fast-import.con commit rearrange delta search progress reporting (e334977)
   1/*
   2Format of STDIN stream:
   3
   4  stream ::= cmd*;
   5
   6  cmd ::= new_blob
   7        | new_commit
   8        | new_tag
   9        | reset_branch
  10        | checkpoint
  11        | progress
  12        ;
  13
  14  new_blob ::= 'blob' lf
  15    mark?
  16    file_content;
  17  file_content ::= data;
  18
  19  new_commit ::= 'commit' sp ref_str lf
  20    mark?
  21    ('author' sp name '<' email '>' when lf)?
  22    'committer' sp name '<' email '>' when lf
  23    commit_msg
  24    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  25    ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
  26    file_change*
  27    lf?;
  28  commit_msg ::= data;
  29
  30  file_change ::= file_clr
  31    | file_del
  32    | file_rnm
  33    | file_cpy
  34    | file_obm
  35    | file_inm;
  36  file_clr ::= 'deleteall' lf;
  37  file_del ::= 'D' sp path_str lf;
  38  file_rnm ::= 'R' sp path_str sp path_str lf;
  39  file_cpy ::= 'C' sp path_str sp path_str lf;
  40  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  41  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  42    data;
  43
  44  new_tag ::= 'tag' sp tag_str lf
  45    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
  46    'tagger' sp name '<' email '>' when lf
  47    tag_msg;
  48  tag_msg ::= data;
  49
  50  reset_branch ::= 'reset' sp ref_str lf
  51    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  52    lf?;
  53
  54  checkpoint ::= 'checkpoint' lf
  55    lf?;
  56
  57  progress ::= 'progress' sp not_lf* lf
  58    lf?;
  59
  60     # note: the first idnum in a stream should be 1 and subsequent
  61     # idnums should not have gaps between values as this will cause
  62     # the stream parser to reserve space for the gapped values.  An
  63     # idnum can be updated in the future to a new object by issuing
  64     # a new mark directive with the old idnum.
  65     #
  66  mark ::= 'mark' sp idnum lf;
  67  data ::= (delimited_data | exact_data)
  68    lf?;
  69
  70    # note: delim may be any string but must not contain lf.
  71    # data_line may contain any data but must not be exactly
  72    # delim.
  73  delimited_data ::= 'data' sp '<<' delim lf
  74    (data_line lf)*
  75    delim lf;
  76
  77     # note: declen indicates the length of binary_data in bytes.
  78     # declen does not include the lf preceeding the binary data.
  79     #
  80  exact_data ::= 'data' sp declen lf
  81    binary_data;
  82
  83     # note: quoted strings are C-style quoting supporting \c for
  84     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  85     # is the signed byte value in octal.  Note that the only
  86     # characters which must actually be escaped to protect the
  87     # stream formatting is: \, " and LF.  Otherwise these values
  88     # are UTF8.
  89     #
  90  ref_str     ::= ref;
  91  sha1exp_str ::= sha1exp;
  92  tag_str     ::= tag;
  93  path_str    ::= path    | '"' quoted(path)    '"' ;
  94  mode        ::= '100644' | '644'
  95                | '100755' | '755'
  96                | '120000'
  97                ;
  98
  99  declen ::= # unsigned 32 bit value, ascii base10 notation;
 100  bigint ::= # unsigned integer value, ascii base10 notation;
 101  binary_data ::= # file content, not interpreted;
 102
 103  when         ::= raw_when | rfc2822_when;
 104  raw_when     ::= ts sp tz;
 105  rfc2822_when ::= # Valid RFC 2822 date and time;
 106
 107  sp ::= # ASCII space character;
 108  lf ::= # ASCII newline (LF) character;
 109
 110     # note: a colon (':') must precede the numerical value assigned to
 111     # an idnum.  This is to distinguish it from a ref or tag name as
 112     # GIT does not permit ':' in ref or tag strings.
 113     #
 114  idnum   ::= ':' bigint;
 115  path    ::= # GIT style file path, e.g. "a/b/c";
 116  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 117  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 118  sha1exp ::= # Any valid GIT SHA1 expression;
 119  hexsha1 ::= # SHA1 in hexadecimal format;
 120
 121     # note: name and email are UTF8 strings, however name must not
 122     # contain '<' or lf and email must not contain any of the
 123     # following: '<', '>', lf.
 124     #
 125  name  ::= # valid GIT author/committer name;
 126  email ::= # valid GIT author/committer email;
 127  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 128  tz    ::= # GIT style timezone;
 129
 130     # note: comments may appear anywhere in the input, except
 131     # within a data command.  Any form of the data command
 132     # always escapes the related input from comment processing.
 133     #
 134     # In case it is not clear, the '#' that starts the comment
 135     # must be the first character on that the line (an lf have
 136     # preceeded it).
 137     #
 138  comment ::= '#' not_lf* lf;
 139  not_lf  ::= # Any byte that is not ASCII newline (LF);
 140*/
 141
 142#include "builtin.h"
 143#include "cache.h"
 144#include "object.h"
 145#include "blob.h"
 146#include "tree.h"
 147#include "commit.h"
 148#include "delta.h"
 149#include "pack.h"
 150#include "refs.h"
 151#include "csum-file.h"
 152#include "strbuf.h"
 153#include "quote.h"
 154
 155#define PACK_ID_BITS 16
 156#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 157
 158struct object_entry
 159{
 160        struct object_entry *next;
 161        uint32_t offset;
 162        unsigned type : TYPE_BITS;
 163        unsigned pack_id : PACK_ID_BITS;
 164        unsigned char sha1[20];
 165};
 166
 167struct object_entry_pool
 168{
 169        struct object_entry_pool *next_pool;
 170        struct object_entry *next_free;
 171        struct object_entry *end;
 172        struct object_entry entries[FLEX_ARRAY]; /* more */
 173};
 174
 175struct mark_set
 176{
 177        union {
 178                struct object_entry *marked[1024];
 179                struct mark_set *sets[1024];
 180        } data;
 181        unsigned int shift;
 182};
 183
 184struct last_object
 185{
 186        void *data;
 187        unsigned long len;
 188        uint32_t offset;
 189        unsigned int depth;
 190        unsigned no_free:1;
 191};
 192
 193struct mem_pool
 194{
 195        struct mem_pool *next_pool;
 196        char *next_free;
 197        char *end;
 198        char space[FLEX_ARRAY]; /* more */
 199};
 200
 201struct atom_str
 202{
 203        struct atom_str *next_atom;
 204        unsigned short str_len;
 205        char str_dat[FLEX_ARRAY]; /* more */
 206};
 207
 208struct tree_content;
 209struct tree_entry
 210{
 211        struct tree_content *tree;
 212        struct atom_str* name;
 213        struct tree_entry_ms
 214        {
 215                uint16_t mode;
 216                unsigned char sha1[20];
 217        } versions[2];
 218};
 219
 220struct tree_content
 221{
 222        unsigned int entry_capacity; /* must match avail_tree_content */
 223        unsigned int entry_count;
 224        unsigned int delta_depth;
 225        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 226};
 227
 228struct avail_tree_content
 229{
 230        unsigned int entry_capacity; /* must match tree_content */
 231        struct avail_tree_content *next_avail;
 232};
 233
 234struct branch
 235{
 236        struct branch *table_next_branch;
 237        struct branch *active_next_branch;
 238        const char *name;
 239        struct tree_entry branch_tree;
 240        uintmax_t last_commit;
 241        unsigned active : 1;
 242        unsigned pack_id : PACK_ID_BITS;
 243        unsigned char sha1[20];
 244};
 245
 246struct tag
 247{
 248        struct tag *next_tag;
 249        const char *name;
 250        unsigned int pack_id;
 251        unsigned char sha1[20];
 252};
 253
 254struct dbuf
 255{
 256        void *buffer;
 257        size_t capacity;
 258};
 259
 260struct hash_list
 261{
 262        struct hash_list *next;
 263        unsigned char sha1[20];
 264};
 265
 266typedef enum {
 267        WHENSPEC_RAW = 1,
 268        WHENSPEC_RFC2822,
 269        WHENSPEC_NOW,
 270} whenspec_type;
 271
 272struct recent_command
 273{
 274        struct recent_command *prev;
 275        struct recent_command *next;
 276        char *buf;
 277};
 278
 279/* Configured limits on output */
 280static unsigned long max_depth = 10;
 281static off_t max_packsize = (1LL << 32) - 1;
 282static int force_update;
 283
 284/* Stats and misc. counters */
 285static uintmax_t alloc_count;
 286static uintmax_t marks_set_count;
 287static uintmax_t object_count_by_type[1 << TYPE_BITS];
 288static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 289static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 290static unsigned long object_count;
 291static unsigned long branch_count;
 292static unsigned long branch_load_count;
 293static int failure;
 294static FILE *pack_edges;
 295
 296/* Memory pools */
 297static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 298static size_t total_allocd;
 299static struct mem_pool *mem_pool;
 300
 301/* Atom management */
 302static unsigned int atom_table_sz = 4451;
 303static unsigned int atom_cnt;
 304static struct atom_str **atom_table;
 305
 306/* The .pack file being generated */
 307static unsigned int pack_id;
 308static struct packed_git *pack_data;
 309static struct packed_git **all_packs;
 310static unsigned long pack_size;
 311
 312/* Table of objects we've written. */
 313static unsigned int object_entry_alloc = 5000;
 314static struct object_entry_pool *blocks;
 315static struct object_entry *object_table[1 << 16];
 316static struct mark_set *marks;
 317static const char* mark_file;
 318
 319/* Our last blob */
 320static struct last_object last_blob;
 321
 322/* Tree management */
 323static unsigned int tree_entry_alloc = 1000;
 324static void *avail_tree_entry;
 325static unsigned int avail_tree_table_sz = 100;
 326static struct avail_tree_content **avail_tree_table;
 327static struct dbuf old_tree;
 328static struct dbuf new_tree;
 329
 330/* Branch data */
 331static unsigned long max_active_branches = 5;
 332static unsigned long cur_active_branches;
 333static unsigned long branch_table_sz = 1039;
 334static struct branch **branch_table;
 335static struct branch *active_branches;
 336
 337/* Tag data */
 338static struct tag *first_tag;
 339static struct tag *last_tag;
 340
 341/* Input stream parsing */
 342static whenspec_type whenspec = WHENSPEC_RAW;
 343static struct strbuf command_buf;
 344static int unread_command_buf;
 345static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 346static struct recent_command *cmd_tail = &cmd_hist;
 347static struct recent_command *rc_free;
 348static unsigned int cmd_save = 100;
 349static uintmax_t next_mark;
 350static struct dbuf new_data;
 351
 352static void write_branch_report(FILE *rpt, struct branch *b)
 353{
 354        fprintf(rpt, "%s:\n", b->name);
 355
 356        fprintf(rpt, "  status      :");
 357        if (b->active)
 358                fputs(" active", rpt);
