3609c24288dedb5b5e1b46d77ebfce6c305a434e
   1/*
   2Format of STDIN stream:
   3
   4  stream ::= cmd*;
   5
   6  cmd ::= new_blob
   7        | new_commit
   8        | new_tag
   9        | reset_branch
  10        | checkpoint
  11        | progress
  12        ;
  13
  14  new_blob ::= 'blob' lf
  15    mark?
  16    file_content;
  17  file_content ::= data;
  18
  19  new_commit ::= 'commit' sp ref_str lf
  20    mark?
  21    ('author' sp name '<' email '>' when lf)?
  22    'committer' sp name '<' email '>' when lf
  23    commit_msg
  24    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  25    ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
  26    file_change*
  27    lf?;
  28  commit_msg ::= data;
  29
  30  file_change ::= file_clr
  31    | file_del
  32    | file_rnm
  33    | file_cpy
  34    | file_obm
  35    | file_inm;
  36  file_clr ::= 'deleteall' lf;
  37  file_del ::= 'D' sp path_str lf;
  38  file_rnm ::= 'R' sp path_str sp path_str lf;
  39  file_cpy ::= 'C' sp path_str sp path_str lf;
  40  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  41  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  42    data;
  43
  44  new_tag ::= 'tag' sp tag_str lf
  45    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
  46    'tagger' sp name '<' email '>' when lf
  47    tag_msg;
  48  tag_msg ::= data;
  49
  50  reset_branch ::= 'reset' sp ref_str lf
  51    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  52    lf?;
  53
  54  checkpoint ::= 'checkpoint' lf
  55    lf?;
  56
  57  progress ::= 'progress' sp not_lf* lf
  58    lf?;
  59
  60     # note: the first idnum in a stream should be 1 and subsequent
  61     # idnums should not have gaps between values as this will cause
  62     # the stream parser to reserve space for the gapped values.  An
  63     # idnum can be updated in the future to a new object by issuing
  64     # a new mark directive with the old idnum.
  65     #
  66  mark ::= 'mark' sp idnum lf;
  67  data ::= (delimited_data | exact_data)
  68    lf?;
  69
  70    # note: delim may be any string but must not contain lf.
  71    # data_line may contain any data but must not be exactly
  72    # delim.
  73  delimited_data ::= 'data' sp '<<' delim lf
  74    (data_line lf)*
  75    delim lf;
  76
  77     # note: declen indicates the length of binary_data in bytes.
  78     # declen does not include the lf preceeding the binary data.
  79     #
  80  exact_data ::= 'data' sp declen lf
  81    binary_data;
  82
  83     # note: quoted strings are C-style quoting supporting \c for
  84     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  85     # is the signed byte value in octal.  Note that the only
  86     # characters which must actually be escaped to protect the
  87     # stream formatting is: \, " and LF.  Otherwise these values
  88     # are UTF8.
  89     #
  90  ref_str     ::= ref;
  91  sha1exp_str ::= sha1exp;
  92  tag_str     ::= tag;
  93  path_str    ::= path    | '"' quoted(path)    '"' ;
  94  mode        ::= '100644' | '644'
  95                | '100755' | '755'
  96                | '120000'
  97                ;
  98
  99  declen ::= # unsigned 32 bit value, ascii base10 notation;
 100  bigint ::= # unsigned integer value, ascii base10 notation;
 101  binary_data ::= # file content, not interpreted;
 102
 103  when         ::= raw_when | rfc2822_when;
 104  raw_when     ::= ts sp tz;
 105  rfc2822_when ::= # Valid RFC 2822 date and time;
 106
 107  sp ::= # ASCII space character;
 108  lf ::= # ASCII newline (LF) character;
 109
 110     # note: a colon (':') must precede the numerical value assigned to
 111     # an idnum.  This is to distinguish it from a ref or tag name as
 112     # GIT does not permit ':' in ref or tag strings.
 113     #
 114  idnum   ::= ':' bigint;
 115  path    ::= # GIT style file path, e.g. "a/b/c";
 116  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 117  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 118  sha1exp ::= # Any valid GIT SHA1 expression;
 119  hexsha1 ::= # SHA1 in hexadecimal format;
 120
 121     # note: name and email are UTF8 strings, however name must not
 122     # contain '<' or lf and email must not contain any of the
 123     # following: '<', '>', lf.
