fast-import.con commit fast-import pull request (7e5dcea)
   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, va_list params)
 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        vfprintf(rpt, err, params);
 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        va_list x_params;
 446
 447        va_copy(x_params, params);
 448        fputs("fatal: ", stderr);
 449        vfprintf(stderr, err, params);
 450        fputc('\n', stderr);
 451
 452        if (!zombie) {
 453                zombie = 1;
 454                write_crash_report(err, x_params);
 455        }
 456        va_end(x_params);
 457        exit(128);
 458}
 459
 460static void alloc_objects(unsigned int cnt)
 461{
 462        struct object_entry_pool *b;
 463
 464        b = xmalloc(sizeof(struct object_entry_pool)
 465                + cnt * sizeof(struct object_entry));
 466        b->next_pool = blocks;
 467        b->next_free = b->entries;
 468        b->end = b->entries + cnt;
 469        blocks = b;
 470        alloc_count += cnt;
 471}
 472
 473static struct object_entry *new_object(unsigned char *sha1)
 474{
 475        struct object_entry *e;
 476
 477        if (blocks->next_free == blocks->end)
 478                alloc_objects(object_entry_alloc);
 479
 480        e = blocks->next_free++;
 481        hashcpy(e->sha1, sha1);
 482        return e;
 483}
 484
 485static struct object_entry *find_object(unsigned char *sha1)
 486{
 487        unsigned int h = sha1[0] << 8 | sha1[1];
 488        struct object_entry *e;
 489        for (e = object_table[h]; e; e = e->next)
 490                if (!hashcmp(sha1, e->sha1))
 491                        return e;
 492        return NULL;
 493}
 494
 495static struct object_entry *insert_object(unsigned char *sha1)
 496{
 497        unsigned int h = sha1[0] << 8 | sha1[1];
 498        struct object_entry *e = object_table[h];
 499        struct object_entry *p = NULL;
 500
 501        while (e) {
 502                if (!hashcmp(sha1, e->sha1))
 503                        return e;
 504                p = e;
 505                e = e->next;
 506        }
 507
 508        e = new_object(sha1);
 509        e->next = NULL;
 510        e->offset = 0;
 511        if (p)
 512                p->next = e;
 513        else
 514                object_table[h] = e;
 515        return e;
 516}
 517
 518static unsigned int hc_str(const char *s, size_t len)
 519{
 520        unsigned int r = 0;
 521        while (len-- > 0)
 522                r = r * 31 + *s++;
 523        return r;
 524}
 525
 526static void *pool_alloc(size_t len)
 527{
 528        struct mem_pool *p;
 529        void *r;
 530
 531        for (p = mem_pool; p; p = p->next_pool)
 532                if ((p->end - p->next_free >= len))
 533                        break;
 534
 535        if (!p) {
 536                if (len >= (mem_pool_alloc/2)) {
 537                        total_allocd += len;
 538                        return xmalloc(len);
 539                }
 540                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 541                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 542                p->next_pool = mem_pool;
 543                p->next_free = p->space;
 544                p->end = p->next_free + mem_pool_alloc;
 545                mem_pool = p;
 546        }
 547
 548        r = p->next_free;
 549        /* round out to a pointer alignment */
 550        if (len & (sizeof(void*) - 1))
 551                len += sizeof(void*) - (len & (sizeof(void*) - 1));
 552        p->next_free += len;
 553        return r;
 554}
 555
 556static void *pool_calloc(size_t count, size_t size)
 557{
 558        size_t len = count * size;
 559        void *r = pool_alloc(len);
 560        memset(r, 0, len);
 561        return r;
 562}
 563
 564static char *pool_strdup(const char *s)
 565{
 566        char *r = pool_alloc(strlen(s) + 1);
 567        strcpy(r, s);
 568        return r;
 569}
 570
 571static void size_dbuf(struct dbuf *b, size_t maxlen)
 572{
 573        if (b->buffer) {
 574                if (b->capacity >= maxlen)
 575                        return;
 576                free(b->buffer);
 577        }
 578        b->capacity = ((maxlen / 1024) + 1) * 1024;
 579        b->buffer = xmalloc(b->capacity);
 580}
 581
 582static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 583{
 584        struct mark_set *s = marks;
 585        while ((idnum >> s->shift) >= 1024) {
 586                s = pool_calloc(1, sizeof(struct mark_set));
 587                s->shift = marks->shift + 10;
 588                s->data.sets[0] = marks;
 589                marks = s;
 590        }
 591        while (s->shift) {
 592                uintmax_t i = idnum >> s->shift;
 593                idnum -= i << s->shift;
 594                if (!s->data.sets[i]) {
 595                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 596                        s->data.sets[i]->shift = s->shift - 10;
 597                }
 598                s = s->data.sets[i];
 599        }
 600        if (!s->data.marked[idnum])
 601                marks_set_count++;
 602        s->data.marked[idnum] = oe;
 603}
 604
 605static struct object_entry *find_mark(uintmax_t idnum)
 606{
 607        uintmax_t orig_idnum = idnum;
 608        struct mark_set *s = marks;
 609        struct object_entry *oe = NULL;
 610        if ((idnum >> s->shift) < 1024) {
 611                while (s && s->shift) {
 612                        uintmax_t i = idnum >> s->shift;
 613                        idnum -= i << s->shift;
 614                        s = s->data.sets[i];
 615                }
 616                if (s)
 617                        oe = s->data.marked[idnum];
 618        }
 619        if (!oe)
 620                die("mark :%" PRIuMAX " not declared", orig_idnum);
 621        return oe;
 622}
 623
 624static struct atom_str *to_atom(const char *s, unsigned short len)
 625{
 626        unsigned int hc = hc_str(s, len) % atom_table_sz;
 627        struct atom_str *c;
 628
 629        for (c = atom_table[hc]; c; c = c->next_atom)
 630                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 631                        return c;
 632
 633        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 634        c->str_len = len;
 635        strncpy(c->str_dat, s, len);
 636        c->str_dat[len] = 0;
 637        c->next_atom = atom_table[hc];
 638        atom_table[hc] = c;
 639        atom_cnt++;
 640        return c;
 641}
 642
 643static struct branch *lookup_branch(const char *name)
 644{
 645        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 646        struct branch *b;
 647
 648        for (b = branch_table[hc]; b; b = b->table_next_branch)
 649                if (!strcmp(name, b->name))
 650                        return b;
 651        return NULL;
 652}
 653
 654static struct branch *new_branch(const char *name)
 655{
 656        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 657        struct branch* b = lookup_branch(name);
 658
 659        if (b)
 660                die("Invalid attempt to create duplicate branch: %s", name);
 661        switch (check_ref_format(name)) {
 662        case  0: break; /* its valid */
 663        case -2: break; /* valid, but too few '/', allow anyway */
 664        default:
 665                die("Branch name doesn't conform to GIT standards: %s", name);
 666        }
 667
 668        b = pool_calloc(1, sizeof(struct branch));
 669        b->name = pool_strdup(name);
 670        b->table_next_branch = branch_table[hc];
 671        b->branch_tree.versions[0].mode = S_IFDIR;
 672        b->branch_tree.versions[1].mode = S_IFDIR;
 673        b->active = 0;
 674        b->pack_id = MAX_PACK_ID;
 675        branch_table[hc] = b;
 676        branch_count++;
 677        return b;
 678}
 679
 680static unsigned int hc_entries(unsigned int cnt)
 681{
 682        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 683        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 684}
 685
 686static struct tree_content *new_tree_content(unsigned int cnt)
 687{
 688        struct avail_tree_content *f, *l = NULL;
 689        struct tree_content *t;
 690        unsigned int hc = hc_entries(cnt);
 691
 692        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 693                if (f->entry_capacity >= cnt)
 694                        break;
 695
 696        if (f) {
 697                if (l)
 698                        l->next_avail = f->next_avail;
 699                else
 700                        avail_tree_table[hc] = f->next_avail;
 701        } else {
 702                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 703                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 704                f->entry_capacity = cnt;
 705        }
 706
 707        t = (struct tree_content*)f;
 708        t->entry_count = 0;
 709        t->delta_depth = 0;
 710        return t;
 711}
 712
 713static void release_tree_entry(struct tree_entry *e);
 714static void release_tree_content(struct tree_content *t)
 715{
 716        struct avail_tree_content *f = (struct avail_tree_content*)t;
 717        unsigned int hc = hc_entries(f->entry_capacity);
 718        f->next_avail = avail_tree_table[hc];
