
1074                        if ((*next)->sha1[0] != i)
1075                                break;
1076                        next++;
1077                }
1078                array[i] = htonl(next - idx);
1079                c = next;
1080        }
1081
1082        f = sha1create("%s", idx_name);
1083        sha1write(f, array, 256 * sizeof(int));
1084        for (c = idx; c != last; c++) {
1085                unsigned int offset = htonl((*c)->offset);
1086                sha1write(f, &offset, 4);
1087                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
1088        }
1089        sha1write(f, pack_sha1, sizeof(pack_sha1));
1090        sha1close(f, NULL, 1);
1091        free(idx);
1092}
1093
1094static void dump_branches()
1095{
1096        static const char *msg = "fast-import";
1097        unsigned int i;
1098        struct branch *b;
1099        struct ref_lock *lock;
1100
1101        for (i = 0; i < branch_table_sz; i++) {
1102                for (b = branch_table[i]; b; b = b->table_next_branch) {
1103                        lock = lock_any_ref_for_update(b->name, NULL, 0);
1104                        if (!lock || write_ref_sha1(lock, b->sha1, msg) < 0)
1105                                die("Can't write %s", b->name);
1106                }
1107        }
1108}
1109
1110static void dump_tags()
1111{
1112        static const char *msg = "fast-import";
1113        struct tag *t;
1114        struct ref_lock *lock;
1115        char path[PATH_MAX];
1116
1117        for (t = first_tag; t; t = t->next_tag) {
1118                sprintf(path, "refs/tags/%s", t->name);
1119                lock = lock_any_ref_for_update(path, NULL, 0);
1120                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1121                        die("Can't write %s", path);
1122        }
1123}
1124
1125static void read_next_command()
1126{
1127        read_line(&command_buf, stdin, '\n');
1128}
1129
1130static void cmd_mark()
1131{
1132        if (!strncmp("mark :", command_buf.buf, 6)) {
1133                next_mark = strtoul(command_buf.buf + 6, NULL, 10);
1134                read_next_command();
1135        }
1136        else
1137                next_mark = 0;
1138}
1139
1140static void* cmd_data (size_t *size)
1141{
1142        size_t n = 0;
1143        void *buffer;
1144        size_t length;
1145
1146        if (strncmp("data ", command_buf.buf, 5))
1147                die("Expected 'data n' command, found: %s", command_buf.buf);
1148
1149        length = strtoul(command_buf.buf + 5, NULL, 10);
1150        buffer = xmalloc(length);
1151
1152        while (n < length) {
1153                size_t s = fread((char*)buffer + n, 1, length - n, stdin);
1154                if (!s && feof(stdin))
1155                        die("EOF in data (%lu bytes remaining)", length - n);
1156                n += s;
1157        }
1158
1159        if (fgetc(stdin) != '\n')
1160                die("An lf did not trail the binary data as expected.");
1161
1162        *size = length;
1163        return buffer;
1164}
1165
1166static void cmd_new_blob()
1167{
1168        size_t l;
1169        void *d;
1170
1171        read_next_command();
1172        cmd_mark();
1173        d = cmd_data(&l);
1174
1175        if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
1176                free(d);
1177}
1178
1179static void unload_one_branch()
1180{
1181        while (cur_active_branches
1182                && cur_active_branches >= max_active_branches) {
1183                unsigned long min_commit = ULONG_MAX;
1184                struct branch *e, *l = NULL, *p = NULL;
1185
1186                for (e = active_branches; e; e = e->active_next_branch) {
1187                        if (e->last_commit < min_commit) {
1188                                p = l;
1189                                min_commit = e->last_commit;
1190                        }
1191                        l = e;
1192                }
1193
1194                if (p) {
1195                        e = p->active_next_branch;
1196                        p->active_next_branch = e->active_next_branch;
1197                } else {
1198                        e = active_branches;
1199                        active_branches = e->active_next_branch;
1200                }