 124     #
 125  name  ::= # valid GIT author/committer name;
 126  email ::= # valid GIT author/committer email;
 127  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 128  tz    ::= # GIT style timezone;
 129
 130     # note: comments may appear anywhere in the input, except
 131     # within a data command.  Any form of the data command
 132     # always escapes the related input from comment processing.
 133     #
 134     # In case it is not clear, the '#' that starts the comment
 135     # must be the first character on that the line (an lf have
 136     # preceeded it).
 137     #
 138  comment ::= '#' not_lf* lf;
 139  not_lf  ::= # Any byte that is not ASCII newline (LF);
 140*/
 141
 142#include "builtin.h"
 143#include "cache.h"
 144#include "object.h"
 145#include "blob.h"
 146#include "tree.h"
 147#include "commit.h"
 148#include "delta.h"
 149#include "pack.h"
 150#include "refs.h"
 151#include "csum-file.h"
 152#include "quote.h"
 153
 154#define PACK_ID_BITS 16
 155#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 156#define DEPTH_BITS 13
 157#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 158
 159struct object_entry
 160{
 161        struct object_entry *next;
 162        uint32_t offset;
 163        uint32_t type : TYPE_BITS,
 164                pack_id : PACK_ID_BITS,
 165                depth : DEPTH_BITS;
 166        unsigned char sha1[20];
 167};
 168
 169struct object_entry_pool
 170{
 171        struct object_entry_pool *next_pool;
 172        struct object_entry *next_free;
 173        struct object_entry *end;
 174        struct object_entry entries[FLEX_ARRAY]; /* more */
 175};
 176
 177struct mark_set
 178{
 179        union {
 180                struct object_entry *marked[1024];
 181                struct mark_set *sets[1024];
 182        } data;
 183        unsigned int shift;
 184};
 185
 186struct last_object
 187{
 188        struct strbuf data;
 189        uint32_t offset;
 190        unsigned int depth;
 191        unsigned no_swap : 1;
 192};
 193
 194struct mem_pool
 195{
 196        struct mem_pool *next_pool;
 197        char *next_free;
 198        char *end;
 199        uintmax_t space[FLEX_ARRAY]; /* more */
 200};
 201
 202struct atom_str
 203{
 204        struct atom_str *next_atom;
 205        unsigned short str_len;
 206        char str_dat[FLEX_ARRAY]; /* more */
 207};
 208
 209struct tree_content;
 210struct tree_entry
 211{
 212        struct tree_content *tree;
 213        struct atom_str* name;
 214        struct tree_entry_ms
 215        {
 216                uint16_t mode;
 217                unsigned char sha1[20];
 218        } versions[2];
 219};
 220
 221struct tree_content
 222{
 223        unsigned int entry_capacity; /* must match avail_tree_content */
 224        unsigned int entry_count;
 225        unsigned int delta_depth;
 226        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 227};
 228
 229struct avail_tree_content
 230{
 231        unsigned int entry_capacity; /* must match tree_content */
 232        struct avail_tree_content *next_avail;
 233};
 234
 235struct branch
 236{
 237        struct branch *table_next_branch;
 238        struct branch *active_next_branch;
 239        const char *name;
 240        struct tree_entry branch_tree;
 241        uintmax_t last_commit;
 242        unsigned active : 1;
 243        unsigned pack_id : PACK_ID_BITS;
 244        unsigned char sha1[20];
 245};
 246
 247struct tag
 248{
 249        struct tag *next_tag;
 250        const char *name;
 251        unsigned int pack_id;
 252        unsigned char sha1[20];
 253};
 254
 255struct hash_list
 256{
 257        struct hash_list *next;
 258        unsigned char sha1[20];
 259};
 260
 261typedef enum {
 262        WHENSPEC_RAW = 1,
 263        WHENSPEC_RFC2822,
 264        WHENSPEC_NOW,
 265} whenspec_type;
 266
 267struct recent_command
 268{
 269        struct recent_command *prev;
 270        struct recent_command *next;
 271        char *buf;
 272};
 273
 274/* Configured limits on output */
 275static unsigned long max_depth = 10;
 276static off_t max_packsize = (1LL << 32) - 1;
 277static int force_update;
 278
 279/* Stats and misc. counters */
 280static uintmax_t alloc_count;
 281static uintmax_t marks_set_count;
 282static uintmax_t object_count_by_type[1 << TYPE_BITS];
 283static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 284static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 285static unsigned long object_count;
 286static unsigned long branch_count;
 287static unsigned long branch_load_count;
 288static int failure;
 289static FILE *pack_edges;
 290
 291/* Memory pools */