 719        avail_tree_table[hc] = f;
 720}
 721
 722static void release_tree_content_recursive(struct tree_content *t)
 723{
 724        unsigned int i;
 725        for (i = 0; i < t->entry_count; i++)
 726                release_tree_entry(t->entries[i]);
 727        release_tree_content(t);
 728}
 729
 730static struct tree_content *grow_tree_content(
 731        struct tree_content *t,
 732        int amt)
 733{
 734        struct tree_content *r = new_tree_content(t->entry_count + amt);
 735        r->entry_count = t->entry_count;
 736        r->delta_depth = t->delta_depth;
 737        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 738        release_tree_content(t);
 739        return r;
 740}
 741
 742static struct tree_entry *new_tree_entry(void)
 743{
 744        struct tree_entry *e;
 745
 746        if (!avail_tree_entry) {
 747                unsigned int n = tree_entry_alloc;
 748                total_allocd += n * sizeof(struct tree_entry);
 749                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 750                while (n-- > 1) {
 751                        *((void**)e) = e + 1;
 752                        e++;
 753                }
 754                *((void**)e) = NULL;
 755        }
 756
 757        e = avail_tree_entry;
 758        avail_tree_entry = *((void**)e);
 759        return e;
 760}
 761
 762static void release_tree_entry(struct tree_entry *e)
 763{
 764        if (e->tree)
 765                release_tree_content_recursive(e->tree);
 766        *((void**)e) = avail_tree_entry;
 767        avail_tree_entry = e;
 768}
 769
 770static struct tree_content *dup_tree_content(struct tree_content *s)
 771{
 772        struct tree_content *d;
 773        struct tree_entry *a, *b;
 774        unsigned int i;
 775
 776        if (!s)
 777                return NULL;
 778        d = new_tree_content(s->entry_count);
 779        for (i = 0; i < s->entry_count; i++) {
 780                a = s->entries[i];
 781                b = new_tree_entry();
 782                memcpy(b, a, sizeof(*a));
 783                if (a->tree && is_null_sha1(b->versions[1].sha1))
 784                        b->tree = dup_tree_content(a->tree);
 785                else
 786                        b->tree = NULL;
 787                d->entries[i] = b;
 788        }
 789        d->entry_count = s->entry_count;
 790        d->delta_depth = s->delta_depth;
 791
 792        return d;
 793}
 794
 795static void start_packfile(void)
 796{
 797        static char tmpfile[PATH_MAX];
 798        struct packed_git *p;
 799        struct pack_header hdr;
 800        int pack_fd;
 801
 802        snprintf(tmpfile, sizeof(tmpfile),
 803                "%s/tmp_pack_XXXXXX", get_object_directory());
 804        pack_fd = xmkstemp(tmpfile);
 805        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 806        strcpy(p->pack_name, tmpfile);
 807        p->pack_fd = pack_fd;
 808
 809        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 810        hdr.hdr_version = htonl(2);
 811        hdr.hdr_entries = 0;
 812        write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 813
 814        pack_data = p;
 815        pack_size = sizeof(hdr);
 816        object_count = 0;
 817
 818        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 819        all_packs[pack_id] = p;
 820}
 821
 822static int oecmp (const void *a_, const void *b_)
 823{
 824        struct object_entry *a = *((struct object_entry**)a_);
 825        struct object_entry *b = *((struct object_entry**)b_);
 826        return hashcmp(a->sha1, b->sha1);
 827}
 828
 829static char *create_index(void)
 830{
 831        static char tmpfile[PATH_MAX];
 832        SHA_CTX ctx;
 833        struct sha1file *f;
 834        struct object_entry **idx, **c, **last, *e;
 835        struct object_entry_pool *o;
 836        uint32_t array[256];
 837        int i, idx_fd;
 838
 839        /* Build the sorted table of object IDs. */
 840        idx = xmalloc(object_count * sizeof(struct object_entry*));
 841        c = idx;
 842        for (o = blocks; o; o = o->next_pool)
 843                for (e = o->next_free; e-- != o->entries;)
 844                        if (pack_id == e->pack_id)
 845                                *c++ = e;
 846        last = idx + object_count;
 847        if (c != last)
 848                die("internal consistency error creating the index");
 849        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 850
 851        /* Generate the fan-out array. */
 852        c = idx;
 853        for (i = 0; i < 256; i++) {
 854                struct object_entry **next = c;;
 855                while (next < last) {
 856                        if ((*next)->sha1[0] != i)
 857                                break;
 858                        next++;
 859                }
 860                array[i] = htonl(next - idx);
 861                c = next;
 862        }
 863
 864        snprintf(tmpfile, sizeof(tmpfile),
 865                "%s/tmp_idx_XXXXXX", get_object_directory());
 866        idx_fd = xmkstemp(tmpfile);
 867        f = sha1fd(idx_fd, tmpfile);
 868        sha1write(f, array, 256 * sizeof(int));
 869        SHA1_Init(&ctx);
 870        for (c = idx; c != last; c++) {
 871                uint32_t offset = htonl((*c)->offset);
 872                sha1write(f, &offset, 4);
 873                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 874                SHA1_Update(&ctx, (*c)->sha1, 20);
 875        }
 876        sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
 877        sha1close(f, NULL, 1);
 878        free(idx);
 879        SHA1_Final(pack_data->sha1, &ctx);
 880        return tmpfile;
 881}
 882
 883static char *keep_pack(char *curr_index_name)
 884{
 885        static char name[PATH_MAX];
 886        static const char *keep_msg = "fast-import";
 887        int keep_fd;
 888
 889        chmod(pack_data->pack_name, 0444);
 890        chmod(curr_index_name, 0444);
 891
 892        snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 893                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 894        keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
 895        if (keep_fd < 0)
 896                die("cannot create keep file");
 897        write(keep_fd, keep_msg, strlen(keep_msg));
 898        close(keep_fd);
 899
 900        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 901                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 902        if (move_temp_to_file(pack_data->pack_name, name))
 903                die("cannot store pack file");
 904
 905        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 906                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 907        if (move_temp_to_file(curr_index_name, name))
 908                die("cannot store index file");
 909        return name;
 910}
 911
 912static void unkeep_all_packs(void)
 913{
 914        static char name[PATH_MAX];
 915        int k;
 916
 917        for (k = 0; k < pack_id; k++) {
 918                struct packed_git *p = all_packs[k];
 919                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 920                         get_object_directory(), sha1_to_hex(p->sha1));
 921                unlink(name);
 922        }
 923}
 924
 925static void end_packfile(void)
 926{
 927        struct packed_git *old_p = pack_data, *new_p;
 928
 929        if (object_count) {
 930                char *idx_name;
 931                int i;
 932                struct branch *b;
 933                struct tag *t;
 934
 935                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 936                                    pack_data->pack_name, object_count);
 937                close(pack_data->pack_fd);
 938                idx_name = keep_pack(create_index());
 939
 940                /* Register the packfile with core git's machinary. */
 941                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 942                if (!new_p)
 943                        die("core git rejected index %s", idx_name);
 944                new_p->windows = old_p->windows;
 945                all_packs[pack_id] = new_p;
 946                install_packed_git(new_p);
 947
 948                /* Print the boundary */
 949                if (pack_edges) {
 950                        fprintf(pack_edges, "%s:", new_p->pack_name);
 951                        for (i = 0; i < branch_table_sz; i++) {
 952                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 953                                        if (b->pack_id == pack_id)
 954                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 955                                }
 956                        }
 957                        for (t = first_tag; t; t = t->next_tag) {
 958                                if (t->pack_id == pack_id)
 959                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 960                        }
 961                        fputc('\n', pack_edges);
 962                        fflush(pack_edges);
 963                }
 964
 965                pack_id++;
 966        }
 967        else
 968                unlink(old_p->pack_name);
 969        free(old_p);
 970
 971        /* We can't carry a delta across packfiles. */