 359        if (b->branch_tree.tree)
 360                fputs(" loaded", rpt);
 361        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 362                fputs(" dirty", rpt);
 363        fputc('\n', rpt);
 364
 365        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 366        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 367        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 368        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 369
 370        fputs("  last pack   : ", rpt);
 371        if (b->pack_id < MAX_PACK_ID)
 372                fprintf(rpt, "%u", b->pack_id);
 373        fputc('\n', rpt);
 374
 375        fputc('\n', rpt);
 376}
 377
 378static void write_crash_report(const char *err)
 379{
 380        char *loc = git_path("fast_import_crash_%d", getpid());
 381        FILE *rpt = fopen(loc, "w");
 382        struct branch *b;
 383        unsigned long lu;
 384        struct recent_command *rc;
 385
 386        if (!rpt) {
 387                error("can't write crash report %s: %s", loc, strerror(errno));
 388                return;
 389        }
 390
 391        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 392
 393        fprintf(rpt, "fast-import crash report:\n");
 394        fprintf(rpt, "    fast-import process: %d\n", getpid());
 395        fprintf(rpt, "    parent process     : %d\n", getppid());
 396        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 397        fputc('\n', rpt);
 398
 399        fputs("fatal: ", rpt);
 400        fputs(err, rpt);
 401        fputc('\n', rpt);
 402
 403        fputc('\n', rpt);
 404        fputs("Most Recent Commands Before Crash\n", rpt);
 405        fputs("---------------------------------\n", rpt);
 406        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 407                if (rc->next == &cmd_hist)
 408                        fputs("* ", rpt);
 409                else
 410                        fputs("  ", rpt);
 411                fputs(rc->buf, rpt);
 412                fputc('\n', rpt);
 413        }
 414
 415        fputc('\n', rpt);
 416        fputs("Active Branch LRU\n", rpt);
 417        fputs("-----------------\n", rpt);
 418        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 419                cur_active_branches,
 420                max_active_branches);
 421        fputc('\n', rpt);
 422        fputs("  pos  clock name\n", rpt);
 423        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 424        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 425                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 426                        ++lu, b->last_commit, b->name);
 427
 428        fputc('\n', rpt);
 429        fputs("Inactive Branches\n", rpt);
 430        fputs("-----------------\n", rpt);
 431        for (lu = 0; lu < branch_table_sz; lu++) {
 432                for (b = branch_table[lu]; b; b = b->table_next_branch)
 433                        write_branch_report(rpt, b);
 434        }
 435
 436        fputc('\n', rpt);
 437        fputs("-------------------\n", rpt);
 438        fputs("END OF CRASH REPORT\n", rpt);
 439        fclose(rpt);
 440}
 441
 442static NORETURN void die_nicely(const char *err, va_list params)
 443{
 444        static int zombie;
 445        char message[2 * PATH_MAX];
 446
 447        vsnprintf(message, sizeof(message), err, params);
 448        fputs("fatal: ", stderr);
 449        fputs(message, stderr);
 450        fputc('\n', stderr);
 451
 452        if (!zombie) {
 453                zombie = 1;
 454                write_crash_report(message);
 455        }
 456        exit(128);
 457}
 458
 459static void alloc_objects(unsigned int cnt)
 460{
 461        struct object_entry_pool *b;
 462
 463        b = xmalloc(sizeof(struct object_entry_pool)
 464                + cnt * sizeof(struct object_entry));
 465        b->next_pool = blocks;
 466        b->next_free = b->entries;
 467        b->end = b->entries + cnt;
 468        blocks = b;
 469        alloc_count += cnt;
 470}
 471
 472static struct object_entry *new_object(unsigned char *sha1)
 473{
 474        struct object_entry *e;
 475
 476        if (blocks->next_free == blocks->end)
 477                alloc_objects(object_entry_alloc);
 478
 479        e = blocks->next_free++;
 480        hashcpy(e->sha1, sha1);
 481        return e;
 482}
 483
 484static struct object_entry *find_object(unsigned char *sha1)
 485{
 486        unsigned int h = sha1[0] << 8 | sha1[1];
 487        struct object_entry *e;
 488        for (e = object_table[h]; e; e = e->next)
 489                if (!hashcmp(sha1, e->sha1))
 490                        return e;
 491        return NULL;
 492}
 493
 494static struct object_entry *insert_object(unsigned char *sha1)
 495{
 496        unsigned int h = sha1[0] << 8 | sha1[1];
 497        struct object_entry *e = object_table[h];
 498        struct object_entry *p = NULL;
 499
 500        while (e) {
 501                if (!hashcmp(sha1, e->sha1))
 502                        return e;
 503                p = e;
 504                e = e->next;
 505        }
 506
 507        e = new_object(sha1);
 508        e->next = NULL;
 509        e->offset = 0;
 510        if (p)
 511                p->next = e;
 512        else
 513                object_table[h] = e;
 514        return e;
 515}
 516
 517static unsigned int hc_str(const char *s, size_t len)
 518{
 519        unsigned int r = 0;
 520        while (len-- > 0)
 521                r = r * 31 + *s++;
 522        return r;
 523}
 524
 525static void *pool_alloc(size_t len)
 526{
 527        struct mem_pool *p;
 528        void *r;
 529
 530        for (p = mem_pool; p; p = p->next_pool)
 531                if ((p->end - p->next_free >= len))
 532                        break;
 533
 534        if (!p) {
 535                if (len >= (mem_pool_alloc/2)) {
 536                        total_allocd += len;
 537                        return xmalloc(len);
 538                }
 539                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 540                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 541                p->next_pool = mem_pool;
 542                p->next_free = p->space;
 543                p->end = p->next_free + mem_pool_alloc;
 544                mem_pool = p;
 545        }
 546
 547        r = p->next_free;
 548        /* round out to a pointer alignment */
 549        if (len & (sizeof(void*) - 1))
 550                len += sizeof(void*) - (len & (sizeof(void*) - 1));
 551        p->next_free += len;
 552        return r;
 553}
 554
 555static void *pool_calloc(size_t count, size_t size)
 556{
 557        size_t len = count * size;
 558        void *r = pool_alloc(len);
 559        memset(r, 0, len);
 560        return r;
 561}
 562
 563static char *pool_strdup(const char *s)
 564{
 565        char *r = pool_alloc(strlen(s) + 1);
 566        strcpy(r, s);
 567        return r;
 568}
 569
 570static void size_dbuf(struct dbuf *b, size_t maxlen)
 571{
 572        if (b->buffer) {
 573                if (b->capacity >= maxlen)
 574                        return;
 575                free(b->buffer);
 576        }
 577        b->capacity = ((maxlen / 1024) + 1) * 1024;
 578        b->buffer = xmalloc(b->capacity);
 579}
 580
 581static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 582{
 583        struct mark_set *s = marks;
 584        while ((idnum >> s->shift) >= 1024) {
 585                s = pool_calloc(1, sizeof(struct mark_set));
 586                s->shift = marks->shift + 10;
 587                s->data.sets[0] = marks;
 588                marks = s;
 589        }
 590        while (s->shift) {
 591                uintmax_t i = idnum >> s->shift;
 592                idnum -= i << s->shift;
 593                if (!s->data.sets[i]) {
 594                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 595                        s->data.sets[i]->shift = s->shift - 10;
 596                }
 597                s = s->data.sets[i];
 598        }
 599        if (!s->data.marked[idnum])
 600                marks_set_count++;
 601        s->data.marked[idnum] = oe;
 602}
 603
 604static struct object_entry *find_mark(uintmax_t idnum)
 605{
 606        uintmax_t orig_idnum = idnum;
 607        struct mark_set *s = marks;
 608        struct object_entry *oe = NULL;
 609        if ((idnum >> s->shift) < 1024) {
 610                while (s && s->shift) {
 611                        uintmax_t i = idnum >> s->shift;
 612                        idnum -= i << s->shift;
 613                        s = s->data.sets[i];
 614                }
 615                if (s)
 616                        oe = s->data.marked[idnum];
 617        }
 618        if (!oe)
 619                die("mark :%" PRIuMAX " not declared", orig_idnum);
 620        return oe;
 621}
 622
 623static struct atom_str *to_atom(const char *s, unsigned short len)
 624{
 625        unsigned int hc = hc_str(s, len) % atom_table_sz;
 626        struct atom_str *c;
 627
 628        for (c = atom_table[hc]; c; c = c->next_atom)
 629                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 630                        return c;
 631
 632        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 633        c->str_len = len;
 634        strncpy(c->str_dat, s, len);
 635        c->str_dat[len] = 0;
 636        c->next_atom = atom_table[hc];
 637        atom_table[hc] = c;
 638        atom_cnt++;
 639        return c;
 640}
 641
 642static struct branch *lookup_branch(const char *name)
 643{
 644        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 645        struct branch *b;
 646
 647        for (b = branch_table[hc]; b; b = b->table_next_branch)
 648                if (!strcmp(name, b->name))
 649                        return b;
 650        return NULL;
 651}
 652
 653static struct branch *new_branch(const char *name)
 654{
 655        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 656        struct branch* b = lookup_branch(name);
 657
 658        if (b)
 659                die("Invalid attempt to create duplicate branch: %s", name);
 660        switch (check_ref_format(name)) {
 661        case  0: break; /* its valid */
 662        case -2: break; /* valid, but too few '/', allow anyway */
 663        default:
 664                die("Branch name doesn't conform to GIT standards: %s", name);
 665        }
 666
 667        b = pool_calloc(1, sizeof(struct branch));
 668        b->name = pool_strdup(name);
 669        b->table_next_branch = branch_table[hc];
 670        b->branch_tree.versions[0].mode = S_IFDIR;
 671        b->branch_tree.versions[1].mode = S_IFDIR;
 672        b->active = 0;
 673        b->pack_id = MAX_PACK_ID;
 674        branch_table[hc] = b;
 675        branch_count++;
 676        return b;
 677}
 678
 679static unsigned int hc_entries(unsigned int cnt)
 680{
 681        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 682        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 683}
 684