1201                e->active_next_branch = NULL;
1202                if (e->branch_tree.tree) {
1203                        release_tree_content_recursive(e->branch_tree.tree);
1204                        e->branch_tree.tree = NULL;
1205                }
1206                cur_active_branches--;
1207        }
1208}
1209
1210static void load_branch(struct branch *b)
1211{
1212        load_tree(&b->branch_tree);
1213        b->active_next_branch = active_branches;
1214        active_branches = b;
1215        cur_active_branches++;
1216        branch_load_count++;
1217}
1218
1219static void file_change_m(struct branch *b)
1220{
1221        const char *p = command_buf.buf + 2;
1222        char *p_uq;
1223        const char *endp;
1224        struct object_entry *oe;
1225        unsigned char sha1[20];
1226        unsigned int mode;
1227        char type[20];
1228
1229        p = get_mode(p, &mode);
1230        if (!p)
1231                die("Corrupt mode: %s", command_buf.buf);
1232        switch (mode) {
1233        case S_IFREG | 0644:
1234        case S_IFREG | 0755:
1235        case S_IFLNK:
1236        case 0644:
1237        case 0755:
1238                /* ok */
1239                break;
1240        default:
1241                die("Corrupt mode: %s", command_buf.buf);
1242        }
1243
1244        if (*p == ':') {
1245                char *x;
1246                oe = find_mark(strtoul(p + 1, &x, 10));
1247                p = x;
1248        } else {
1249                if (get_sha1_hex(p, sha1))
1250                        die("Invalid SHA1: %s", command_buf.buf);
1251                oe = find_object(sha1);
1252                p += 40;
1253        }
1254        if (*p++ != ' ')
1255                die("Missing space after SHA1: %s", command_buf.buf);
1256
1257        p_uq = unquote_c_style(p, &endp);
1258        if (p_uq) {
1259                if (*endp)
1260                        die("Garbage after path in: %s", command_buf.buf);
1261                p = p_uq;
1262        }
1263
1264        if (oe) {
1265                if (oe->type != OBJ_BLOB)
1266                        die("Not a blob (actually a %s): %s",
1267                                command_buf.buf, type_names[oe->type]);
1268        } else {
1269                if (sha1_object_info(sha1, type, NULL))
1270                        die("Blob not found: %s", command_buf.buf);
1271                if (strcmp(blob_type, type))
1272                        die("Not a blob (actually a %s): %s",
1273                                command_buf.buf, type);
1274        }
1275
1276        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
1277
1278        if (p_uq)
1279                free(p_uq);
1280}
1281
1282static void file_change_d(struct branch *b)
1283{
1284        const char *p = command_buf.buf + 2;
1285        char *p_uq;
1286        const char *endp;
1287
1288        p_uq = unquote_c_style(p, &endp);
1289        if (p_uq) {
1290                if (*endp)
1291                        die("Garbage after path in: %s", command_buf.buf);
1292                p = p_uq;
1293        }
1294        tree_content_remove(&b->branch_tree, p);
1295        if (p_uq)
1296                free(p_uq);
1297}
1298
1299static void cmd_from(struct branch *b)
1300{
1301        const char *from, *endp;
1302        char *str_uq;
1303        struct branch *s;
1304
1305        if (strncmp("from ", command_buf.buf, 5))
1306                return;
1307
1308        if (b->last_commit)
1309                die("Can't reinitailize branch %s", b->name);
1310
1311        from = strchr(command_buf.buf, ' ') + 1;
1312        str_uq = unquote_c_style(from, &endp);
1313        if (str_uq) {
1314                if (*endp)
1315                        die("Garbage after string in: %s", command_buf.buf);
1316                from = str_uq;
1317        }
1318
1319        s = lookup_branch(from);
1320        if (b == s)
1321                die("Can't create a branch from itself: %s", b->name);
1322        else if (s) {
1323                memcpy(b->sha1, s->sha1, 20);
1324                memcpy(b->branch_tree.sha1, s->branch_tree.sha1, 20);
1325        } else if (*from == ':') {
1326                unsigned long idnum = strtoul(from + 1, NULL, 10);
1327                struct object_entry *oe = find_mark(idnum);