 972        free(last_blob.data);
 973        last_blob.data = NULL;
 974        last_blob.len = 0;
 975        last_blob.offset = 0;
 976        last_blob.depth = 0;
 977}
 978
 979static void cycle_packfile(void)
 980{
 981        end_packfile();
 982        start_packfile();
 983}
 984
 985static size_t encode_header(
 986        enum object_type type,
 987        size_t size,
 988        unsigned char *hdr)
 989{
 990        int n = 1;
 991        unsigned char c;
 992
 993        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 994                die("bad type %d", type);
 995
 996        c = (type << 4) | (size & 15);
 997        size >>= 4;
 998        while (size) {
 999                *hdr++ = c | 0x80;
1000                c = size & 0x7f;
1001                size >>= 7;
1002                n++;
1003        }
1004        *hdr = c;
1005        return n;
1006}
1007
1008static int store_object(
1009        enum object_type type,
1010        void *dat,
1011        size_t datlen,
1012        struct last_object *last,
1013        unsigned char *sha1out,
1014        uintmax_t mark)
1015{
1016        void *out, *delta;
1017        struct object_entry *e;
1018        unsigned char hdr[96];
1019        unsigned char sha1[20];
1020        unsigned long hdrlen, deltalen;
1021        SHA_CTX c;
1022        z_stream s;
1023
1024        hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1025                (unsigned long)datlen) + 1;
1026        SHA1_Init(&c);
1027        SHA1_Update(&c, hdr, hdrlen);
1028        SHA1_Update(&c, dat, datlen);
1029        SHA1_Final(sha1, &c);
1030        if (sha1out)
1031                hashcpy(sha1out, sha1);
1032
1033        e = insert_object(sha1);
1034        if (mark)
1035                insert_mark(mark, e);
1036        if (e->offset) {
1037                duplicate_count_by_type[type]++;
1038                return 1;
1039        } else if (find_sha1_pack(sha1, packed_git)) {
1040                e->type = type;
1041                e->pack_id = MAX_PACK_ID;
1042                e->offset = 1; /* just not zero! */
1043                duplicate_count_by_type[type]++;
1044                return 1;
1045        }
1046
1047        if (last && last->data && last->depth < max_depth) {
1048                delta = diff_delta(last->data, last->len,
1049                        dat, datlen,
1050                        &deltalen, 0);
1051                if (delta && deltalen >= datlen) {
1052                        free(delta);
1053                        delta = NULL;
1054                }
1055        } else
1056                delta = NULL;
1057
1058        memset(&s, 0, sizeof(s));
1059        deflateInit(&s, zlib_compression_level);
1060        if (delta) {
1061                s.next_in = delta;
1062                s.avail_in = deltalen;
1063        } else {
1064                s.next_in = dat;
1065                s.avail_in = datlen;
1066        }
1067        s.avail_out = deflateBound(&s, s.avail_in);
1068        s.next_out = out = xmalloc(s.avail_out);
1069        while (deflate(&s, Z_FINISH) == Z_OK)
1070                /* nothing */;
1071        deflateEnd(&s);
1072
1073        /* Determine if we should auto-checkpoint. */
1074        if ((pack_size + 60 + s.total_out) > max_packsize
1075                || (pack_size + 60 + s.total_out) < pack_size) {
1076
1077                /* This new object needs to *not* have the current pack_id. */
1078                e->pack_id = pack_id + 1;
1079                cycle_packfile();
1080
1081                /* We cannot carry a delta into the new pack. */
1082                if (delta) {
1083                        free(delta);
1084                        delta = NULL;
1085
1086                        memset(&s, 0, sizeof(s));
1087                        deflateInit(&s, zlib_compression_level);
1088                        s.next_in = dat;
1089                        s.avail_in = datlen;
1090                        s.avail_out = deflateBound(&s, s.avail_in);
1091                        s.next_out = out = xrealloc(out, s.avail_out);
1092                        while (deflate(&s, Z_FINISH) == Z_OK)
1093                                /* nothing */;
1094                        deflateEnd(&s);
1095                }
1096        }
1097
1098        e->type = type;
1099        e->pack_id = pack_id;
1100        e->offset = pack_size;
1101        object_count++;
1102        object_count_by_type[type]++;
1103
1104        if (delta) {
1105                unsigned long ofs = e->offset - last->offset;
1106                unsigned pos = sizeof(hdr) - 1;
1107
1108                delta_count_by_type[type]++;
1109                last->depth++;
1110
1111                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1112                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1113                pack_size += hdrlen;
1114
1115                hdr[pos] = ofs & 127;
1116                while (ofs >>= 7)
1117                        hdr[--pos] = 128 | (--ofs & 127);
1118                write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1119                pack_size += sizeof(hdr) - pos;
1120        } else {
1121                if (last)
1122                        last->depth = 0;
1123                hdrlen = encode_header(type, datlen, hdr);
1124                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1125                pack_size += hdrlen;
1126        }
1127
1128        write_or_die(pack_data->pack_fd, out, s.total_out);
1129        pack_size += s.total_out;
1130
1131        free(out);
1132        free(delta);
1133        if (last) {
1134                if (!last->no_free)
1135                        free(last->data);
1136                last->data = dat;
1137                last->offset = e->offset;
1138                last->len = datlen;
1139        }
1140        return 0;
1141}
1142
1143static void *gfi_unpack_entry(
1144        struct object_entry *oe,
1145        unsigned long *sizep)
1146{
1147        enum object_type type;
1148        struct packed_git *p = all_packs[oe->pack_id];
1149        if (p == pack_data)
1150                p->pack_size = pack_size + 20;
1151        return unpack_entry(p, oe->offset, &type, sizep);
1152}
1153
1154static const char *get_mode(const char *str, uint16_t *modep)
1155{
1156        unsigned char c;
1157        uint16_t mode = 0;
1158
1159        while ((c = *str++) != ' ') {
1160                if (c < '0' || c > '7')
1161                        return NULL;
1162                mode = (mode << 3) + (c - '0');
1163        }
1164        *modep = mode;
1165        return str;
1166}
1167
1168static void load_tree(struct tree_entry *root)
1169{
1170        unsigned char* sha1 = root->versions[1].sha1;
1171        struct object_entry *myoe;
1172        struct tree_content *t;
1173        unsigned long size;
1174        char *buf;
1175        const char *c;
1176
1177        root->tree = t = new_tree_content(8);
1178        if (is_null_sha1(sha1))
1179                return;
1180
1181        myoe = find_object(sha1);
1182        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1183                if (myoe->type != OBJ_TREE)
1184                        die("Not a tree: %s", sha1_to_hex(sha1));
1185                t->delta_depth = 0;
1186                buf = gfi_unpack_entry(myoe, &size);
1187        } else {
1188                enum object_type type;
1189                buf = read_sha1_file(sha1, &type, &size);
1190                if (!buf || type != OBJ_TREE)
1191                        die("Can't load tree %s", sha1_to_hex(sha1));
1192        }
1193
1194        c = buf;
1195        while (c != (buf + size)) {
1196                struct tree_entry *e = new_tree_entry();
1197
1198                if (t->entry_count == t->entry_capacity)
1199                        root->tree = t = grow_tree_content(t, t->entry_count);
1200                t->entries[t->entry_count++] = e;
1201
1202                e->tree = NULL;
1203                c = get_mode(c, &e->versions[1].mode);
1204                if (!c)
1205                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1206                e->versions[0].mode = e->versions[1].mode;
1207                e->name = to_atom(c, strlen(c));
1208                c += e->name->str_len + 1;
1209                hashcpy(e->versions[0].sha1, (unsigned char*)c);
1210                hashcpy(e->versions[1].sha1, (unsigned char*)c);
1211                c += 20;
1212        }
1213        free(buf);
1214}
1215
1216static int tecmp0 (const void *_a, const void *_b)
1217{
1218        struct tree_entry *a = *((struct tree_entry**)_a);
1219        struct tree_entry *b = *((struct tree_entry**)_b);
1220        return base_name_compare(
1221                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1222                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1223}
1224
1225static int tecmp1 (const void *_a, const void *_b)
1226{
1227        struct tree_entry *a = *((struct tree_entry**)_a);
1228        struct tree_entry *b = *((struct tree_entry**)_b);
1229        return base_name_compare(
1230                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1231                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1232}
1233
1234static void mktree(struct tree_content *t,
1235        int v,