 292static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 293static size_t total_allocd;
 294static struct mem_pool *mem_pool;
 295
 296/* Atom management */
 297static unsigned int atom_table_sz = 4451;
 298static unsigned int atom_cnt;
 299static struct atom_str **atom_table;
 300
 301/* The .pack file being generated */
 302static unsigned int pack_id;
 303static struct packed_git *pack_data;
 304static struct packed_git **all_packs;
 305static unsigned long pack_size;
 306
 307/* Table of objects we've written. */
 308static unsigned int object_entry_alloc = 5000;
 309static struct object_entry_pool *blocks;
 310static struct object_entry *object_table[1 << 16];
 311static struct mark_set *marks;
 312static const char* mark_file;
 313
 314/* Our last blob */
 315static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 316
 317/* Tree management */
 318static unsigned int tree_entry_alloc = 1000;
 319static void *avail_tree_entry;
 320static unsigned int avail_tree_table_sz = 100;
 321static struct avail_tree_content **avail_tree_table;
 322static struct strbuf old_tree = STRBUF_INIT;
 323static struct strbuf new_tree = STRBUF_INIT;
 324
 325/* Branch data */
 326static unsigned long max_active_branches = 5;
 327static unsigned long cur_active_branches;
 328static unsigned long branch_table_sz = 1039;
 329static struct branch **branch_table;
 330static struct branch *active_branches;
 331
 332/* Tag data */
 333static struct tag *first_tag;
 334static struct tag *last_tag;
 335
 336/* Input stream parsing */
 337static whenspec_type whenspec = WHENSPEC_RAW;
 338static struct strbuf command_buf = STRBUF_INIT;
 339static int unread_command_buf;
 340static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 341static struct recent_command *cmd_tail = &cmd_hist;
 342static struct recent_command *rc_free;
 343static unsigned int cmd_save = 100;
 344static uintmax_t next_mark;
 345static struct strbuf new_data = STRBUF_INIT;
 346
 347static void write_branch_report(FILE *rpt, struct branch *b)
 348{
 349        fprintf(rpt, "%s:\n", b->name);
 350
 351        fprintf(rpt, "  status      :");
 352        if (b->active)
 353                fputs(" active", rpt);
 354        if (b->branch_tree.tree)
 355                fputs(" loaded", rpt);
 356        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 357                fputs(" dirty", rpt);
 358        fputc('\n', rpt);
 359
 360        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 361        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 362        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 363        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 364
 365        fputs("  last pack   : ", rpt);
 366        if (b->pack_id < MAX_PACK_ID)
 367                fprintf(rpt, "%u", b->pack_id);
 368        fputc('\n', rpt);
 369
 370        fputc('\n', rpt);
 371}
 372
 373static void write_crash_report(const char *err)
 374{
 375        char *loc = git_path("fast_import_crash_%d", getpid());
 376        FILE *rpt = fopen(loc, "w");
 377        struct branch *b;
 378        unsigned long lu;
 379        struct recent_command *rc;
 380
 381        if (!rpt) {
 382                error("can't write crash report %s: %s", loc, strerror(errno));
 383                return;
 384        }
 385
 386        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 387
 388        fprintf(rpt, "fast-import crash report:\n");
 389        fprintf(rpt, "    fast-import process: %d\n", getpid());
 390        fprintf(rpt, "    parent process     : %d\n", getppid());
 391        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 392        fputc('\n', rpt);
 393
 394        fputs("fatal: ", rpt);
 395        fputs(err, rpt);
 396        fputc('\n', rpt);
 397
 398        fputc('\n', rpt);
 399        fputs("Most Recent Commands Before Crash\n", rpt);
 400        fputs("---------------------------------\n", rpt);
 401        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 402                if (rc->next == &cmd_hist)
 403                        fputs("* ", rpt);
 404                else
 405                        fputs("  ", rpt);
 406                fputs(rc->buf, rpt);
 407                fputc('\n', rpt);
 408        }
 409
 410        fputc('\n', rpt);
 411        fputs("Active Branch LRU\n", rpt);
 412        fputs("-----------------\n", rpt);
 413        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 414                cur_active_branches,
 415                max_active_branches);
 416        fputc('\n', rpt);
 417        fputs("  pos  clock name\n", rpt);