1328                unsigned long size;
1329                char *buf;
1330                if (oe->type != OBJ_COMMIT)
1331                        die("Mark :%lu not a commit", idnum);
1332                memcpy(b->sha1, oe->sha1, 20);
1333                buf = unpack_entry(oe->offset, &size);
1334                if (!buf || size < 46)
1335                        die("Not a valid commit: %s", from);
1336                if (memcmp("tree ", buf, 5)
1337                        || get_sha1_hex(buf + 5, b->branch_tree.sha1))
1338                        die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1339                free(buf);
1340        } else if (!get_sha1(from, b->sha1)) {
1341                if (!memcmp(b->sha1, null_sha1, 20))
1342                        memcpy(b->branch_tree.sha1, null_sha1, 20);
1343                else {
1344                        unsigned long size;
1345                        char *buf;
1346
1347                        buf = read_object_with_reference(b->sha1,
1348                                type_names[OBJ_COMMIT], &size, b->sha1);
1349                        if (!buf || size < 46)
1350                                die("Not a valid commit: %s", from);
1351                        if (memcmp("tree ", buf, 5)
1352                                || get_sha1_hex(buf + 5, b->branch_tree.sha1))
1353                                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1354                        free(buf);
1355                }
1356        } else
1357                die("Invalid ref name or SHA1 expression: %s", from);
1358
1359        read_next_command();
1360}
1361
1362static void cmd_new_commit()
1363{
1364        struct branch *b;
1365        void *msg;
1366        size_t msglen;
1367        char *str_uq;
1368        const char *endp;
1369        char *sp;
1370        char *author = NULL;
1371        char *committer = NULL;
1372        char *body;
1373
1374        /* Obtain the branch name from the rest of our command */
1375        sp = strchr(command_buf.buf, ' ') + 1;
1376        str_uq = unquote_c_style(sp, &endp);
1377        if (str_uq) {
1378                if (*endp)
1379                        die("Garbage after ref in: %s", command_buf.buf);
1380                sp = str_uq;
1381        }
1382        b = lookup_branch(sp);
1383        if (!b)
1384                b = new_branch(sp);
1385        if (str_uq)
1386                free(str_uq);
1387
1388        read_next_command();
1389        cmd_mark();
1390        if (!strncmp("author ", command_buf.buf, 7)) {
1391                author = strdup(command_buf.buf);
1392                read_next_command();
1393        }
1394        if (!strncmp("committer ", command_buf.buf, 10)) {
1395                committer = strdup(command_buf.buf);
1396                read_next_command();
1397        }
1398        if (!committer)
1399                die("Expected committer but didn't get one");
1400        msg = cmd_data(&msglen);
1401        read_next_command();
1402        cmd_from(b);
1403
1404        /* ensure the branch is active/loaded */
1405        if (!b->branch_tree.tree || !max_active_branches) {
1406                unload_one_branch();
1407                load_branch(b);
1408        }
1409
1410        /* file_change* */
1411        for (;;) {
1412                if (1 == command_buf.len)
1413                        break;
1414                else if (!strncmp("M ", command_buf.buf, 2))
1415                        file_change_m(b);
1416                else if (!strncmp("D ", command_buf.buf, 2))
1417                        file_change_d(b);
1418                else
1419                        die("Unsupported file_change: %s", command_buf.buf);
1420                read_next_command();
1421        }
1422
1423        /* build the tree and the commit */
1424        store_tree(&b->branch_tree);
1425        body = xmalloc(97 + msglen
1426                + (author
1427                        ? strlen(author) + strlen(committer)
1428                        : 2 * strlen(committer)));
1429        sp = body;
1430        sp += sprintf(sp, "tree %s\n", sha1_to_hex(b->branch_tree.sha1));
1431        if (memcmp(b->sha1, null_sha1, 20))
1432                sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
1433        if (author)