1236        unsigned long *szp,
1237        struct dbuf *b)
1238{
1239        size_t maxlen = 0;
1240        unsigned int i;
1241        char *c;
1242
1243        if (!v)
1244                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1245        else
1246                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1247
1248        for (i = 0; i < t->entry_count; i++) {
1249                if (t->entries[i]->versions[v].mode)
1250                        maxlen += t->entries[i]->name->str_len + 34;
1251        }
1252
1253        size_dbuf(b, maxlen);
1254        c = b->buffer;
1255        for (i = 0; i < t->entry_count; i++) {
1256                struct tree_entry *e = t->entries[i];
1257                if (!e->versions[v].mode)
1258                        continue;
1259                c += sprintf(c, "%o", (unsigned int)e->versions[v].mode);
1260                *c++ = ' ';
1261                strcpy(c, e->name->str_dat);
1262                c += e->name->str_len + 1;
1263                hashcpy((unsigned char*)c, e->versions[v].sha1);
1264                c += 20;
1265        }
1266        *szp = c - (char*)b->buffer;
1267}
1268
1269static void store_tree(struct tree_entry *root)
1270{
1271        struct tree_content *t = root->tree;
1272        unsigned int i, j, del;
1273        unsigned long new_len;
1274        struct last_object lo;
1275        struct object_entry *le;
1276
1277        if (!is_null_sha1(root->versions[1].sha1))
1278                return;
1279
1280        for (i = 0; i < t->entry_count; i++) {
1281                if (t->entries[i]->tree)
1282                        store_tree(t->entries[i]);
1283        }
1284
1285        le = find_object(root->versions[0].sha1);
1286        if (!S_ISDIR(root->versions[0].mode)
1287                || !le
1288                || le->pack_id != pack_id) {
1289                lo.data = NULL;
1290                lo.depth = 0;
1291                lo.no_free = 0;
1292        } else {
1293                mktree(t, 0, &lo.len, &old_tree);
1294                lo.data = old_tree.buffer;
1295                lo.offset = le->offset;
1296                lo.depth = t->delta_depth;
1297                lo.no_free = 1;
1298        }
1299
1300        mktree(t, 1, &new_len, &new_tree);
1301        store_object(OBJ_TREE, new_tree.buffer, new_len,
1302                &lo, root->versions[1].sha1, 0);
1303
1304        t->delta_depth = lo.depth;
1305        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1306                struct tree_entry *e = t->entries[i];
1307                if (e->versions[1].mode) {
1308                        e->versions[0].mode = e->versions[1].mode;
1309                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1310                        t->entries[j++] = e;
1311                } else {
1312                        release_tree_entry(e);
1313                        del++;
1314                }
1315        }
1316        t->entry_count -= del;
1317}
1318
1319static int tree_content_set(
1320        struct tree_entry *root,
1321        const char *p,
1322        const unsigned char *sha1,
1323        const uint16_t mode,
1324        struct tree_content *subtree)
1325{
1326        struct tree_content *t = root->tree;
1327        const char *slash1;
1328        unsigned int i, n;
1329        struct tree_entry *e;
1330
1331        slash1 = strchr(p, '/');
1332        if (slash1)
1333                n = slash1 - p;
1334        else
1335                n = strlen(p);
1336        if (!n)
1337                die("Empty path component found in input");
1338        if (!slash1 && !S_ISDIR(mode) && subtree)
1339                die("Non-directories cannot have subtrees");
1340
1341        for (i = 0; i < t->entry_count; i++) {
1342                e = t->entries[i];
1343                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1344                        if (!slash1) {
1345                                if (!S_ISDIR(mode)
1346                                                && e->versions[1].mode == mode
1347                                                && !hashcmp(e->versions[1].sha1, sha1))
1348                                        return 0;
1349                                e->versions[1].mode = mode;
1350                                hashcpy(e->versions[1].sha1, sha1);
1351                                if (e->tree)
1352                                        release_tree_content_recursive(e->tree);
1353                                e->tree = subtree;
1354                                hashclr(root->versions[1].sha1);
1355                                return 1;
1356                        }
1357                        if (!S_ISDIR(e->versions[1].mode)) {
1358                                e->tree = new_tree_content(8);
1359                                e->versions[1].mode = S_IFDIR;
1360                        }
1361                        if (!e->tree)
1362                                load_tree(e);
1363                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1364                                hashclr(root->versions[1].sha1);
1365                                return 1;
1366                        }
1367                        return 0;
1368                }
1369        }
1370
1371        if (t->entry_count == t->entry_capacity)
1372                root->tree = t = grow_tree_content(t, t->entry_count);
1373        e = new_tree_entry();
1374        e->name = to_atom(p, n);
1375        e->versions[0].mode = 0;
1376        hashclr(e->versions[0].sha1);
1377        t->entries[t->entry_count++] = e;
1378        if (slash1) {
1379                e->tree = new_tree_content(8);
1380                e->versions[1].mode = S_IFDIR;
1381                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1382        } else {
1383                e->tree = subtree;
1384                e->versions[1].mode = mode;
1385                hashcpy(e->versions[1].sha1, sha1);
1386        }
1387        hashclr(root->versions[1].sha1);
1388        return 1;
1389}
1390
1391static int tree_content_remove(
1392        struct tree_entry *root,
1393        const char *p,
1394        struct tree_entry *backup_leaf)
1395{
1396        struct tree_content *t = root->tree;
1397        const char *slash1;
1398        unsigned int i, n;
1399        struct tree_entry *e;
1400
1401        slash1 = strchr(p, '/');
1402        if (slash1)
1403                n = slash1 - p;
1404        else
1405                n = strlen(p);
1406
1407        for (i = 0; i < t->entry_count; i++) {
1408                e = t->entries[i];
1409                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1410                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1411                                goto del_entry;
1412                        if (!e->tree)
1413                                load_tree(e);
1414                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1415                                for (n = 0; n < e->tree->entry_count; n++) {
1416                                        if (e->tree->entries[n]->versions[1].mode) {
1417                                                hashclr(root->versions[1].sha1);
1418                                                return 1;
1419                                        }
1420                                }
1421                                backup_leaf = NULL;
1422                                goto del_entry;
1423                        }
1424                        return 0;
1425                }
1426        }
1427        return 0;
1428
1429del_entry:
1430        if (backup_leaf)
1431                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1432        else if (e->tree)
1433                release_tree_content_recursive(e->tree);
1434        e->tree = NULL;
1435        e->versions[1].mode = 0;
1436        hashclr(e->versions[1].sha1);
1437        hashclr(root->versions[1].sha1);
1438        return 1;
1439}
1440
1441static int tree_content_get(
1442        struct tree_entry *root,
1443        const char *p,
1444        struct tree_entry *leaf)
1445{
1446        struct tree_content *t = root->tree;
1447        const char *slash1;
1448        unsigned int i, n;
1449        struct tree_entry *e;
1450
1451        slash1 = strchr(p, '/');
1452        if (slash1)
1453                n = slash1 - p;
1454        else
1455                n = strlen(p);
1456
1457        for (i = 0; i < t->entry_count; i++) {
1458                e = t->entries[i];
1459                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1460                        if (!slash1) {
1461                                memcpy(leaf, e, sizeof(*leaf));
1462                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1463                                        leaf->tree = dup_tree_content(e->tree);
1464                                else
1465                                        leaf->tree = NULL;
1466                                return 1;
1467                        }
1468                        if (!S_ISDIR(e->versions[1].mode))
1469                                return 0;
1470                        if (!e->tree)
1471                                load_tree(e);
1472                        return tree_content_get(e, slash1 + 1, leaf);