 685static struct tree_content *new_tree_content(unsigned int cnt)
 686{
 687        struct avail_tree_content *f, *l = NULL;
 688        struct tree_content *t;
 689        unsigned int hc = hc_entries(cnt);
 690
 691        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 692                if (f->entry_capacity >= cnt)
 693                        break;
 694
 695        if (f) {
 696                if (l)
 697                        l->next_avail = f->next_avail;
 698                else
 699                        avail_tree_table[hc] = f->next_avail;
 700        } else {
 701                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 702                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 703                f->entry_capacity = cnt;
 704        }
 705
 706        t = (struct tree_content*)f;
 707        t->entry_count = 0;
 708        t->delta_depth = 0;
 709        return t;
 710}
 711
 712static void release_tree_entry(struct tree_entry *e);
 713static void release_tree_content(struct tree_content *t)
 714{
 715        struct avail_tree_content *f = (struct avail_tree_content*)t;
 716        unsigned int hc = hc_entries(f->entry_capacity);
 717        f->next_avail = avail_tree_table[hc];
 718        avail_tree_table[hc] = f;
 719}
 720
 721static void release_tree_content_recursive(struct tree_content *t)
 722{
 723        unsigned int i;
 724        for (i = 0; i < t->entry_count; i++)
 725                release_tree_entry(t->entries[i]);
 726        release_tree_content(t);
 727}
 728
 729static struct tree_content *grow_tree_content(
 730        struct tree_content *t,
 731        int amt)
 732{
 733        struct tree_content *r = new_tree_content(t->entry_count + amt);
 734        r->entry_count = t->entry_count;
 735        r->delta_depth = t->delta_depth;
 736        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 737        release_tree_content(t);
 738        return r;
 739}
 740
 741static struct tree_entry *new_tree_entry(void)
 742{
 743        struct tree_entry *e;
 744
 745        if (!avail_tree_entry) {
 746                unsigned int n = tree_entry_alloc;
 747                total_allocd += n * sizeof(struct tree_entry);
 748                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 749                while (n-- > 1) {
 750                        *((void**)e) = e + 1;
 751                        e++;
 752                }
 753                *((void**)e) = NULL;
 754        }
 755
 756        e = avail_tree_entry;
 757        avail_tree_entry = *((void**)e);
 758        return e;
 759}
 760
 761static void release_tree_entry(struct tree_entry *e)
 762{
 763        if (e->tree)
 764                release_tree_content_recursive(e->tree);
 765        *((void**)e) = avail_tree_entry;
 766        avail_tree_entry = e;
 767}
 768
 769static struct tree_content *dup_tree_content(struct tree_content *s)
 770{
 771        struct tree_content *d;
 772        struct tree_entry *a, *b;
 773        unsigned int i;
 774
 775        if (!s)
 776                return NULL;
 777        d = new_tree_content(s->entry_count);
 778        for (i = 0; i < s->entry_count; i++) {
 779                a = s->entries[i];
 780                b = new_tree_entry();
 781                memcpy(b, a, sizeof(*a));
 782                if (a->tree && is_null_sha1(b->versions[1].sha1))
 783                        b->tree = dup_tree_content(a->tree);
 784                else
 785                        b->tree = NULL;
 786                d->entries[i] = b;
 787        }
 788        d->entry_count = s->entry_count;
 789        d->delta_depth = s->delta_depth;
 790
 791        return d;
 792}
 793
 794static void start_packfile(void)
 795{
 796        static char tmpfile[PATH_MAX];
 797        struct packed_git *p;
 798        struct pack_header hdr;
 799        int pack_fd;
 800
 801        snprintf(tmpfile, sizeof(tmpfile),
 802                "%s/tmp_pack_XXXXXX", get_object_directory());
 803        pack_fd = xmkstemp(tmpfile);
 804        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 805        strcpy(p->pack_name, tmpfile);
 806        p->pack_fd = pack_fd;
 807
 808        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 809        hdr.hdr_version = htonl(2);
 810        hdr.hdr_entries = 0;
 811        write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 812
 813        pack_data = p;
 814        pack_size = sizeof(hdr);
 815        object_count = 0;
 816
 817        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 818        all_packs[pack_id] = p;
 819}
 820
 821static int oecmp (const void *a_, const void *b_)
 822{
 823        struct object_entry *a = *((struct object_entry**)a_);
 824        struct object_entry *b = *((struct object_entry**)b_);
 825        return hashcmp(a->sha1, b->sha1);
 826}
 827
 828static char *create_index(void)
 829{
 830        static char tmpfile[PATH_MAX];
 831        SHA_CTX ctx;
 832        struct sha1file *f;
 833        struct object_entry **idx, **c, **last, *e;
 834        struct object_entry_pool *o;
 835        uint32_t array[256];
 836        int i, idx_fd;
 837
 838        /* Build the sorted table of object IDs. */
 839        idx = xmalloc(object_count * sizeof(struct object_entry*));
 840        c = idx;
 841        for (o = blocks; o; o = o->next_pool)
 842                for (e = o->next_free; e-- != o->entries;)
 843                        if (pack_id == e->pack_id)
 844                                *c++ = e;
 845        last = idx + object_count;
 846        if (c != last)
 847                die("internal consistency error creating the index");
 848        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 849
 850        /* Generate the fan-out array. */
 851        c = idx;
 852        for (i = 0; i < 256; i++) {
 853                struct object_entry **next = c;;
 854                while (next < last) {
 855                        if ((*next)->sha1[0] != i)
 856                                break;
 857                        next++;
 858                }
 859                array[i] = htonl(next - idx);
 860                c = next;
 861        }
 862
 863        snprintf(tmpfile, sizeof(tmpfile),
 864                "%s/tmp_idx_XXXXXX", get_object_directory());
 865        idx_fd = xmkstemp(tmpfile);
 866        f = sha1fd(idx_fd, tmpfile);
 867        sha1write(f, array, 256 * sizeof(int));
 868        SHA1_Init(&ctx);
 869        for (c = idx; c != last; c++) {
 870                uint32_t offset = htonl((*c)->offset);
 871                sha1write(f, &offset, 4);
 872                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 873                SHA1_Update(&ctx, (*c)->sha1, 20);
 874        }
 875        sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
 876        sha1close(f, NULL, 1);
 877        free(idx);
 878        SHA1_Final(pack_data->sha1, &ctx);
 879        return tmpfile;
 880}
 881
 882static char *keep_pack(char *curr_index_name)
 883{
 884        static char name[PATH_MAX];
 885        static const char *keep_msg = "fast-import";
 886        int keep_fd;
 887
 888        chmod(pack_data->pack_name, 0444);
 889        chmod(curr_index_name, 0444);
 890
 891        snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 892                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 893        keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
 894        if (keep_fd < 0)
 895                die("cannot create keep file");
 896        write(keep_fd, keep_msg, strlen(keep_msg));
 897        close(keep_fd);
 898
 899        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 900                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 901        if (move_temp_to_file(pack_data->pack_name, name))
 902                die("cannot store pack file");
 903
 904        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 905                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 906        if (move_temp_to_file(curr_index_name, name))
 907                die("cannot store index file");
 908        return name;
 909}
 910
 911static void unkeep_all_packs(void)
 912{
 913        static char name[PATH_MAX];
 914        int k;
 915
 916        for (k = 0; k < pack_id; k++) {
 917                struct packed_git *p = all_packs[k];
 918                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 919                         get_object_directory(), sha1_to_hex(p->sha1));
 920                unlink(name);
 921        }
 922}
 923
 924static void end_packfile(void)
 925{
 926        struct packed_git *old_p = pack_data, *new_p;
 927
 928        if (object_count) {
 929                char *idx_name;
 930                int i;
 931                struct branch *b;
 932                struct tag *t;
 933
 934                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 935                                    pack_data->pack_name, object_count);
 936                close(pack_data->pack_fd);
 937                idx_name = keep_pack(create_index());
 938
 939                /* Register the packfile with core git's machinary. */
 940                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 941                if (!new_p)
 942                        die("core git rejected index %s", idx_name);
 943                new_p->windows = old_p->windows;
 944                all_packs[pack_id] = new_p;
 945                install_packed_git(new_p);
 946
 947                /* Print the boundary */
 948                if (pack_edges) {
 949                        fprintf(pack_edges, "%s:", new_p->pack_name);
 950                        for (i = 0; i < branch_table_sz; i++) {
 951                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 952                                        if (b->pack_id == pack_id)
 953                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 954                                }
 955                        }
 956                        for (t = first_tag; t; t = t->next_tag) {
 957                                if (t->pack_id == pack_id)
 958                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 959                        }
 960                        fputc('\n', pack_edges);
 961                        fflush(pack_edges);
 962                }
 963
 964                pack_id++;
 965        }
 966        else
 967                unlink(old_p->pack_name);
 968        free(old_p);
 969
 970        /* We can't carry a delta across packfiles. */
 971        free(last_blob.data);
 972        last_blob.data = NULL;
 973        last_blob.len = 0;
 974        last_blob.offset = 0;
 975        last_blob.depth = 0;
 976}
 977
 978static void cycle_packfile(void)
 979{
 980        end_packfile();
 981        start_packfile();
 982}
 983
 984static size_t encode_header(
 985        enum object_type type,
 986        size_t size,
 987        unsigned char *hdr)
 988{
 989        int n = 1;
 990        unsigned char c;
 991