1434                sp += sprintf(sp, "%s\n", author);
1435        else
1436                sp += sprintf(sp, "author %s\n", committer + 10);
1437        sp += sprintf(sp, "%s\n\n", committer);
1438        memcpy(sp, msg, msglen);
1439        sp += msglen;
1440        if (author)
1441                free(author);
1442        free(committer);
1443        free(msg);
1444
1445        store_object(OBJ_COMMIT, body, sp - body, NULL, b->sha1, next_mark);
1446        free(body);
1447        b->last_commit = object_count_by_type[OBJ_COMMIT];
1448}
1449
1450static void cmd_new_tag()
1451{
1452        char *str_uq;
1453        const char *endp;
1454        char *sp;
1455        const char *from;
1456        char *tagger;
1457        struct branch *s;
1458        void *msg;
1459        size_t msglen;
1460        char *body;
1461        struct tag *t;
1462        unsigned char sha1[20];
1463
1464        /* Obtain the new tag name from the rest of our command */
1465        sp = strchr(command_buf.buf, ' ') + 1;
1466        str_uq = unquote_c_style(sp, &endp);
1467        if (str_uq) {
1468                if (*endp)
1469                        die("Garbage after tag name in: %s", command_buf.buf);
1470                sp = str_uq;
1471        }
1472        t = pool_alloc(sizeof(struct tag));
1473        t->next_tag = NULL;
1474        t->name = pool_strdup(sp);
1475        if (last_tag)
1476                last_tag->next_tag = t;
1477        else
1478                first_tag = t;
1479        last_tag = t;
1480        if (str_uq)
1481                free(str_uq);
1482        read_next_command();
1483
1484        /* from ... */
1485        if (strncmp("from ", command_buf.buf, 5))
1486                die("Expected from command, got %s", command_buf.buf);
1487
1488        from = strchr(command_buf.buf, ' ') + 1;
1489        str_uq = unquote_c_style(from, &endp);
1490        if (str_uq) {
1491                if (*endp)
1492                        die("Garbage after string in: %s", command_buf.buf);
1493                from = str_uq;
1494        }
1495
1496        s = lookup_branch(from);
1497        if (s) {
1498                memcpy(sha1, s->sha1, 20);
1499        } else if (*from == ':') {
1500                unsigned long idnum = strtoul(from + 1, NULL, 10);
1501                struct object_entry *oe = find_mark(idnum);
1502                if (oe->type != OBJ_COMMIT)
1503                        die("Mark :%lu not a commit", idnum);
1504                memcpy(sha1, oe->sha1, 20);
1505        } else if (!get_sha1(from, sha1)) {
1506                unsigned long size;
1507                char *buf;
1508
1509                buf = read_object_with_reference(sha1,
1510                        type_names[OBJ_COMMIT], &size, sha1);
1511                if (!buf || size < 46)
1512                        die("Not a valid commit: %s", from);
1513                free(buf);
1514        } else
1515                die("Invalid ref name or SHA1 expression: %s", from);
1516
1517        if (str_uq)
1518                free(str_uq);
1519        read_next_command();
1520
1521        /* tagger ... */
1522        if (strncmp("tagger ", command_buf.buf, 7))
1523                die("Expected tagger command, got %s", command_buf.buf);
1524        tagger = strdup(command_buf.buf);
1525
1526        /* tag payload/message */
1527        read_next_command();
1528        msg = cmd_data(&msglen);
1529
1530        /* build the tag object */
1531        body = xmalloc(67 + strlen(t->name) + strlen(tagger) + msglen);
1532        sp = body;
1533        sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
1534        sp += sprintf(sp, "type %s\n", type_names[OBJ_COMMIT]);
1535        sp += sprintf(sp, "tag %s\n", t->name);
1536        sp += sprintf(sp, "%s\n\n", tagger);
1537        memcpy(sp, msg, msglen);
1538        sp += msglen;
1539        free(tagger);
1540        free(msg);
1541
1542        store_object(OBJ_TAG, body, sp - body, NULL, t->sha1, 0);
1543        free(body);
1544}
1545
1546static const char fast_import_usage[] =
1547"git-fast-import [--objects=n] [--depth=n] [--active-branches=n] temp.pack";
1548
1549int main(int argc, const char **argv)
1550{
1551        const char *base_name;
1552        int i;
1553        unsigned long est_obj_cnt = 1000;
1554        char *pack_name;