1473                }
1474        }
1475        return 0;
1476}
1477
1478static int update_branch(struct branch *b)
1479{
1480        static const char *msg = "fast-import";
1481        struct ref_lock *lock;
1482        unsigned char old_sha1[20];
1483
1484        if (read_ref(b->name, old_sha1))
1485                hashclr(old_sha1);
1486        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1487        if (!lock)
1488                return error("Unable to lock %s", b->name);
1489        if (!force_update && !is_null_sha1(old_sha1)) {
1490                struct commit *old_cmit, *new_cmit;
1491
1492                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1493                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1494                if (!old_cmit || !new_cmit) {
1495                        unlock_ref(lock);
1496                        return error("Branch %s is missing commits.", b->name);
1497                }
1498
1499                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1500                        unlock_ref(lock);
1501                        warning("Not updating %s"
1502                                " (new tip %s does not contain %s)",
1503                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1504                        return -1;
1505                }
1506        }
1507        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1508                return error("Unable to update %s", b->name);
1509        return 0;
1510}
1511
1512static void dump_branches(void)
1513{
1514        unsigned int i;
1515        struct branch *b;
1516
1517        for (i = 0; i < branch_table_sz; i++) {
1518                for (b = branch_table[i]; b; b = b->table_next_branch)
1519                        failure |= update_branch(b);
1520        }
1521}
1522
1523static void dump_tags(void)
1524{
1525        static const char *msg = "fast-import";
1526        struct tag *t;
1527        struct ref_lock *lock;
1528        char ref_name[PATH_MAX];
1529
1530        for (t = first_tag; t; t = t->next_tag) {
1531                sprintf(ref_name, "tags/%s", t->name);
1532                lock = lock_ref_sha1(ref_name, NULL);
1533                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1534                        failure |= error("Unable to update %s", ref_name);
1535        }
1536}
1537
1538static void dump_marks_helper(FILE *f,
1539        uintmax_t base,
1540        struct mark_set *m)
1541{
1542        uintmax_t k;
1543        if (m->shift) {
1544                for (k = 0; k < 1024; k++) {
1545                        if (m->data.sets[k])
1546                                dump_marks_helper(f, (base + k) << m->shift,
1547                                        m->data.sets[k]);
1548                }
1549        } else {
1550                for (k = 0; k < 1024; k++) {
1551                        if (m->data.marked[k])
1552                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1553                                        sha1_to_hex(m->data.marked[k]->sha1));
1554                }
1555        }
1556}
1557
1558static void dump_marks(void)
1559{
1560        static struct lock_file mark_lock;
1561        int mark_fd;
1562        FILE *f;
1563
1564        if (!mark_file)
1565                return;
1566
1567        mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1568        if (mark_fd < 0) {
1569                failure |= error("Unable to write marks file %s: %s",
1570                        mark_file, strerror(errno));
1571                return;
1572        }
1573
1574        f = fdopen(mark_fd, "w");
1575        if (!f) {
1576                rollback_lock_file(&mark_lock);
1577                failure |= error("Unable to write marks file %s: %s",
1578                        mark_file, strerror(errno));
1579                return;
1580        }
1581
1582        dump_marks_helper(f, 0, marks);
1583        fclose(f);
1584        if (commit_lock_file(&mark_lock))
1585                failure |= error("Unable to write marks file %s: %s",
1586                        mark_file, strerror(errno));
1587}
1588
1589static void read_next_command(void)
1590{
1591        do {
1592                if (unread_command_buf) {
1593                        unread_command_buf = 0;
1594                        if (command_buf.eof)
1595                                return;
1596                } else {
1597                        struct recent_command *rc;
1598
1599                        command_buf.buf = NULL;
1600                        read_line(&command_buf, stdin, '\n');
1601                        if (command_buf.eof)
1602                                return;
1603
1604                        rc = rc_free;
1605                        if (rc)
1606                                rc_free = rc->next;
1607                        else {
1608                                rc = cmd_hist.next;
1609                                cmd_hist.next = rc->next;
1610                                cmd_hist.next->prev = &cmd_hist;
1611                                free(rc->buf);
1612                        }
1613
1614                        rc->buf = command_buf.buf;
1615                        rc->prev = cmd_tail;
1616                        rc->next = cmd_hist.prev;
1617                        rc->prev->next = rc;
1618                        cmd_tail = rc;
1619                }
1620        } while (command_buf.buf[0] == '#');
1621}
1622
1623static void skip_optional_lf(void)
1624{
1625        int term_char = fgetc(stdin);
1626        if (term_char != '\n' && term_char != EOF)
1627                ungetc(term_char, stdin);
1628}
1629
1630static void cmd_mark(void)
1631{
1632        if (!prefixcmp(command_buf.buf, "mark :")) {
1633                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1634                read_next_command();
1635        }
1636        else
1637                next_mark = 0;
1638}
1639
1640static void *cmd_data (size_t *size)
1641{
1642        size_t length;
1643        char *buffer;
1644
1645        if (prefixcmp(command_buf.buf, "data "))
1646                die("Expected 'data n' command, found: %s", command_buf.buf);
1647
1648        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1649                char *term = xstrdup(command_buf.buf + 5 + 2);
1650                size_t sz = 8192, term_len = command_buf.len - 5 - 2;
1651                length = 0;
1652                buffer = xmalloc(sz);
1653                command_buf.buf = NULL;
1654                for (;;) {
1655                        read_line(&command_buf, stdin, '\n');
1656                        if (command_buf.eof)
1657                                die("EOF in data (terminator '%s' not found)", term);
1658                        if (term_len == command_buf.len
1659                                && !strcmp(term, command_buf.buf))
1660                                break;
1661                        ALLOC_GROW(buffer, length + command_buf.len, sz);
1662                        memcpy(buffer + length,
1663                                command_buf.buf,
1664                                command_buf.len - 1);
1665                        length += command_buf.len - 1;
1666                        buffer[length++] = '\n';
1667                }
1668                free(term);
1669        }
1670        else {
1671                size_t n = 0;
1672                length = strtoul(command_buf.buf + 5, NULL, 10);
1673                buffer = xmalloc(length);
1674                while (n < length) {
1675                        size_t s = fread(buffer + n, 1, length - n, stdin);
1676                        if (!s && feof(stdin))
1677                                die("EOF in data (%lu bytes remaining)",
1678                                        (unsigned long)(length - n));
1679                        n += s;
1680                }
1681        }
1682
1683        skip_optional_lf();
1684        *size = length;
1685        return buffer;
1686}
1687
1688static int validate_raw_date(const char *src, char *result, int maxlen)
1689{
1690        const char *orig_src = src;
1691        char *endp, sign;
1692
1693        strtoul(src, &endp, 10);
1694        if (endp == src || *endp != ' ')
1695                return -1;
1696
1697        src = endp + 1;
1698        if (*src != '-' && *src != '+')
1699                return -1;
1700        sign = *src;
1701
1702        strtoul(src + 1, &endp, 10);
1703        if (endp == src || *endp || (endp - orig_src) >= maxlen)
1704                return -1;
1705
1706        strcpy(result, orig_src);
1707        return 0;
1708}
1709
1710static char *parse_ident(const char *buf)
1711{
1712        const char *gt;
1713        size_t name_len;
1714        char *ident;
1715
1716        gt = strrchr(buf, '>');
1717        if (!gt)
1718                die("Missing > in ident string: %s", buf);
1719        gt++;
1720        if (*gt != ' ')
1721                die("Missing space after > in ident string: %s", buf);
1722        gt++;
1723        name_len = gt - buf;
1724        ident = xmalloc(name_len + 24);
1725        strncpy(ident, buf, name_len);
1726
1727        switch (whenspec) {
1728        case WHENSPEC_RAW:
1729                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1730                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1731                break;