 992        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 993                die("bad type %d", type);
 994
 995        c = (type << 4) | (size & 15);
 996        size >>= 4;
 997        while (size) {
 998                *hdr++ = c | 0x80;
 999                c = size & 0x7f;
1000                size >>= 7;
1001                n++;
1002        }
1003        *hdr = c;
1004        return n;
1005}
1006
1007static int store_object(
1008        enum object_type type,
1009        void *dat,
1010        size_t datlen,
1011        struct last_object *last,
1012        unsigned char *sha1out,
1013        uintmax_t mark)
1014{
1015        void *out, *delta;
1016        struct object_entry *e;
1017        unsigned char hdr[96];
1018        unsigned char sha1[20];
1019        unsigned long hdrlen, deltalen;
1020        SHA_CTX c;
1021        z_stream s;
1022
1023        hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1024                (unsigned long)datlen) + 1;
1025        SHA1_Init(&c);
1026        SHA1_Update(&c, hdr, hdrlen);
1027        SHA1_Update(&c, dat, datlen);
1028        SHA1_Final(sha1, &c);
1029        if (sha1out)
1030                hashcpy(sha1out, sha1);
1031
1032        e = insert_object(sha1);
1033        if (mark)
1034                insert_mark(mark, e);
1035        if (e->offset) {
1036                duplicate_count_by_type[type]++;
1037                return 1;
1038        } else if (find_sha1_pack(sha1, packed_git)) {
1039                e->type = type;
1040                e->pack_id = MAX_PACK_ID;
1041                e->offset = 1; /* just not zero! */
1042                duplicate_count_by_type[type]++;
1043                return 1;
1044        }
1045
1046        if (last && last->data && last->depth < max_depth) {
1047                delta = diff_delta(last->data, last->len,
1048                        dat, datlen,
1049                        &deltalen, 0);
1050                if (delta && deltalen >= datlen) {
1051                        free(delta);
1052                        delta = NULL;
1053                }
1054        } else
1055                delta = NULL;
1056
1057        memset(&s, 0, sizeof(s));
1058        deflateInit(&s, zlib_compression_level);
1059        if (delta) {
1060                s.next_in = delta;
1061                s.avail_in = deltalen;
1062        } else {
1063                s.next_in = dat;
1064                s.avail_in = datlen;
1065        }
1066        s.avail_out = deflateBound(&s, s.avail_in);
1067        s.next_out = out = xmalloc(s.avail_out);
1068        while (deflate(&s, Z_FINISH) == Z_OK)
1069                /* nothing */;
1070        deflateEnd(&s);
1071
1072        /* Determine if we should auto-checkpoint. */
1073        if ((pack_size + 60 + s.total_out) > max_packsize
1074                || (pack_size + 60 + s.total_out) < pack_size) {
1075
1076                /* This new object needs to *not* have the current pack_id. */
1077                e->pack_id = pack_id + 1;
1078                cycle_packfile();
1079
1080                /* We cannot carry a delta into the new pack. */
1081                if (delta) {
1082                        free(delta);
1083                        delta = NULL;
1084
1085                        memset(&s, 0, sizeof(s));
1086                        deflateInit(&s, zlib_compression_level);
1087                        s.next_in = dat;
1088                        s.avail_in = datlen;
1089                        s.avail_out = deflateBound(&s, s.avail_in);
1090                        s.next_out = out = xrealloc(out, s.avail_out);
1091                        while (deflate(&s, Z_FINISH) == Z_OK)
1092                                /* nothing */;
1093                        deflateEnd(&s);
1094                }
1095        }
1096
1097        e->type = type;
1098        e->pack_id = pack_id;
1099        e->offset = pack_size;
1100        object_count++;
1101        object_count_by_type[type]++;
1102
1103        if (delta) {
1104                unsigned long ofs = e->offset - last->offset;
1105                unsigned pos = sizeof(hdr) - 1;
1106
1107                delta_count_by_type[type]++;
1108                last->depth++;
1109
1110                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1111                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1112                pack_size += hdrlen;
1113
1114                hdr[pos] = ofs & 127;
1115                while (ofs >>= 7)
1116                        hdr[--pos] = 128 | (--ofs & 127);
1117                write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1118                pack_size += sizeof(hdr) - pos;
1119        } else {
1120                if (last)
1121                        last->depth = 0;
1122                hdrlen = encode_header(type, datlen, hdr);
1123                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1124                pack_size += hdrlen;
1125        }
1126
1127        write_or_die(pack_data->pack_fd, out, s.total_out);
1128        pack_size += s.total_out;
1129
1130        free(out);
1131        free(delta);
1132        if (last) {
1133                if (!last->no_free)
1134                        free(last->data);
1135                last->data = dat;
1136                last->offset = e->offset;
1137                last->len = datlen;
1138        }
1139        return 0;
1140}
1141
1142static void *gfi_unpack_entry(
1143        struct object_entry *oe,
1144        unsigned long *sizep)
1145{
1146        enum object_type type;
1147        struct packed_git *p = all_packs[oe->pack_id];
1148        if (p == pack_data)
1149                p->pack_size = pack_size + 20;
1150        return unpack_entry(p, oe->offset, &type, sizep);
1151}
1152
1153static const char *get_mode(const char *str, uint16_t *modep)
1154{
1155        unsigned char c;
1156        uint16_t mode = 0;
1157
1158        while ((c = *str++) != ' ') {
1159                if (c < '0' || c > '7')
1160                        return NULL;
1161                mode = (mode << 3) + (c - '0');
1162        }
1163        *modep = mode;
1164        return str;
1165}
1166
1167static void load_tree(struct tree_entry *root)
1168{
1169        unsigned char* sha1 = root->versions[1].sha1;
1170        struct object_entry *myoe;
1171        struct tree_content *t;
1172        unsigned long size;
1173        char *buf;
1174        const char *c;
1175
1176        root->tree = t = new_tree_content(8);
1177        if (is_null_sha1(sha1))
1178                return;
1179
1180        myoe = find_object(sha1);
1181        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1182                if (myoe->type != OBJ_TREE)
1183                        die("Not a tree: %s", sha1_to_hex(sha1));
1184                t->delta_depth = 0;
1185                buf = gfi_unpack_entry(myoe, &size);
1186        } else {
1187                enum object_type type;
1188                buf = read_sha1_file(sha1, &type, &size);
1189                if (!buf || type != OBJ_TREE)
1190                        die("Can't load tree %s", sha1_to_hex(sha1));
1191        }
1192
1193        c = buf;
1194        while (c != (buf + size)) {
1195                struct tree_entry *e = new_tree_entry();
1196
1197                if (t->entry_count == t->entry_capacity)
1198                        root->tree = t = grow_tree_content(t, t->entry_count);
1199                t->entries[t->entry_count++] = e;
1200
1201                e->tree = NULL;
1202                c = get_mode(c, &e->versions[1].mode);
1203                if (!c)
1204                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1205                e->versions[0].mode = e->versions[1].mode;
1206                e->name = to_atom(c, strlen(c));
1207                c += e->name->str_len + 1;
1208                hashcpy(e->versions[0].sha1, (unsigned char*)c);
1209                hashcpy(e->versions[1].sha1, (unsigned char*)c);
1210                c += 20;
1211        }
1212        free(buf);
1213}
1214
1215static int tecmp0 (const void *_a, const void *_b)
1216{
1217        struct tree_entry *a = *((struct tree_entry**)_a);
1218        struct tree_entry *b = *((struct tree_entry**)_b);
1219        return base_name_compare(
1220                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1221                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1222}
1223
1224static int tecmp1 (const void *_a, const void *_b)
1225{
1226        struct tree_entry *a = *((struct tree_entry**)_a);
1227        struct tree_entry *b = *((struct tree_entry**)_b);
1228        return base_name_compare(
1229                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1230                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1231}
1232
1233static void mktree(struct tree_content *t,
1234        int v,
1235        unsigned long *szp,
1236        struct dbuf *b)
1237{
1238        size_t maxlen = 0;
1239        unsigned int i;
1240        char *c;
1241
1242        if (!v)
1243                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1244        else
1245                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1246
1247        for (i = 0; i < t->entry_count; i++) {
1248                if (t->entries[i]->versions[v].mode)
1249                        maxlen += t->entries[i]->name->str_len + 34;
1250        }
1251
1252        size_dbuf(b, maxlen);
1253        c = b->buffer;
1254        for (i = 0; i < t->entry_count; i++) {
1255                struct tree_entry *e = t->entries[i];
1256                if (!e->versions[v].mode)
1257                        continue;
1258                c += sprintf(c, "%o", (unsigned int)e->versions[v].mode);
1259                *c++ = ' ';
1260                strcpy(c, e->name->str_dat);
1261                c += e->name->str_len + 1;
1262                hashcpy((unsigned char*)c, e->versions[v].sha1);
1263                c += 20;
1264        }
1265        *szp = c - (char*)b->buffer;
1266}
1267
1268static void store_tree(struct tree_entry *root)
1269{
1270        struct tree_content *t = root->tree;
1271        unsigned int i, j, del;
1272        unsigned long new_len;
1273        struct last_object lo;
1274        struct object_entry *le;
1275
1276        if (!is_null_sha1(root->versions[1].sha1))
1277                return;
1278
1279        for (i = 0; i < t->entry_count; i++) {
1280                if (t->entries[i]->tree)
1281                        store_tree(t->entries[i]);
1282        }
1283
1284        le = find_object(root->versions[0].sha1);
1285        if (!S_ISDIR(root->versions[0].mode)
1286                || !le
1287                || le->pack_id != pack_id) {
1288                lo.data = NULL;
1289                lo.depth = 0;
1290                lo.no_free = 0;
1291        } else {
1292                mktree(t, 0, &lo.len, &old_tree);
1293                lo.data = old_tree.buffer;
1294                lo.offset = le->offset;
1295                lo.depth = t->delta_depth;
1296                lo.no_free = 1;
1297        }
1298
1299        mktree(t, 1, &new_len, &new_tree);