 418        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 419        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 420                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 421                        ++lu, b->last_commit, b->name);
 422
 423        fputc('\n', rpt);
 424        fputs("Inactive Branches\n", rpt);
 425        fputs("-----------------\n", rpt);
 426        for (lu = 0; lu < branch_table_sz; lu++) {
 427                for (b = branch_table[lu]; b; b = b->table_next_branch)
 428                        write_branch_report(rpt, b);
 429        }
 430
 431        fputc('\n', rpt);
 432        fputs("-------------------\n", rpt);
 433        fputs("END OF CRASH REPORT\n", rpt);
 434        fclose(rpt);
 435}
 436
 437static NORETURN void die_nicely(const char *err, va_list params)
 438{
 439        static int zombie;
 440        char message[2 * PATH_MAX];
 441
 442        vsnprintf(message, sizeof(message), err, params);
 443        fputs("fatal: ", stderr);
 444        fputs(message, stderr);
 445        fputc('\n', stderr);
 446
 447        if (!zombie) {
 448                zombie = 1;
 449                write_crash_report(message);
 450        }
 451        exit(128);
 452}
 453
 454static void alloc_objects(unsigned int cnt)
 455{
 456        struct object_entry_pool *b;
 457
 458        b = xmalloc(sizeof(struct object_entry_pool)
 459                + cnt * sizeof(struct object_entry));
 460        b->next_pool = blocks;
 461        b->next_free = b->entries;
 462        b->end = b->entries + cnt;
 463        blocks = b;
 464        alloc_count += cnt;
 465}
 466
 467static struct object_entry *new_object(unsigned char *sha1)
 468{
 469        struct object_entry *e;
 470
 471        if (blocks->next_free == blocks->end)
 472                alloc_objects(object_entry_alloc);
 473
 474        e = blocks->next_free++;
 475        hashcpy(e->sha1, sha1);
 476        return e;
 477}
 478
 479static struct object_entry *find_object(unsigned char *sha1)
 480{
 481        unsigned int h = sha1[0] << 8 | sha1[1];
 482        struct object_entry *e;
 483        for (e = object_table[h]; e; e = e->next)
 484                if (!hashcmp(sha1, e->sha1))
 485                        return e;
 486        return NULL;
 487}
 488
 489static struct object_entry *insert_object(unsigned char *sha1)
 490{
 491        unsigned int h = sha1[0] << 8 | sha1[1];
 492        struct object_entry *e = object_table[h];
 493        struct object_entry *p = NULL;
 494
 495        while (e) {
 496                if (!hashcmp(sha1, e->sha1))
 497                        return e;
 498                p = e;
 499                e = e->next;
 500        }
 501
 502        e = new_object(sha1);
 503        e->next = NULL;
 504        e->offset = 0;
 505        if (p)
 506                p->next = e;
 507        else
 508                object_table[h] = e;
 509        return e;
 510}
 511
 512static unsigned int hc_str(const char *s, size_t len)
 513{
 514        unsigned int r = 0;
 515        while (len-- > 0)
 516                r = r * 31 + *s++;
 517        return r;
 518}
 519
 520static void *pool_alloc(size_t len)
 521{
 522        struct mem_pool *p;
 523        void *r;
 524
 525        for (p = mem_pool; p; p = p->next_pool)
 526                if ((p->end - p->next_free >= len))
 527                        break;
 528
 529        if (!p) {
 530                if (len >= (mem_pool_alloc/2)) {
 531                        total_allocd += len;
 532                        return xmalloc(len);
 533                }
 534                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 535                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 536                p->next_pool = mem_pool;
 537                p->next_free = (char *) p->space;
 538                p->end = p->next_free + mem_pool_alloc;
 539                mem_pool = p;
 540        }
 541
 542        r = p->next_free;
 543        /* round out to a 'uintmax_t' alignment */
 544        if (len & (sizeof(uintmax_t) - 1))
 545                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 546        p->next_free += len;
 547        return r;
 548}
 549
 550static void *pool_calloc(size_t count, size_t size)
 551{
 552        size_t len = count * size;
 553        void *r = pool_alloc(len);
 554        memset(r, 0, len);
 555        return r;
 556}
 557
 558static char *pool_strdup(const char *s)
 559{
 560        char *r = pool_alloc(strlen(s) + 1);
 561        strcpy(r, s);
 562        return r;
 563}
 564
 565static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 566{
 567        struct mark_set *s = marks;
 568        while ((idnum >> s->shift) >= 1024) {
 569                s = pool_calloc(1, sizeof(struct mark_set));