1732        case WHENSPEC_RFC2822:
1733                if (parse_date(gt, ident + name_len, 24) < 0)
1734                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1735                break;
1736        case WHENSPEC_NOW:
1737                if (strcmp("now", gt))
1738                        die("Date in ident must be 'now': %s", buf);
1739                datestamp(ident + name_len, 24);
1740                break;
1741        }
1742
1743        return ident;
1744}
1745
1746static void cmd_new_blob(void)
1747{
1748        size_t l;
1749        void *d;
1750
1751        read_next_command();
1752        cmd_mark();
1753        d = cmd_data(&l);
1754
1755        if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
1756                free(d);
1757}
1758
1759static void unload_one_branch(void)
1760{
1761        while (cur_active_branches
1762                && cur_active_branches >= max_active_branches) {
1763                uintmax_t min_commit = ULONG_MAX;
1764                struct branch *e, *l = NULL, *p = NULL;
1765
1766                for (e = active_branches; e; e = e->active_next_branch) {
1767                        if (e->last_commit < min_commit) {
1768                                p = l;
1769                                min_commit = e->last_commit;
1770                        }
1771                        l = e;
1772                }
1773
1774                if (p) {
1775                        e = p->active_next_branch;
1776                        p->active_next_branch = e->active_next_branch;
1777                } else {
1778                        e = active_branches;
1779                        active_branches = e->active_next_branch;
1780                }
1781                e->active = 0;
1782                e->active_next_branch = NULL;
1783                if (e->branch_tree.tree) {
1784                        release_tree_content_recursive(e->branch_tree.tree);
1785                        e->branch_tree.tree = NULL;
1786                }
1787                cur_active_branches--;
1788        }
1789}
1790
1791static void load_branch(struct branch *b)
1792{
1793        load_tree(&b->branch_tree);
1794        if (!b->active) {
1795                b->active = 1;
1796                b->active_next_branch = active_branches;
1797                active_branches = b;
1798                cur_active_branches++;
1799                branch_load_count++;
1800        }
1801}
1802
1803static void file_change_m(struct branch *b)
1804{
1805        const char *p = command_buf.buf + 2;
1806        char *p_uq;
1807        const char *endp;
1808        struct object_entry *oe = oe;
1809        unsigned char sha1[20];
1810        uint16_t mode, inline_data = 0;
1811
1812        p = get_mode(p, &mode);
1813        if (!p)
1814                die("Corrupt mode: %s", command_buf.buf);
1815        switch (mode) {
1816        case S_IFREG | 0644:
1817        case S_IFREG | 0755:
1818        case S_IFLNK:
1819        case 0644:
1820        case 0755:
1821                /* ok */
1822                break;
1823        default:
1824                die("Corrupt mode: %s", command_buf.buf);
1825        }
1826
1827        if (*p == ':') {
1828                char *x;
1829                oe = find_mark(strtoumax(p + 1, &x, 10));
1830                hashcpy(sha1, oe->sha1);
1831                p = x;
1832        } else if (!prefixcmp(p, "inline")) {
1833                inline_data = 1;
1834                p += 6;
1835        } else {
1836                if (get_sha1_hex(p, sha1))
1837                        die("Invalid SHA1: %s", command_buf.buf);
1838                oe = find_object(sha1);
1839                p += 40;
1840        }
1841        if (*p++ != ' ')
1842                die("Missing space after SHA1: %s", command_buf.buf);
1843
1844        p_uq = unquote_c_style(p, &endp);
1845        if (p_uq) {
1846                if (*endp)
1847                        die("Garbage after path in: %s", command_buf.buf);
1848                p = p_uq;
1849        }
1850
1851        if (inline_data) {
1852                size_t l;
1853                void *d;
1854                if (!p_uq)
1855                        p = p_uq = xstrdup(p);
1856                read_next_command();
1857                d = cmd_data(&l);
1858                if (store_object(OBJ_BLOB, d, l, &last_blob, sha1, 0))
1859                        free(d);
1860        } else if (oe) {
1861                if (oe->type != OBJ_BLOB)
1862                        die("Not a blob (actually a %s): %s",
1863                                command_buf.buf, typename(oe->type));
1864        } else {
1865                enum object_type type = sha1_object_info(sha1, NULL);
1866                if (type < 0)
1867                        die("Blob not found: %s", command_buf.buf);
1868                if (type != OBJ_BLOB)
1869                        die("Not a blob (actually a %s): %s",
1870                            typename(type), command_buf.buf);
1871        }
1872
1873        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1874        free(p_uq);
1875}
1876
1877static void file_change_d(struct branch *b)
1878{
1879        const char *p = command_buf.buf + 2;
1880        char *p_uq;
1881        const char *endp;
1882
1883        p_uq = unquote_c_style(p, &endp);
1884        if (p_uq) {
1885                if (*endp)
1886                        die("Garbage after path in: %s", command_buf.buf);
1887                p = p_uq;
1888        }
1889        tree_content_remove(&b->branch_tree, p, NULL);
1890        free(p_uq);
1891}
1892
1893static void file_change_cr(struct branch *b, int rename)
1894{
1895        const char *s, *d;
1896        char *s_uq, *d_uq;
1897        const char *endp;
1898        struct tree_entry leaf;
1899
1900        s = command_buf.buf + 2;
1901        s_uq = unquote_c_style(s, &endp);
1902        if (s_uq) {
1903                if (*endp != ' ')
1904                        die("Missing space after source: %s", command_buf.buf);
1905        }
1906        else {
1907                endp = strchr(s, ' ');
1908                if (!endp)
1909                        die("Missing space after source: %s", command_buf.buf);
1910                s_uq = xmalloc(endp - s + 1);
1911                memcpy(s_uq, s, endp - s);
1912                s_uq[endp - s] = 0;
1913        }
1914        s = s_uq;
1915
1916        endp++;
1917        if (!*endp)
1918                die("Missing dest: %s", command_buf.buf);
1919
1920        d = endp;
1921        d_uq = unquote_c_style(d, &endp);
1922        if (d_uq) {
1923                if (*endp)
1924                        die("Garbage after dest in: %s", command_buf.buf);
1925                d = d_uq;
1926        }
1927
1928        memset(&leaf, 0, sizeof(leaf));
1929        if (rename)
1930                tree_content_remove(&b->branch_tree, s, &leaf);
1931        else
1932                tree_content_get(&b->branch_tree, s, &leaf);
1933        if (!leaf.versions[1].mode)
1934                die("Path %s not in branch", s);
1935        tree_content_set(&b->branch_tree, d,
1936                leaf.versions[1].sha1,
1937                leaf.versions[1].mode,
1938                leaf.tree);
1939
1940        free(s_uq);
1941        free(d_uq);
1942}
1943
1944static void file_change_deleteall(struct branch *b)
1945{
1946        release_tree_content_recursive(b->branch_tree.tree);
1947        hashclr(b->branch_tree.versions[0].sha1);
1948        hashclr(b->branch_tree.versions[1].sha1);
1949        load_tree(&b->branch_tree);
1950}
1951
1952static void cmd_from_commit(struct branch *b, char *buf, unsigned long size)
1953{
1954        if (!buf || size < 46)
1955                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
1956        if (memcmp("tree ", buf, 5)
1957                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1958                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1959        hashcpy(b->branch_tree.versions[0].sha1,
1960                b->branch_tree.versions[1].sha1);
1961}
1962
1963static void cmd_from_existing(struct branch *b)
1964{
1965        if (is_null_sha1(b->sha1)) {
1966                hashclr(b->branch_tree.versions[0].sha1);
1967                hashclr(b->branch_tree.versions[1].sha1);
1968        } else {
1969                unsigned long size;
1970                char *buf;
1971
1972                buf = read_object_with_reference(b->sha1,
1973                        commit_type, &size, b->sha1);
1974                cmd_from_commit(b, buf, size);
1975                free(buf);
1976        }
1977}
1978
1979static int cmd_from(struct branch *b)
1980{
1981        const char *from;
1982        struct branch *s;
1983
1984        if (prefixcmp(command_buf.buf, "from "))
1985                return 0;
1986
1987        if (b->branch_tree.tree) {
1988                release_tree_content_recursive(b->branch_tree.tree);
1989                b->branch_tree.tree = NULL;
1990        }
1991
1992        from = strchr(command_buf.buf, ' ') + 1;
1993        s = lookup_branch(from);
1994        if (b == s)
1995                die("Can't create a branch from itself: %s", b->name);
1996        else if (s) {
1997                unsigned char *t = s->branch_tree.versions[1].sha1;