1555        char *idx_name;
1556        struct stat sb;
1557
1558        setup_ident();
1559        git_config(git_default_config);
1560
1561        for (i = 1; i < argc; i++) {
1562                const char *a = argv[i];
1563
1564                if (*a != '-' || !strcmp(a, "--"))
1565                        break;
1566                else if (!strncmp(a, "--objects=", 10))
1567                        est_obj_cnt = strtoul(a + 10, NULL, 0);
1568                else if (!strncmp(a, "--depth=", 8))
1569                        max_depth = strtoul(a + 8, NULL, 0);
1570                else if (!strncmp(a, "--active-branches=", 18))
1571                        max_active_branches = strtoul(a + 18, NULL, 0);
1572                else
1573                        die("unknown option %s", a);
1574        }
1575        if ((i+1) != argc)
1576                usage(fast_import_usage);
1577        base_name = argv[i];
1578
1579        pack_name = xmalloc(strlen(base_name) + 6);
1580        sprintf(pack_name, "%s.pack", base_name);
1581        idx_name = xmalloc(strlen(base_name) + 5);
1582        sprintf(idx_name, "%s.idx", base_name);
1583
1584        pack_fd = open(pack_name, O_RDWR|O_CREAT|O_EXCL, 0666);
1585        if (pack_fd < 0)
1586                die("Can't create %s: %s", pack_name, strerror(errno));
1587
1588        init_pack_header();
1589        alloc_objects(est_obj_cnt);
1590        strbuf_init(&command_buf);
1591
1592        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
1593        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
1594        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
1595        marks = pool_calloc(1, sizeof(struct mark_set));
1596
1597        for (;;) {
1598                read_next_command();
1599                if (command_buf.eof)
1600                        break;
1601                else if (!strcmp("blob", command_buf.buf))
1602                        cmd_new_blob();
1603                else if (!strncmp("commit ", command_buf.buf, 7))
1604                        cmd_new_commit();
1605                else if (!strncmp("tag ", command_buf.buf, 4))
1606                        cmd_new_tag();
1607                else
1608                        die("Unsupported command: %s", command_buf.buf);
1609        }
1610
1611        fixup_header_footer();
1612        close(pack_fd);
1613        write_index(idx_name);
1614        dump_branches();
1615        dump_tags();
1616
1617        fprintf(stderr, "%s statistics:\n", argv[0]);
1618        fprintf(stderr, "---------------------------------------------------\n");
1619        fprintf(stderr, "Alloc'd objects: %10lu (%10lu overflow  )\n", alloc_count, alloc_count - est_obj_cnt);
1620        fprintf(stderr, "Total objects:   %10lu (%10lu duplicates)\n", object_count, duplicate_count);
1621        fprintf(stderr, "      blobs  :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB]);
1622        fprintf(stderr, "      trees  :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE]);
1623        fprintf(stderr, "      commits:   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT]);
1624        fprintf(stderr, "      tags   :   %10lu (%10lu duplicates)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG]);
1625        fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
1626        fprintf(stderr, "      marks:     %10u (%10lu unique    )\n", (1 << marks->shift) * 1024, marks_set_count);
1627        fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
1628        fprintf(stderr, "Memory total:    %10lu KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
1629        fprintf(stderr, "       pools:    %10lu KiB\n", total_allocd/1024);
1630        fprintf(stderr, "     objects:    %10lu KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
1631        fprintf(stderr, "Pack remaps:     %10lu\n", remap_count);
1632        fprintf(stderr, "---------------------------------------------------\n");
1633
1634        stat(pack_name, &sb);
1635        fprintf(stderr, "Pack size:       %10lu KiB\n", (unsigned long)(sb.st_size/1024));
1636        stat(idx_name, &sb);
1637        fprintf(stderr, "Index size:      %10lu KiB\n", (unsigned long)(sb.st_size/1024));
1638
1639        fprintf(stderr, "\n");
1640
1641        return 0;
1642}