 570                s->shift = marks->shift + 10;
 571                s->data.sets[0] = marks;
 572                marks = s;
 573        }
 574        while (s->shift) {
 575                uintmax_t i = idnum >> s->shift;
 576                idnum -= i << s->shift;
 577                if (!s->data.sets[i]) {
 578                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 579                        s->data.sets[i]->shift = s->shift - 10;
 580                }
 581                s = s->data.sets[i];
 582        }
 583        if (!s->data.marked[idnum])
 584                marks_set_count++;
 585        s->data.marked[idnum] = oe;
 586}
 587
 588static struct object_entry *find_mark(uintmax_t idnum)
 589{
 590        uintmax_t orig_idnum = idnum;
 591        struct mark_set *s = marks;
 592        struct object_entry *oe = NULL;
 593        if ((idnum >> s->shift) < 1024) {
 594                while (s && s->shift) {
 595                        uintmax_t i = idnum >> s->shift;
 596                        idnum -= i << s->shift;
 597                        s = s->data.sets[i];
 598                }
 599                if (s)
 600                        oe = s->data.marked[idnum];
 601        }
 602        if (!oe)
 603                die("mark :%" PRIuMAX " not declared", orig_idnum);
 604        return oe;
 605}
 606
 607static struct atom_str *to_atom(const char *s, unsigned short len)
 608{
 609        unsigned int hc = hc_str(s, len) % atom_table_sz;
 610        struct atom_str *c;
 611
 612        for (c = atom_table[hc]; c; c = c->next_atom)
 613                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 614                        return c;
 615
 616        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 617        c->str_len = len;
 618        strncpy(c->str_dat, s, len);
 619        c->str_dat[len] = 0;
 620        c->next_atom = atom_table[hc];
 621        atom_table[hc] = c;
 622        atom_cnt++;
 623        return c;
 624}
 625
 626static struct branch *lookup_branch(const char *name)
 627{
 628        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 629        struct branch *b;
 630
 631        for (b = branch_table[hc]; b; b = b->table_next_branch)
 632                if (!strcmp(name, b->name))
 633                        return b;
 634        return NULL;
 635}
 636
 637static struct branch *new_branch(const char *name)
 638{
 639        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 640        struct branch* b = lookup_branch(name);
 641
 642        if (b)
 643                die("Invalid attempt to create duplicate branch: %s", name);
 644        switch (check_ref_format(name)) {
 645        case 0: break; /* its valid */
 646        case CHECK_REF_FORMAT_ONELEVEL:
 647                break; /* valid, but too few '/', allow anyway */
 648        default:
 649                die("Branch name doesn't conform to GIT standards: %s", name);
 650        }
 651
 652        b = pool_calloc(1, sizeof(struct branch));
 653        b->name = pool_strdup(name);
 654        b->table_next_branch = branch_table[hc];
 655        b->branch_tree.versions[0].mode = S_IFDIR;
 656        b->branch_tree.versions[1].mode = S_IFDIR;
 657        b->active = 0;
 658        b->pack_id = MAX_PACK_ID;
 659        branch_table[hc] = b;
 660        branch_count++;
 661        return b;
 662}
 663
 664static unsigned int hc_entries(unsigned int cnt)
 665{
 666        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 667        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 668}
 669
 670static struct tree_content *new_tree_content(unsigned int cnt)
 671{
 672        struct avail_tree_content *f, *l = NULL;
 673        struct tree_content *t;
 674        unsigned int hc = hc_entries(cnt);
 675
 676        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 677                if (f->entry_capacity >= cnt)
 678                        break;
 679
 680        if (f) {
 681                if (l)
 682                        l->next_avail = f->next_avail;
 683                else
 684                        avail_tree_table[hc] = f->next_avail;
 685        } else {
 686                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 687                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 688                f->entry_capacity = cnt;
 689        }
 690
 691        t = (struct tree_content*)f;
 692        t->entry_count = 0;
 693        t->delta_depth = 0;
 694        return t;
 695}
 696
 697static void release_tree_entry(struct tree_entry *e);
 698static void release_tree_content(struct tree_content *t)
 699{
 700        struct avail_tree_content *f = (struct avail_tree_content*)t;