1998                hashcpy(b->sha1, s->sha1);
1999                hashcpy(b->branch_tree.versions[0].sha1, t);
2000                hashcpy(b->branch_tree.versions[1].sha1, t);
2001        } else if (*from == ':') {
2002                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2003                struct object_entry *oe = find_mark(idnum);
2004                if (oe->type != OBJ_COMMIT)
2005                        die("Mark :%" PRIuMAX " not a commit", idnum);
2006                hashcpy(b->sha1, oe->sha1);
2007                if (oe->pack_id != MAX_PACK_ID) {
2008                        unsigned long size;
2009                        char *buf = gfi_unpack_entry(oe, &size);
2010                        cmd_from_commit(b, buf, size);
2011                        free(buf);
2012                } else
2013                        cmd_from_existing(b);
2014        } else if (!get_sha1(from, b->sha1))
2015                cmd_from_existing(b);
2016        else
2017                die("Invalid ref name or SHA1 expression: %s", from);
2018
2019        read_next_command();
2020        return 1;
2021}
2022
2023static struct hash_list *cmd_merge(unsigned int *count)
2024{
2025        struct hash_list *list = NULL, *n, *e = e;
2026        const char *from;
2027        struct branch *s;
2028
2029        *count = 0;
2030        while (!prefixcmp(command_buf.buf, "merge ")) {
2031                from = strchr(command_buf.buf, ' ') + 1;
2032                n = xmalloc(sizeof(*n));
2033                s = lookup_branch(from);
2034                if (s)
2035                        hashcpy(n->sha1, s->sha1);
2036                else if (*from == ':') {
2037                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2038                        struct object_entry *oe = find_mark(idnum);
2039                        if (oe->type != OBJ_COMMIT)
2040                                die("Mark :%" PRIuMAX " not a commit", idnum);
2041                        hashcpy(n->sha1, oe->sha1);
2042                } else if (!get_sha1(from, n->sha1)) {
2043                        unsigned long size;
2044                        char *buf = read_object_with_reference(n->sha1,
2045                                commit_type, &size, n->sha1);
2046                        if (!buf || size < 46)
2047                                die("Not a valid commit: %s", from);
2048                        free(buf);
2049                } else
2050                        die("Invalid ref name or SHA1 expression: %s", from);
2051
2052                n->next = NULL;
2053                if (list)
2054                        e->next = n;
2055                else
2056                        list = n;
2057                e = n;
2058                (*count)++;
2059                read_next_command();
2060        }
2061        return list;
2062}
2063
2064static void cmd_new_commit(void)
2065{
2066        struct branch *b;
2067        void *msg;
2068        size_t msglen;
2069        char *sp;
2070        char *author = NULL;
2071        char *committer = NULL;
2072        struct hash_list *merge_list = NULL;
2073        unsigned int merge_count;
2074
2075        /* Obtain the branch name from the rest of our command */
2076        sp = strchr(command_buf.buf, ' ') + 1;
2077        b = lookup_branch(sp);
2078        if (!b)
2079                b = new_branch(sp);
2080
2081        read_next_command();
2082        cmd_mark();
2083        if (!prefixcmp(command_buf.buf, "author ")) {
2084                author = parse_ident(command_buf.buf + 7);
2085                read_next_command();
2086        }
2087        if (!prefixcmp(command_buf.buf, "committer ")) {
2088                committer = parse_ident(command_buf.buf + 10);
2089                read_next_command();
2090        }
2091        if (!committer)
2092                die("Expected committer but didn't get one");
2093        msg = cmd_data(&msglen);
2094        read_next_command();
2095        cmd_from(b);
2096        merge_list = cmd_merge(&merge_count);
2097
2098        /* ensure the branch is active/loaded */
2099        if (!b->branch_tree.tree || !max_active_branches) {
2100                unload_one_branch();
2101                load_branch(b);
2102        }
2103
2104        /* file_change* */
2105        while (!command_buf.eof && command_buf.len > 1) {
2106                if (!prefixcmp(command_buf.buf, "M "))
2107                        file_change_m(b);
2108                else if (!prefixcmp(command_buf.buf, "D "))
2109                        file_change_d(b);
2110                else if (!prefixcmp(command_buf.buf, "R "))
2111                        file_change_cr(b, 1);
2112                else if (!prefixcmp(command_buf.buf, "C "))
2113                        file_change_cr(b, 0);
2114                else if (!strcmp("deleteall", command_buf.buf))
2115                        file_change_deleteall(b);
2116                else {
2117                        unread_command_buf = 1;
2118                        break;
2119                }
2120                read_next_command();
2121        }
2122
2123        /* build the tree and the commit */
2124        store_tree(&b->branch_tree);
2125        hashcpy(b->branch_tree.versions[0].sha1,
2126                b->branch_tree.versions[1].sha1);
2127        size_dbuf(&new_data, 114 + msglen
2128                + merge_count * 49
2129                + (author
2130                        ? strlen(author) + strlen(committer)
2131                        : 2 * strlen(committer)));
2132        sp = new_data.buffer;
2133        sp += sprintf(sp, "tree %s\n",
2134                sha1_to_hex(b->branch_tree.versions[1].sha1));
2135        if (!is_null_sha1(b->sha1))
2136                sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
2137        while (merge_list) {
2138                struct hash_list *next = merge_list->next;
2139                sp += sprintf(sp, "parent %s\n", sha1_to_hex(merge_list->sha1));
2140                free(merge_list);
2141                merge_list = next;
2142        }
2143        sp += sprintf(sp, "author %s\n", author ? author : committer);
2144        sp += sprintf(sp, "committer %s\n", committer);
2145        *sp++ = '\n';
2146        memcpy(sp, msg, msglen);
2147        sp += msglen;
2148        free(author);
2149        free(committer);
2150        free(msg);
2151
2152        if (!store_object(OBJ_COMMIT,
2153                new_data.buffer, sp - (char*)new_data.buffer,
2154                NULL, b->sha1, next_mark))
2155                b->pack_id = pack_id;
2156        b->last_commit = object_count_by_type[OBJ_COMMIT];
2157}
2158
2159static void cmd_new_tag(void)
2160{
2161        char *sp;
2162        const char *from;
2163        char *tagger;
2164        struct branch *s;
2165        void *msg;
2166        size_t msglen;
2167        struct tag *t;
2168        uintmax_t from_mark = 0;
2169        unsigned char sha1[20];
2170
2171        /* Obtain the new tag name from the rest of our command */
2172        sp = strchr(command_buf.buf, ' ') + 1;
2173        t = pool_alloc(sizeof(struct tag));
2174        t->next_tag = NULL;
2175        t->name = pool_strdup(sp);
2176        if (last_tag)
2177                last_tag->next_tag = t;
2178        else
2179                first_tag = t;
2180        last_tag = t;
2181        read_next_command();
2182
2183        /* from ... */
2184        if (prefixcmp(command_buf.buf, "from "))
2185                die("Expected from command, got %s", command_buf.buf);
2186        from = strchr(command_buf.buf, ' ') + 1;
2187        s = lookup_branch(from);
2188        if (s) {
2189                hashcpy(sha1, s->sha1);
2190        } else if (*from == ':') {
2191                struct object_entry *oe;
2192                from_mark = strtoumax(from + 1, NULL, 10);
2193                oe = find_mark(from_mark);
2194                if (oe->type != OBJ_COMMIT)
2195                        die("Mark :%" PRIuMAX " not a commit", from_mark);
2196                hashcpy(sha1, oe->sha1);
2197        } else if (!get_sha1(from, sha1)) {
2198                unsigned long size;
2199                char *buf;
2200
2201                buf = read_object_with_reference(sha1,
2202                        commit_type, &size, sha1);
2203                if (!buf || size < 46)
2204                        die("Not a valid commit: %s", from);
2205                free(buf);
2206        } else
2207                die("Invalid ref name or SHA1 expression: %s", from);
2208        read_next_command();
2209
2210        /* tagger ... */
2211        if (prefixcmp(command_buf.buf, "tagger "))
2212                die("Expected tagger command, got %s", command_buf.buf);
2213        tagger = parse_ident(command_buf.buf + 7);
2214
2215        /* tag payload/message */
2216        read_next_command();
2217        msg = cmd_data(&msglen);
2218
2219        /* build the tag object */
2220        size_dbuf(&new_data, 67+strlen(t->name)+strlen(tagger)+msglen);
2221        sp = new_data.buffer;
2222        sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
2223        sp += sprintf(sp, "type %s\n", commit_type);
2224        sp += sprintf(sp, "tag %s\n", t->name);
2225        sp += sprintf(sp, "tagger %s\n", tagger);
2226        *sp++ = '\n';
2227        memcpy(sp, msg, msglen);