1300        store_object(OBJ_TREE, new_tree.buffer, new_len,
1301                &lo, root->versions[1].sha1, 0);
1302
1303        t->delta_depth = lo.depth;
1304        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1305                struct tree_entry *e = t->entries[i];
1306                if (e->versions[1].mode) {
1307                        e->versions[0].mode = e->versions[1].mode;
1308                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1309                        t->entries[j++] = e;
1310                } else {
1311                        release_tree_entry(e);
1312                        del++;
1313                }
1314        }
1315        t->entry_count -= del;
1316}
1317
1318static int tree_content_set(
1319        struct tree_entry *root,
1320        const char *p,
1321        const unsigned char *sha1,
1322        const uint16_t mode,
1323        struct tree_content *subtree)
1324{
1325        struct tree_content *t = root->tree;
1326        const char *slash1;
1327        unsigned int i, n;
1328        struct tree_entry *e;
1329
1330        slash1 = strchr(p, '/');
1331        if (slash1)
1332                n = slash1 - p;
1333        else
1334                n = strlen(p);
1335        if (!n)
1336                die("Empty path component found in input");
1337        if (!slash1 && !S_ISDIR(mode) && subtree)
1338                die("Non-directories cannot have subtrees");
1339
1340        for (i = 0; i < t->entry_count; i++) {
1341                e = t->entries[i];
1342                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1343                        if (!slash1) {
1344                                if (!S_ISDIR(mode)
1345                                                && e->versions[1].mode == mode
1346                                                && !hashcmp(e->versions[1].sha1, sha1))
1347                                        return 0;
1348                                e->versions[1].mode = mode;
1349                                hashcpy(e->versions[1].sha1, sha1);
1350                                if (e->tree)
1351                                        release_tree_content_recursive(e->tree);
1352                                e->tree = subtree;
1353                                hashclr(root->versions[1].sha1);
1354                                return 1;
1355                        }
1356                        if (!S_ISDIR(e->versions[1].mode)) {
1357                                e->tree = new_tree_content(8);
1358                                e->versions[1].mode = S_IFDIR;
1359                        }
1360                        if (!e->tree)
1361                                load_tree(e);
1362                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1363                                hashclr(root->versions[1].sha1);
1364                                return 1;
1365                        }
1366                        return 0;
1367                }
1368        }
1369
1370        if (t->entry_count == t->entry_capacity)
1371                root->tree = t = grow_tree_content(t, t->entry_count);
1372        e = new_tree_entry();
1373        e->name = to_atom(p, n);
1374        e->versions[0].mode = 0;
1375        hashclr(e->versions[0].sha1);
1376        t->entries[t->entry_count++] = e;
1377        if (slash1) {
1378                e->tree = new_tree_content(8);
1379                e->versions[1].mode = S_IFDIR;
1380                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1381        } else {
1382                e->tree = subtree;
1383                e->versions[1].mode = mode;
1384                hashcpy(e->versions[1].sha1, sha1);
1385        }
1386        hashclr(root->versions[1].sha1);
1387        return 1;
1388}
1389
1390static int tree_content_remove(
1391        struct tree_entry *root,
1392        const char *p,
1393        struct tree_entry *backup_leaf)
1394{
1395        struct tree_content *t = root->tree;
1396        const char *slash1;
1397        unsigned int i, n;
1398        struct tree_entry *e;
1399
1400        slash1 = strchr(p, '/');
1401        if (slash1)
1402                n = slash1 - p;
1403        else
1404                n = strlen(p);
1405
1406        for (i = 0; i < t->entry_count; i++) {
1407                e = t->entries[i];
1408                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1409                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1410                                goto del_entry;
1411                        if (!e->tree)
1412                                load_tree(e);
1413                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1414                                for (n = 0; n < e->tree->entry_count; n++) {
1415                                        if (e->tree->entries[n]->versions[1].mode) {
1416                                                hashclr(root->versions[1].sha1);
1417                                                return 1;
1418                                        }
1419                                }
1420                                backup_leaf = NULL;
1421                                goto del_entry;
1422                        }
1423                        return 0;
1424                }
1425        }
1426        return 0;
1427
1428del_entry:
1429        if (backup_leaf)
1430                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1431        else if (e->tree)
1432                release_tree_content_recursive(e->tree);
1433        e->tree = NULL;
1434        e->versions[1].mode = 0;
1435        hashclr(e->versions[1].sha1);
1436        hashclr(root->versions[1].sha1);
1437        return 1;
1438}
1439
1440static int tree_content_get(
1441        struct tree_entry *root,
1442        const char *p,
1443        struct tree_entry *leaf)
1444{
1445        struct tree_content *t = root->tree;
1446        const char *slash1;
1447        unsigned int i, n;
1448        struct tree_entry *e;
1449
1450        slash1 = strchr(p, '/');
1451        if (slash1)
1452                n = slash1 - p;
1453        else
1454                n = strlen(p);
1455
1456        for (i = 0; i < t->entry_count; i++) {
1457                e = t->entries[i];
1458                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1459                        if (!slash1) {
1460                                memcpy(leaf, e, sizeof(*leaf));
1461                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1462                                        leaf->tree = dup_tree_content(e->tree);
1463                                else
1464                                        leaf->tree = NULL;
1465                                return 1;
1466                        }
1467                        if (!S_ISDIR(e->versions[1].mode))
1468                                return 0;
1469                        if (!e->tree)
1470                                load_tree(e);
1471                        return tree_content_get(e, slash1 + 1, leaf);
1472                }
1473        }
1474        return 0;
1475}
1476
1477static int update_branch(struct branch *b)
1478{
1479        static const char *msg = "fast-import";
1480        struct ref_lock *lock;
1481        unsigned char old_sha1[20];
1482
1483        if (read_ref(b->name, old_sha1))
1484                hashclr(old_sha1);
1485        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1486        if (!lock)
1487                return error("Unable to lock %s", b->name);
1488        if (!force_update && !is_null_sha1(old_sha1)) {
1489                struct commit *old_cmit, *new_cmit;
1490
1491                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1492                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1493                if (!old_cmit || !new_cmit) {
1494                        unlock_ref(lock);
1495                        return error("Branch %s is missing commits.", b->name);
1496                }
1497
1498                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1499                        unlock_ref(lock);
1500                        warning("Not updating %s"
1501                                " (new tip %s does not contain %s)",
1502                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1503                        return -1;
1504                }
1505        }
1506        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1507                return error("Unable to update %s", b->name);
1508        return 0;
1509}
1510
1511static void dump_branches(void)
1512{
1513        unsigned int i;
1514        struct branch *b;
1515
1516        for (i = 0; i < branch_table_sz; i++) {
1517                for (b = branch_table[i]; b; b = b->table_next_branch)
1518                        failure |= update_branch(b);
1519        }
1520}
1521
1522static void dump_tags(void)
1523{
1524        static const char *msg = "fast-import";
1525        struct tag *t;
1526        struct ref_lock *lock;
1527        char ref_name[PATH_MAX];
1528
1529        for (t = first_tag; t; t = t->next_tag) {
1530                sprintf(ref_name, "tags/%s", t->name);
1531                lock = lock_ref_sha1(ref_name, NULL);
1532                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1533                        failure |= error("Unable to update %s", ref_name);
1534        }
1535}
1536
1537static void dump_marks_helper(FILE *f,
1538        uintmax_t base,
1539        struct mark_set *m)
1540{
1541        uintmax_t k;
1542        if (m->shift) {
1543                for (k = 0; k < 1024; k++) {
1544                        if (m->data.sets[k])
1545                                dump_marks_helper(f, (base + k) << m->shift,
1546                                        m->data.sets[k]);
1547                }
1548        } else {
1549                for (k = 0; k < 1024; k++) {
1550                        if (m->data.marked[k])
1551                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1552                                        sha1_to_hex(m->data.marked[k]->sha1));
1553                }
1554        }
1555}
1556
1557static void dump_marks(void)
1558{
1559        static struct lock_file mark_lock;
1560        int mark_fd;
1561        FILE *f;
1562
1563        if (!mark_file)
1564                return;
1565
1566        mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1567        if (mark_fd < 0) {
1568                failure |= error("Unable to write marks file %s: %s",
1569                        mark_file, strerror(errno));
1570                return;
1571        }
1572
1573        f = fdopen(mark_fd, "w");
1574        if (!f) {
1575                rollback_lock_file(&mark_lock);
1576                failure |= error("Unable to write marks file %s: %s",
1577                        mark_file, strerror(errno));
1578                return;
1579        }
1580
1581        dump_marks_helper(f, 0, marks);
1582        fclose(f);
1583        if (commit_lock_file(&mark_lock))