 701        unsigned int hc = hc_entries(f->entry_capacity);
 702        f->next_avail = avail_tree_table[hc];
 703        avail_tree_table[hc] = f;
 704}
 705
 706static void release_tree_content_recursive(struct tree_content *t)
 707{
 708        unsigned int i;
 709        for (i = 0; i < t->entry_count; i++)
 710                release_tree_entry(t->entries[i]);
 711        release_tree_content(t);
 712}
 713
 714static struct tree_content *grow_tree_content(
 715        struct tree_content *t,
 716        int amt)
 717{
 718        struct tree_content *r = new_tree_content(t->entry_count + amt);
 719        r->entry_count = t->entry_count;
 720        r->delta_depth = t->delta_depth;
 721        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 722        release_tree_content(t);
 723        return r;
 724}
 725
 726static struct tree_entry *new_tree_entry(void)
 727{
 728        struct tree_entry *e;
 729
 730        if (!avail_tree_entry) {
 731                unsigned int n = tree_entry_alloc;
 732                total_allocd += n * sizeof(struct tree_entry);
 733                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 734                while (n-- > 1) {
 735                        *((void**)e) = e + 1;
 736                        e++;
 737                }
 738                *((void**)e) = NULL;
 739        }
 740
 741        e = avail_tree_entry;
 742        avail_tree_entry = *((void**)e);
 743        return e;
 744}
 745
 746static void release_tree_entry(struct tree_entry *e)
 747{
 748        if (e->tree)
 749                release_tree_content_recursive(e->tree);
 750        *((void**)e) = avail_tree_entry;
 751        avail_tree_entry = e;
 752}
 753
 754static struct tree_content *dup_tree_content(struct tree_content *s)
 755{
 756        struct tree_content *d;
 757        struct tree_entry *a, *b;
 758        unsigned int i;
 759
 760        if (!s)
 761                return NULL;
 762        d = new_tree_content(s->entry_count);
 763        for (i = 0; i < s->entry_count; i++) {
 764                a = s->entries[i];
 765                b = new_tree_entry();
 766                memcpy(b, a, sizeof(*a));
 767                if (a->tree && is_null_sha1(b->versions[1].sha1))
 768                        b->tree = dup_tree_content(a->tree);
 769                else
 770                        b->tree = NULL;
 771                d->entries[i] = b;
 772        }
 773        d->entry_count = s->entry_count;
 774        d->delta_depth = s->delta_depth;
 775
 776        return d;
 777}
 778
 779static void start_packfile(void)
 780{
 781        static char tmpfile[PATH_MAX];
 782        struct packed_git *p;
 783        struct pack_header hdr;
 784        int pack_fd;
 785
 786        snprintf(tmpfile, sizeof(tmpfile),
 787                "%s/tmp_pack_XXXXXX", get_object_directory());
 788        pack_fd = xmkstemp(tmpfile);
 789        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 790        strcpy(p->pack_name, tmpfile);
 791        p->pack_fd = pack_fd;
 792
 793        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 794        hdr.hdr_version = htonl(2);
 795        hdr.hdr_entries = 0;
 796        write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 797
 798        pack_data = p;
 799        pack_size = sizeof(hdr);
 800        object_count = 0;
 801
 802        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 803        all_packs[pack_id] = p;
 804}
 805
 806static int oecmp (const void *a_, const void *b_)
 807{
 808        struct object_entry *a = *((struct object_entry**)a_);
 809        struct object_entry *b = *((struct object_entry**)b_);
 810        return hashcmp(a->sha1, b->sha1);
 811}
 812
 813static char *create_index(void)
 814{
 815        static char tmpfile[PATH_MAX];
 816        SHA_CTX ctx;
 817        struct sha1file *f;
 818        struct object_entry **idx, **c, **last, *e;
 819        struct object_entry_pool *o;
 820        uint32_t array[256];
 821        int i, idx_fd;
 822
 823        /* Build the sorted table of object IDs. */
 824        idx = xmalloc(object_count * sizeof(struct object_entry*));
 825        c = idx;
 826        for (o = blocks; o; o = o->next_pool)
 827                for (e = o->next_free; e-- != o->entries;)
 828                        if (pack_id == e->pack_id)
 829                                *c++ = e;
 830        last = idx + object_count;
 831        if (c != last)
 832                die("internal consistency error creating the index");
 833        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 834
 835        /* Generate the fan-out array. */
 836        c = idx;