2228        sp += msglen;
2229        free(tagger);
2230        free(msg);
2231
2232        if (store_object(OBJ_TAG, new_data.buffer,
2233                sp - (char*)new_data.buffer,
2234                NULL, t->sha1, 0))
2235                t->pack_id = MAX_PACK_ID;
2236        else
2237                t->pack_id = pack_id;
2238}
2239
2240static void cmd_reset_branch(void)
2241{
2242        struct branch *b;
2243        char *sp;
2244
2245        /* Obtain the branch name from the rest of our command */
2246        sp = strchr(command_buf.buf, ' ') + 1;
2247        b = lookup_branch(sp);
2248        if (b) {
2249                hashclr(b->sha1);
2250                hashclr(b->branch_tree.versions[0].sha1);
2251                hashclr(b->branch_tree.versions[1].sha1);
2252                if (b->branch_tree.tree) {
2253                        release_tree_content_recursive(b->branch_tree.tree);
2254                        b->branch_tree.tree = NULL;
2255                }
2256        }
2257        else
2258                b = new_branch(sp);
2259        read_next_command();
2260        if (!cmd_from(b) && command_buf.len > 1)
2261                unread_command_buf = 1;
2262}
2263
2264static void cmd_checkpoint(void)
2265{
2266        if (object_count) {
2267                cycle_packfile();
2268                dump_branches();
2269                dump_tags();
2270                dump_marks();
2271        }
2272        skip_optional_lf();
2273}
2274
2275static void cmd_progress(void)
2276{
2277        fwrite(command_buf.buf, 1, command_buf.len - 1, stdout);
2278        fputc('\n', stdout);
2279        fflush(stdout);
2280        skip_optional_lf();
2281}
2282
2283static void import_marks(const char *input_file)
2284{
2285        char line[512];
2286        FILE *f = fopen(input_file, "r");
2287        if (!f)
2288                die("cannot read %s: %s", input_file, strerror(errno));
2289        while (fgets(line, sizeof(line), f)) {
2290                uintmax_t mark;
2291                char *end;
2292                unsigned char sha1[20];
2293                struct object_entry *e;
2294
2295                end = strchr(line, '\n');
2296                if (line[0] != ':' || !end)
2297                        die("corrupt mark line: %s", line);
2298                *end = 0;
2299                mark = strtoumax(line + 1, &end, 10);
2300                if (!mark || end == line + 1
2301                        || *end != ' ' || get_sha1(end + 1, sha1))
2302                        die("corrupt mark line: %s", line);
2303                e = find_object(sha1);
2304                if (!e) {
2305                        enum object_type type = sha1_object_info(sha1, NULL);
2306                        if (type < 0)
2307                                die("object not found: %s", sha1_to_hex(sha1));
2308                        e = insert_object(sha1);
2309                        e->type = type;
2310                        e->pack_id = MAX_PACK_ID;
2311                        e->offset = 1; /* just not zero! */
2312                }
2313                insert_mark(mark, e);
2314        }
2315        fclose(f);
2316}
2317
2318static const char fast_import_usage[] =
2319"git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2320
2321int main(int argc, const char **argv)
2322{
2323        unsigned int i, show_stats = 1;
2324
2325        git_config(git_default_config);
2326        alloc_objects(object_entry_alloc);
2327        strbuf_init(&command_buf);
2328        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2329        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2330        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2331        marks = pool_calloc(1, sizeof(struct mark_set));
2332
2333        for (i = 1; i < argc; i++) {
2334                const char *a = argv[i];
2335
2336                if (*a != '-' || !strcmp(a, "--"))
2337                        break;
2338                else if (!prefixcmp(a, "--date-format=")) {
2339                        const char *fmt = a + 14;
2340                        if (!strcmp(fmt, "raw"))
2341                                whenspec = WHENSPEC_RAW;
2342                        else if (!strcmp(fmt, "rfc2822"))
2343                                whenspec = WHENSPEC_RFC2822;
2344                        else if (!strcmp(fmt, "now"))
2345                                whenspec = WHENSPEC_NOW;
2346                        else
2347                                die("unknown --date-format argument %s", fmt);
2348                }
2349                else if (!prefixcmp(a, "--max-pack-size="))
2350                        max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2351                else if (!prefixcmp(a, "--depth="))
2352                        max_depth = strtoul(a + 8, NULL, 0);
2353                else if (!prefixcmp(a, "--active-branches="))
2354                        max_active_branches = strtoul(a + 18, NULL, 0);
2355                else if (!prefixcmp(a, "--import-marks="))
2356                        import_marks(a + 15);
2357                else if (!prefixcmp(a, "--export-marks="))
2358                        mark_file = a + 15;
2359                else if (!prefixcmp(a, "--export-pack-edges=")) {
2360                        if (pack_edges)
2361                                fclose(pack_edges);
2362                        pack_edges = fopen(a + 20, "a");
2363                        if (!pack_edges)
2364                                die("Cannot open %s: %s", a + 20, strerror(errno));
2365                } else if (!strcmp(a, "--force"))
2366                        force_update = 1;
2367                else if (!strcmp(a, "--quiet"))
2368                        show_stats = 0;
2369                else if (!strcmp(a, "--stats"))
2370                        show_stats = 1;
2371                else
2372                        die("unknown option %s", a);
2373        }
2374        if (i != argc)
2375                usage(fast_import_usage);
2376
2377        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2378        for (i = 0; i < (cmd_save - 1); i++)
2379                rc_free[i].next = &rc_free[i + 1];
2380        rc_free[cmd_save - 1].next = NULL;
2381
2382        prepare_packed_git();
2383        start_packfile();
2384        set_die_routine(die_nicely);
2385        for (;;) {
2386                read_next_command();
2387                if (command_buf.eof)
2388                        break;
2389                else if (!strcmp("blob", command_buf.buf))
2390                        cmd_new_blob();
2391                else if (!prefixcmp(command_buf.buf, "commit "))
2392                        cmd_new_commit();
2393                else if (!prefixcmp(command_buf.buf, "tag "))
2394                        cmd_new_tag();
2395                else if (!prefixcmp(command_buf.buf, "reset "))
2396                        cmd_reset_branch();
2397                else if (!strcmp("checkpoint", command_buf.buf))
2398                        cmd_checkpoint();
2399                else if (!prefixcmp(command_buf.buf, "progress "))
2400                        cmd_progress();
2401                else
2402                        die("Unsupported command: %s", command_buf.buf);
2403        }
2404        end_packfile();
2405
2406        dump_branches();
2407        dump_tags();
2408        unkeep_all_packs();
2409        dump_marks();
2410
2411        if (pack_edges)
2412                fclose(pack_edges);
2413
2414        if (show_stats) {
2415                uintmax_t total_count = 0, duplicate_count = 0;
2416                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2417                        total_count += object_count_by_type[i];
2418                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2419                        duplicate_count += duplicate_count_by_type[i];
2420
2421                fprintf(stderr, "%s statistics:\n", argv[0]);
2422                fprintf(stderr, "---------------------------------------------------------------------\n");
2423                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2424                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2425                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]);
2426                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]);
2427                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]);
2428                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]);
2429                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2430                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2431                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2432                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2433                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2434                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2435                fprintf(stderr, "---------------------------------------------------------------------\n");
2436                pack_report();
2437                fprintf(stderr, "---------------------------------------------------------------------\n");
2438                fprintf(stderr, "\n");
2439        }
2440
2441        return failure ? 1 : 0;
2442}