1584                failure |= error("Unable to write marks file %s: %s",
1585                        mark_file, strerror(errno));
1586}
1587
1588static void read_next_command(void)
1589{
1590        do {
1591                if (unread_command_buf) {
1592                        unread_command_buf = 0;
1593                        if (command_buf.eof)
1594                                return;
1595                } else {
1596                        struct recent_command *rc;
1597
1598                        command_buf.buf = NULL;
1599                        read_line(&command_buf, stdin, '\n');
1600                        if (command_buf.eof)
1601                                return;
1602
1603                        rc = rc_free;
1604                        if (rc)
1605                                rc_free = rc->next;
1606                        else {
1607                                rc = cmd_hist.next;
1608                                cmd_hist.next = rc->next;
1609                                cmd_hist.next->prev = &cmd_hist;
1610                                free(rc->buf);
1611                        }
1612
1613                        rc->buf = command_buf.buf;
1614                        rc->prev = cmd_tail;
1615                        rc->next = cmd_hist.prev;
1616                        rc->prev->next = rc;
1617                        cmd_tail = rc;
1618                }
1619        } while (command_buf.buf[0] == '#');
1620}
1621
1622static void skip_optional_lf(void)
1623{
1624        int term_char = fgetc(stdin);
1625        if (term_char != '\n' && term_char != EOF)
1626                ungetc(term_char, stdin);
1627}
1628
1629static void cmd_mark(void)
1630{
1631        if (!prefixcmp(command_buf.buf, "mark :")) {
1632                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1633                read_next_command();
1634        }
1635        else
1636                next_mark = 0;
1637}
1638
1639static void *cmd_data (size_t *size)
1640{
1641        size_t length;
1642        char *buffer;
1643
1644        if (prefixcmp(command_buf.buf, "data "))
1645                die("Expected 'data n' command, found: %s", command_buf.buf);
1646
1647        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1648                char *term = xstrdup(command_buf.buf + 5 + 2);
1649                size_t sz = 8192, term_len = command_buf.len - 5 - 2;
1650                length = 0;
1651                buffer = xmalloc(sz);
1652                command_buf.buf = NULL;
1653                for (;;) {
1654                        read_line(&command_buf, stdin, '\n');
1655                        if (command_buf.eof)
1656                                die("EOF in data (terminator '%s' not found)", term);
1657                        if (term_len == command_buf.len
1658                                && !strcmp(term, command_buf.buf))
1659                                break;
1660                        ALLOC_GROW(buffer, length + command_buf.len, sz);
1661                        memcpy(buffer + length,
1662                                command_buf.buf,
1663                                command_buf.len - 1);
1664                        length += command_buf.len - 1;
1665                        buffer[length++] = '\n';
1666                }
1667                free(term);
1668        }
1669        else {
1670                size_t n = 0;
1671                length = strtoul(command_buf.buf + 5, NULL, 10);
1672                buffer = xmalloc(length);
1673                while (n < length) {
1674                        size_t s = fread(buffer + n, 1, length - n, stdin);
1675                        if (!s && feof(stdin))
1676                                die("EOF in data (%lu bytes remaining)",
1677                                        (unsigned long)(length - n));
1678                        n += s;
1679                }
1680        }
1681
1682        skip_optional_lf();
1683        *size = length;
1684        return buffer;
1685}
1686
1687static int validate_raw_date(const char *src, char *result, int maxlen)
1688{
1689        const char *orig_src = src;
1690        char *endp, sign;
1691
1692        strtoul(src, &endp, 10);
1693        if (endp == src || *endp != ' ')
1694                return -1;
1695
1696        src = endp + 1;
1697        if (*src != '-' && *src != '+')
1698                return -1;
1699        sign = *src;
1700
1701        strtoul(src + 1, &endp, 10);
1702        if (endp == src || *endp || (endp - orig_src) >= maxlen)
1703                return -1;
1704
1705        strcpy(result, orig_src);
1706        return 0;
1707}
1708
1709static char *parse_ident(const char *buf)
1710{
1711        const char *gt;
1712        size_t name_len;
1713        char *ident;
1714
1715        gt = strrchr(buf, '>');
1716        if (!gt)
1717                die("Missing > in ident string: %s", buf);
1718        gt++;
1719        if (*gt != ' ')
1720                die("Missing space after > in ident string: %s", buf);
1721        gt++;
1722        name_len = gt - buf;
1723        ident = xmalloc(name_len + 24);
1724        strncpy(ident, buf, name_len);
1725
1726        switch (whenspec) {
1727        case WHENSPEC_RAW:
1728                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1729                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1730                break;
1731        case WHENSPEC_RFC2822:
1732                if (parse_date(gt, ident + name_len, 24) < 0)
1733                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1734                break;
1735        case WHENSPEC_NOW:
1736                if (strcmp("now", gt))
1737                        die("Date in ident must be 'now': %s", buf);
1738                datestamp(ident + name_len, 24);
1739                break;
1740        }
1741
1742        return ident;
1743}
1744
1745static void cmd_new_blob(void)
1746{
1747        size_t l;
1748        void *d;
1749
1750        read_next_command();
1751        cmd_mark();
1752        d = cmd_data(&l);
1753
1754        if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
1755                free(d);
1756}
1757
1758static void unload_one_branch(void)
1759{
1760        while (cur_active_branches
1761                && cur_active_branches >= max_active_branches) {
1762                uintmax_t min_commit = ULONG_MAX;
1763                struct branch *e, *l = NULL, *p = NULL;
1764
1765                for (e = active_branches; e; e = e->active_next_branch) {
1766                        if (e->last_commit < min_commit) {
1767                                p = l;
1768                                min_commit = e->last_commit;
1769                        }
1770                        l = e;
1771                }
1772
1773                if (p) {
1774                        e = p->active_next_branch;
1775                        p->active_next_branch = e->active_next_branch;
1776                } else {
1777                        e = active_branches;
1778                        active_branches = e->active_next_branch;
1779                }
1780                e->active = 0;
1781                e->active_next_branch = NULL;
1782                if (e->branch_tree.tree) {
1783                        release_tree_content_recursive(e->branch_tree.tree);
1784                        e->branch_tree.tree = NULL;
1785                }
1786                cur_active_branches--;
1787        }
1788}
1789
1790static void load_branch(struct branch *b)
1791{
1792        load_tree(&b->branch_tree);
1793        if (!b->active) {
1794                b->active = 1;
1795                b->active_next_branch = active_branches;
1796                active_branches = b;
1797                cur_active_branches++;
1798                branch_load_count++;
1799        }
1800}
1801
1802static void file_change_m(struct branch *b)
1803{
1804        const char *p = command_buf.buf + 2;
1805        char *p_uq;
1806        const char *endp;
1807        struct object_entry *oe = oe;
1808        unsigned char sha1[20];
1809        uint16_t mode, inline_data = 0;
1810
1811        p = get_mode(p, &mode);
1812        if (!p)
1813                die("Corrupt mode: %s", command_buf.buf);
1814        switch (mode) {
1815        case S_IFREG | 0644:
1816        case S_IFREG | 0755:
1817        case S_IFLNK:
1818        case 0644:
1819        case 0755:
1820                /* ok */
1821                break;
1822        default:
1823                die("Corrupt mode: %s", command_buf.buf);
1824        }
1825
1826        if (*p == ':') {
1827                char *x;
1828                oe = find_mark(strtoumax(p + 1, &x, 10));
1829                hashcpy(sha1, oe->sha1);
1830                p = x;
1831        } else if (!prefixcmp(p, "inline")) {
1832                inline_data = 1;
1833                p += 6;
1834        } else {
1835                if (get_sha1_hex(p, sha1))
1836                        die("Invalid SHA1: %s", command_buf.buf);
1837                oe = find_object(sha1);
1838                p += 40;
1839        }
1840        if (*p++ != ' ')
1841                die("Missing space after SHA1: %s", command_buf.buf);
1842
1843        p_uq = unquote_c_style(p, &endp);
1844        if (p_uq) {
1845                if (*endp)
1846                        die("Garbage after path in: %s", command_buf.buf);
1847                p = p_uq;
1848        }
1849
1850        if (inline_data) {
1851                size_t l;
1852                void *d;
1853                if (!p_uq)
1854                        p = p_uq = xstrdup(p);
1855                read_next_command();
1856                d = cmd_data(&l);
1857                if (store_object(OBJ_BLOB, d, l, &last_blob, sha1, 0))
1858                        free(d);
1859        } else if (oe) {
1860                if (oe->type != OBJ_BLOB)
1861                        die("Not a blob (actually a %s): %s",
1862                                command_buf.buf, typename(oe->type));
1863        } else {
1864                enum object_type type = sha1_object_info(sha1, NULL);
1865                if (type < 0)
1866                        die("Blob not found: %s", command_buf.buf);
1867                if (type != OBJ_BLOB)
1868                        die("Not a blob (actually a %s): %s",
1869                            typename(type), command_buf.buf);
1870        }
1871
1872        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1873        free(p_uq);
1874}
1875
1876static void file_change_d(struct branch *b)
1877{
1878        const char *p = command_buf.buf + 2;
1879        char *p_uq;
1880        const char *endp;
1881
1882        p_uq = unquote_c_style(p, &endp);
1883        if (p_uq) {
1884                if (*endp)
1885                        die("Garbage after path in: %s", command_buf.buf);
1886                p = p_uq;
1887        }
1888        tree_content_remove(&b->branch_tree, p, NULL);
1889        free(p_uq);
1890}
1891
1892static void file_change_cr(struct branch *b, int rename)
1893{
1894        const char *s, *d;
1895        char *s_uq, *d_uq;