 837        for (i = 0; i < 256; i++) {
 838                struct object_entry **next = c;;
 839                while (next < last) {
 840                        if ((*next)->sha1[0] != i)
 841                                break;
 842                        next++;
 843                }
 844                array[i] = htonl(next - idx);
 845                c = next;
 846        }
 847
 848        snprintf(tmpfile, sizeof(tmpfile),
 849                "%s/tmp_idx_XXXXXX", get_object_directory());
 850        idx_fd = xmkstemp(tmpfile);
 851        f = sha1fd(idx_fd, tmpfile);
 852        sha1write(f, array, 256 * sizeof(int));
 853        SHA1_Init(&ctx);
 854        for (c = idx; c != last; c++) {
 855                uint32_t offset = htonl((*c)->offset);
 856                sha1write(f, &offset, 4);
 857                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 858                SHA1_Update(&ctx, (*c)->sha1, 20);
 859        }
 860        sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
 861        sha1close(f, NULL, 1);
 862        free(idx);
 863        SHA1_Final(pack_data->sha1, &ctx);
 864        return tmpfile;
 865}
 866
 867static char *keep_pack(char *curr_index_name)
 868{
 869        static char name[PATH_MAX];
 870        static const char *keep_msg = "fast-import";
 871        int keep_fd;
 872
 873        chmod(pack_data->pack_name, 0444);
 874        chmod(curr_index_name, 0444);
 875
 876        snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 877                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 878        keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
 879        if (keep_fd < 0)
 880                die("cannot create keep file");
 881        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 882        if (close(keep_fd))
 883                die("failed to write keep file");
 884
 885        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 886                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 887        if (move_temp_to_file(pack_data->pack_name, name))
 888                die("cannot store pack file");
 889
 890        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 891                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 892        if (move_temp_to_file(curr_index_name, name))
 893                die("cannot store index file");
 894        return name;
 895}
 896
 897static void unkeep_all_packs(void)
 898{
 899        static char name[PATH_MAX];
 900        int k;
 901
 902        for (k = 0; k < pack_id; k++) {
 903                struct packed_git *p = all_packs[k];
 904                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 905                         get_object_directory(), sha1_to_hex(p->sha1));
 906                unlink(name);
 907        }
 908}
 909
 910static void end_packfile(void)
 911{
 912        struct packed_git *old_p = pack_data, *new_p;
 913
 914        if (object_count) {
 915                char *idx_name;
 916                int i;
 917                struct branch *b;
 918                struct tag *t;
 919
 920                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 921                                    pack_data->pack_name, object_count);
 922                close(pack_data->pack_fd);
 923                idx_name = keep_pack(create_index());
 924
 925                /* Register the packfile with core git's machinary. */
 926                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 927                if (!new_p)
 928                        die("core git rejected index %s", idx_name);
 929                new_p->windows = old_p->windows;
 930                all_packs[pack_id] = new_p;
 931                install_packed_git(new_p);
 932
 933                /* Print the boundary */
 934                if (pack_edges) {
 935                        fprintf(pack_edges, "%s:", new_p->pack_name);
 936                        for (i = 0; i < branch_table_sz; i++) {
 937                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 938                                        if (b->pack_id == pack_id)
 939                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 940                                }
 941                        }
 942                        for (t = first_tag; t; t = t->next_tag) {
 943                                if (t->pack_id == pack_id)
 944                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 945                        }
 946                        fputc('\n', pack_edges);
 947                        fflush(pack_edges);
 948                }
 949
 950                pack_id++;