1896        const char *endp;
1897        struct tree_entry leaf;
1898
1899        s = command_buf.buf + 2;
1900        s_uq = unquote_c_style(s, &endp);
1901        if (s_uq) {
1902                if (*endp != ' ')
1903                        die("Missing space after source: %s", command_buf.buf);
1904        }
1905        else {
1906                endp = strchr(s, ' ');
1907                if (!endp)
1908                        die("Missing space after source: %s", command_buf.buf);
1909                s_uq = xmalloc(endp - s + 1);
1910                memcpy(s_uq, s, endp - s);
1911                s_uq[endp - s] = 0;
1912        }
1913        s = s_uq;
1914
1915        endp++;
1916        if (!*endp)
1917                die("Missing dest: %s", command_buf.buf);
1918
1919        d = endp;
1920        d_uq = unquote_c_style(d, &endp);
1921        if (d_uq) {
1922                if (*endp)
1923                        die("Garbage after dest in: %s", command_buf.buf);
1924                d = d_uq;
1925        }
1926
1927        memset(&leaf, 0, sizeof(leaf));
1928        if (rename)
1929                tree_content_remove(&b->branch_tree, s, &leaf);
1930        else
1931                tree_content_get(&b->branch_tree, s, &leaf);
1932        if (!leaf.versions[1].mode)
1933                die("Path %s not in branch", s);
1934        tree_content_set(&b->branch_tree, d,
1935                leaf.versions[1].sha1,
1936                leaf.versions[1].mode,
1937                leaf.tree);
1938
1939        free(s_uq);
1940        free(d_uq);
1941}
1942
1943static void file_change_deleteall(struct branch *b)
1944{
1945        release_tree_content_recursive(b->branch_tree.tree);
1946        hashclr(b->branch_tree.versions[0].sha1);
1947        hashclr(b->branch_tree.versions[1].sha1);
1948        load_tree(&b->branch_tree);
1949}
1950
1951static void cmd_from_commit(struct branch *b, char *buf, unsigned long size)
1952{
1953        if (!buf || size < 46)
1954                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
1955        if (memcmp("tree ", buf, 5)
1956                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1957                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1958        hashcpy(b->branch_tree.versions[0].sha1,
1959                b->branch_tree.versions[1].sha1);
1960}
1961
1962static void cmd_from_existing(struct branch *b)
1963{
1964        if (is_null_sha1(b->sha1)) {
1965                hashclr(b->branch_tree.versions[0].sha1);
1966                hashclr(b->branch_tree.versions[1].sha1);
1967        } else {
1968                unsigned long size;
1969                char *buf;
1970
1971                buf = read_object_with_reference(b->sha1,
1972                        commit_type, &size, b->sha1);
1973                cmd_from_commit(b, buf, size);
1974                free(buf);
1975        }
1976}
1977
1978static int cmd_from(struct branch *b)
1979{
1980        const char *from;
1981        struct branch *s;
1982
1983        if (prefixcmp(command_buf.buf, "from "))
1984                return 0;
1985
1986        if (b->branch_tree.tree) {
1987                release_tree_content_recursive(b->branch_tree.tree);
1988                b->branch_tree.tree = NULL;
1989        }
1990
1991        from = strchr(command_buf.buf, ' ') + 1;
1992        s = lookup_branch(from);
1993        if (b == s)
1994                die("Can't create a branch from itself: %s", b->name);
1995        else if (s) {
1996                unsigned char *t = s->branch_tree.versions[1].sha1;
1997                hashcpy(b->sha1, s->sha1);
1998                hashcpy(b->branch_tree.versions[0].sha1, t);
1999                hashcpy(b->branch_tree.versions[1].sha1, t);
2000        } else if (*from == ':') {
2001                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2002                struct object_entry *oe = find_mark(idnum);
2003                if (oe->type != OBJ_COMMIT)
2004                        die("Mark :%" PRIuMAX " not a commit", idnum);
2005                hashcpy(b->sha1, oe->sha1);
2006                if (oe->pack_id != MAX_PACK_ID) {
2007                        unsigned long size;
2008                        char *buf = gfi_unpack_entry(oe, &size);
2009                        cmd_from_commit(b, buf, size);
2010                        free(buf);
2011                } else
2012                        cmd_from_existing(b);
2013        } else if (!get_sha1(from, b->sha1))
2014                cmd_from_existing(b);
2015        else
2016                die("Invalid ref name or SHA1 expression: %s", from);
2017
2018        read_next_command();
2019        return 1;
2020}
2021
2022static struct hash_list *cmd_merge(unsigned int *count)
2023{
2024        struct hash_list *list = NULL, *n, *e = e;
2025        const char *from;
2026        struct branch *s;
2027
2028        *count = 0;
2029        while (!prefixcmp(command_buf.buf, "merge ")) {
2030                from = strchr(command_buf.buf, ' ') + 1;
2031                n = xmalloc(sizeof(*n));
2032                s = lookup_branch(from);
2033                if (s)
2034                        hashcpy(n->sha1, s->sha1);
2035                else if (*from == ':') {
2036                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2037                        struct object_entry *oe = find_mark(idnum);
2038                        if (oe->type != OBJ_COMMIT)
2039                                die("Mark :%" PRIuMAX " not a commit", idnum);
2040                        hashcpy(n->sha1, oe->sha1);
2041                } else if (!get_sha1(from, n->sha1)) {
2042                        unsigned long size;
2043                        char *buf = read_object_with_reference(n->sha1,
2044                                commit_type, &size, n->sha1);
2045                        if (!buf || size < 46)
2046                                die("Not a valid commit: %s", from);
2047                        free(buf);
2048                } else
2049                        die("Invalid ref name or SHA1 expression: %s", from);
2050
2051                n->next = NULL;
2052                if (list)
2053                        e->next = n;
2054                else
2055                        list = n;
2056                e = n;
2057                (*count)++;
2058                read_next_command();
2059        }
2060        return list;
2061}
2062
2063static void cmd_new_commit(void)
2064{
2065        struct branch *b;
2066        void *msg;
2067        size_t msglen;
2068        char *sp;
2069        char *author = NULL;
2070        char *committer = NULL;
2071        struct hash_list *merge_list = NULL;
2072        unsigned int merge_count;
2073
2074        /* Obtain the branch name from the rest of our command */
2075        sp = strchr(command_buf.buf, ' ') + 1;
2076        b = lookup_branch(sp);
2077        if (!b)
2078                b = new_branch(sp);
2079
2080        read_next_command();
2081        cmd_mark();
2082        if (!prefixcmp(command_buf.buf, "author ")) {
2083                author = parse_ident(command_buf.buf + 7);
2084                read_next_command();
2085        }
2086        if (!prefixcmp(command_buf.buf, "committer ")) {
2087                committer = parse_ident(command_buf.buf + 10);
2088                read_next_command();
2089        }
2090        if (!committer)
2091                die("Expected committer but didn't get one");
2092        msg = cmd_data(&msglen);
2093        read_next_command();
2094        cmd_from(b);
2095        merge_list = cmd_merge(&merge_count);
2096
2097        /* ensure the branch is active/loaded */
2098        if (!b->branch_tree.tree || !max_active_branches) {
2099                unload_one_branch();
2100                load_branch(b);
2101        }
2102
2103        /* file_change* */
2104        while (!command_buf.eof && command_buf.len > 1) {
2105                if (!prefixcmp(command_buf.buf, "M "))
2106                        file_change_m(b);
2107                else if (!prefixcmp(command_buf.buf, "D "))
2108                        file_change_d(b);
2109                else if (!prefixcmp(command_buf.buf, "R "))
2110                        file_change_cr(b, 1);
2111                else if (!prefixcmp(command_buf.buf, "C "))
2112                        file_change_cr(b, 0);
2113                else if (!strcmp("deleteall", command_buf.buf))
2114                        file_change_deleteall(b);
2115                else {
2116                        unread_command_buf = 1;
2117                        break;
2118                }
2119                read_next_command();
2120        }
2121
2122        /* build the tree and the commit */
2123        store_tree(&b->branch_tree);
2124        hashcpy(b->branch_tree.versions[0].sha1,
2125                b->branch_tree.versions[1].sha1);
2126        size_dbuf(&new_data, 114 + msglen
2127                + merge_count * 49
2128                + (author
2129                        ? strlen(author) + strlen(committer)
2130                        : 2 * strlen(committer)));
2131        sp = new_data.buffer;
2132        sp += sprintf(sp, "tree %s\n",
2133                sha1_to_hex(b->branch_tree.versions[1].sha1));
2134        if (!is_null_sha1(b->sha1))
2135                sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
2136        while (merge_list) {
2137                struct hash_list *next = merge_list->next;
2138                sp += sprintf(sp, "parent %s\n", sha1_to_hex(merge_list->sha1));
2139                free(merge_list);
2140                merge_list = next;
2141        }
2142        sp += sprintf(sp, "author %s\n", author ? author : committer);
2143        sp += sprintf(sp, "committer %s\n", committer);
2144        *sp++ = '\n';
2145        memcpy(sp, msg, msglen);
2146        sp += msglen;
2147        free(author);
2148        free(committer);
2149        free(msg);
2150
2151        if (!store_object(OBJ_COMMIT,
2152                new_data.buffer, sp - (char*)new_data.buffer,
2153                NULL, b->sha1, next_mark))
2154                b->pack_id = pack_id;
2155        b->last_commit = object_count_by_type[OBJ_COMMIT];
2156}
2157
2158static void cmd_new_tag(void)
2159{
2160        char *sp;
2161        const char *from;
2162        char *tagger;
2163        struct branch *s;
2164        void *msg;
2165        size_t msglen;
2166        struct tag *t;
2167        uintmax_t from_mark = 0;
2168        unsigned char sha1[20];
2169
2170        /* Obtain the new tag name from the rest of our command */
2171        sp = strchr(command_buf.buf, ' ') + 1;
2172        t = pool_alloc(sizeof(struct tag));
2173        t->next_tag = NULL;
2174        t->name = pool_strdup(sp);
2175        if (last_tag)
2176                last_tag->next_tag = t;
2177        else
2178                first_tag = t;
2179        last_tag = t;
2180        read_next_command();
2181
2182        /* from ... */
2183        if (prefixcmp(command_buf.buf, "from "))
2184                die("Expected from command, got %s", command_buf.buf);
2185        from = strchr(command_buf.buf, ' ') + 1;
2186        s = lookup_branch(from);
2187        if (s) {
2188                hashcpy(sha1, s->sha1);
2189        } else if (*from == ':') {
2190                struct object_entry *oe;
2191                from_mark = strtoumax(from + 1, NULL, 10);
2192                oe = find_mark(from_mark);
2193                if (oe->type != OBJ_COMMIT)
2194                        die("Mark :%" PRIuMAX " not a commit", from_mark);
2195                hashcpy(sha1, oe->sha1);
2196        } else if (!get_sha1(from, sha1)) {
2197                unsigned long size;
2198                char *buf;
2199
2200                buf = read_object_with_reference(sha1,
2201                        commit_type, &size, sha1);
2202                if (!buf || size < 46)
2203                        die("Not a valid commit: %s", from);
2204                free(buf);
2205        } else
2206                die("Invalid ref name or SHA1 expression: %s", from);
2207        read_next_command();
2208
2209        /* tagger ... */
2210        if (prefixcmp(command_buf.buf, "tagger "))
2211                die("Expected tagger command, got %s", command_buf.buf);
2212        tagger = parse_ident(command_buf.buf + 7);
2213
2214        /* tag payload/message */
2215        read_next_command();
2216        msg = cmd_data(&msglen);
2217
2218        /* build the tag object */
2219        size_dbuf(&new_data, 67+strlen(t->name)+strlen(tagger)+msglen);
2220        sp = new_data.buffer;
2221        sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
2222        sp += sprintf(sp, "type %s\n", commit_type);
2223        sp += sprintf(sp, "tag %s\n", t->name);
2224        sp += sprintf(sp, "tagger %s\n", tagger);
2225        *sp++ = '\n';
2226        memcpy(sp, msg, msglen);
2227        sp += msglen;
2228        free(tagger);
2229        free(msg);
2230
2231        if (store_object(OBJ_TAG, new_data.buffer,
2232                sp - (char*)new_data.buffer,
2233                NULL, t->sha1, 0))
2234                t->pack_id = MAX_PACK_ID;
2235        else
2236                t->pack_id = pack_id;
2237}
2238
2239static void cmd_reset_branch(void)
2240{
2241        struct branch *b;
2242        char *sp;
2243
2244        /* Obtain the branch name from the rest of our command */
2245        sp = strchr(command_buf.buf, ' ') + 1;
2246        b = lookup_branch(sp);
2247        if (b) {
2248                hashclr(b->sha1);
2249                hashclr(b->branch_tree.versions[0].sha1);
2250                hashclr(b->branch_tree.versions[1].sha1);
2251                if (b->branch_tree.tree) {
2252                        release_tree_content_recursive(b->branch_tree.tree);
2253                        b->branch_tree.tree = NULL;
2254                }
2255        }
2256        else
2257                b = new_branch(sp);
2258        read_next_command();
2259        if (!cmd_from(b) && command_buf.len > 1)
2260                unread_command_buf = 1;
2261}
2262
2263static void cmd_checkpoint(void)
2264{
2265        if (object_count) {
2266                cycle_packfile();
2267                dump_branches();
2268                dump_tags();
2269                dump_marks();
2270        }
2271        skip_optional_lf();
2272}
2273
2274static void cmd_progress(void)
2275{
2276        fwrite(command_buf.buf, 1, command_buf.len - 1, stdout);
2277        fputc('\n', stdout);
2278        fflush(stdout);
2279        skip_optional_lf();
2280}
2281
2282static void import_marks(const char *input_file)
2283{
2284        char line[512];
2285        FILE *f = fopen(input_file, "r");
2286        if (!f)
2287                die("cannot read %s: %s", input_file, strerror(errno));
2288        while (fgets(line, sizeof(line), f)) {
2289                uintmax_t mark;
2290                char *end;
2291                unsigned char sha1[20];
2292                struct object_entry *e;
2293
2294                end = strchr(line, '\n');
2295                if (line[0] != ':' || !end)
2296                        die("corrupt mark line: %s", line);
2297                *end = 0;
2298                mark = strtoumax(line + 1, &end, 10);
2299                if (!mark || end == line + 1
2300                        || *end != ' ' || get_sha1(end + 1, sha1))
2301                        die("corrupt mark line: %s", line);
2302                e = find_object(sha1);
2303                if (!e) {
2304                        enum object_type type = sha1_object_info(sha1, NULL);
2305                        if (type < 0)
2306                                die("object not found: %s", sha1_to_hex(sha1));
2307                        e = insert_object(sha1);
2308                        e->type = type;
2309                        e->pack_id = MAX_PACK_ID;
2310                        e->offset = 1; /* just not zero! */
2311                }
2312                insert_mark(mark, e);
2313        }
2314        fclose(f);
2315}
2316
2317static const char fast_import_usage[] =
2318"git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2319
2320int main(int argc, const char **argv)
2321{
2322        unsigned int i, show_stats = 1;
2323
2324        git_config(git_default_config);
2325        alloc_objects(object_entry_alloc);
2326        strbuf_init(&command_buf);
2327        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2328        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2329        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2330        marks = pool_calloc(1, sizeof(struct mark_set));
2331
2332        for (i = 1; i < argc; i++) {
2333                const char *a = argv[i];
2334
2335                if (*a != '-' || !strcmp(a, "--"))
2336                        break;
2337                else if (!prefixcmp(a, "--date-format=")) {
2338                        const char *fmt = a + 14;
2339                        if (!strcmp(fmt, "raw"))
2340                                whenspec = WHENSPEC_RAW;
2341                        else if (!strcmp(fmt, "rfc2822"))
2342                                whenspec = WHENSPEC_RFC2822;
2343                        else if (!strcmp(fmt, "now"))
2344                                whenspec = WHENSPEC_NOW;
2345                        else
2346                                die("unknown --date-format argument %s", fmt);
2347                }
2348                else if (!prefixcmp(a, "--max-pack-size="))
2349                        max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2350                else if (!prefixcmp(a, "--depth="))
2351                        max_depth = strtoul(a + 8, NULL, 0);
2352                else if (!prefixcmp(a, "--active-branches="))
2353                        max_active_branches = strtoul(a + 18, NULL, 0);
2354                else if (!prefixcmp(a, "--import-marks="))
2355                        import_marks(a + 15);
2356                else if (!prefixcmp(a, "--export-marks="))
2357                        mark_file = a + 15;
2358                else if (!prefixcmp(a, "--export-pack-edges=")) {
2359                        if (pack_edges)
2360                                fclose(pack_edges);
2361                        pack_edges = fopen(a + 20, "a");
2362                        if (!pack_edges)
2363                                die("Cannot open %s: %s", a + 20, strerror(errno));
2364                } else if (!strcmp(a, "--force"))
2365                        force_update = 1;
2366                else if (!strcmp(a, "--quiet"))
2367                        show_stats = 0;
2368                else if (!strcmp(a, "--stats"))
2369                        show_stats = 1;
2370                else
2371                        die("unknown option %s", a);
2372        }
2373        if (i != argc)
2374                usage(fast_import_usage);
2375
2376        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2377        for (i = 0; i < (cmd_save - 1); i++)
2378                rc_free[i].next = &rc_free[i + 1];
2379        rc_free[cmd_save - 1].next = NULL;
2380
2381        prepare_packed_git();
2382        start_packfile();
2383        set_die_routine(die_nicely);
2384        for (;;) {
2385                read_next_command();
2386                if (command_buf.eof)
2387                        break;
2388                else if (!strcmp("blob", command_buf.buf))
2389                        cmd_new_blob();
2390                else if (!prefixcmp(command_buf.buf, "commit "))
2391                        cmd_new_commit();
2392                else if (!prefixcmp(command_buf.buf, "tag "))
2393                        cmd_new_tag();
2394                else if (!prefixcmp(command_buf.buf, "reset "))
2395                        cmd_reset_branch();
2396                else if (!strcmp("checkpoint", command_buf.buf))
2397                        cmd_checkpoint();
2398                else if (!prefixcmp(command_buf.buf, "progress "))
2399                        cmd_progress();
2400                else
2401                        die("Unsupported command: %s", command_buf.buf);
2402        }
2403        end_packfile();
2404
2405        dump_branches();
2406        dump_tags();
2407        unkeep_all_packs();
2408        dump_marks();
2409
2410        if (pack_edges)
2411                fclose(pack_edges);
2412
2413        if (show_stats) {
2414                uintmax_t total_count = 0, duplicate_count = 0;
2415                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2416                        total_count += object_count_by_type[i];
2417                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2418                        duplicate_count += duplicate_count_by_type[i];
2419
2420                fprintf(stderr, "%s statistics:\n", argv[0]);
2421                fprintf(stderr, "---------------------------------------------------------------------\n");
2422                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2423                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2424                fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
2425                fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
2426                fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
2427                fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
2428                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2429                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2430                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2431                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2432                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2433                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2434                fprintf(stderr, "---------------------------------------------------------------------\n");
2435                pack_report();
2436                fprintf(stderr, "---------------------------------------------------------------------\n");
2437                fprintf(stderr, "\n");
2438        }
2439
2440        return failure ? 1 : 0;
2441}