 951        }
 952        else
 953                unlink(old_p->pack_name);
 954        free(old_p);
 955
 956        /* We can't carry a delta across packfiles. */
 957        strbuf_release(&last_blob.data);
 958        last_blob.offset = 0;
 959        last_blob.depth = 0;
 960}
 961
 962static void cycle_packfile(void)
 963{
 964        end_packfile();
 965        start_packfile();
 966}
 967
 968static size_t encode_header(
 969        enum object_type type,
 970        size_t size,
 971        unsigned char *hdr)
 972{
 973        int n = 1;
 974        unsigned char c;
 975
 976        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 977                die("bad type %d", type);
 978
 979        c = (type << 4) | (size & 15);
 980        size >>= 4;
 981        while (size) {
 982                *hdr++ = c | 0x80;
 983                c = size & 0x7f;
 984                size >>= 7;
 985                n++;
 986        }
 987        *hdr = c;
 988        return n;
 989}
 990
 991static int store_object(
 992        enum object_type type,
 993        struct strbuf *dat,
 994        struct last_object *last,
 995        unsigned char *sha1out,
 996        uintmax_t mark)
 997{
 998        void *out, *delta;
 999        struct object_entry *e;
1000        unsigned char hdr[96];
1001        unsigned char sha1[20];
1002        unsigned long hdrlen, deltalen;
1003        SHA_CTX c;
1004        z_stream s;
1005
1006        hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1007                (unsigned long)dat->len) + 1;
1008        SHA1_Init(&c);
1009        SHA1_Update(&c, hdr, hdrlen);
1010        SHA1_Update(&c, dat->buf, dat->len);
1011        SHA1_Final(sha1, &c);
1012        if (sha1out)
1013                hashcpy(sha1out, sha1);
1014
1015        e = insert_object(sha1);
1016        if (mark)
1017                insert_mark(mark, e);
1018        if (e->offset) {
1019                duplicate_count_by_type[type]++;
1020                return 1;
1021        } else if (find_sha1_pack(sha1, packed_git)) {
1022                e->type = type;
1023                e->pack_id = MAX_PACK_ID;
1024                e->offset = 1; /* just not zero! */
1025                duplicate_count_by_type[type]++;
1026                return 1;
1027        }
1028
1029        if (last && last->data.buf && last->depth < max_depth) {
1030                delta = diff_delta(last->data.buf, last->data.len,
1031                        dat->buf, dat->len,
1032                        &deltalen, 0);
1033                if (delta && deltalen >= dat->len) {
1034                        free(delta);
1035                        delta = NULL;
1036                }
1037        } else
1038                delta = NULL;
1039
1040        memset(&s, 0, sizeof(s));
1041        deflateInit(&s, zlib_compression_level);
1042        if (delta) {
1043                s.next_in = delta;
1044                s.avail_in = deltalen;
1045        } else {
1046                s.next_in = (void *)dat->buf;
1047                s.avail_in = dat->len;
1048        }
1049        s.avail_out = deflateBound(&s, s.avail_in);
1050        s.next_out = out = xmalloc(s.avail_out);
1051        while (deflate(&s, Z_FINISH) == Z_OK)
1052                /* nothing */;
1053        deflateEnd(&s);
1054
1055        /* Determine if we should auto-checkpoint. */
1056        if ((pack_size + 60 + s.total_out) > max_packsize
1057                || (pack_size + 60 + s.total_out) < pack_size) {
1058
1059                /* This new object needs to *not* have the current pack_id. */
1060                e->pack_id = pack_id + 1;
1061                cycle_packfile();
1062
1063                /* We cannot carry a delta into the new pack. */
1064                if (delta) {
1065                        free(delta);
1066                        delta = NULL;
1067
1068                        memset(&s, 0, sizeof(s));
1069                        deflateInit(&s, zlib_compression_level);
1070                        s.next_in = (void *)dat->buf;
1071                        s.avail_in = dat->len;
1072                        s.avail_out = deflateBound(&s, s.avail_in);
1073                        s.next_out = out = xrealloc(out, s.avail_out);
1074                        while (deflate(&s, Z_FINISH) == Z_OK)
1075                                /* nothing */;
1076                        deflateEnd(&s);
1077                }
1078        }
1079
1080        e->type = type;
1081        e->pack_id = pack_id;
1082        e->offset = pack_size;
1083        object_count++;
1084        object_count_by_type[type]++;
1085
1086        if (delta) {
1087                unsigned long ofs = e->offset - last->offset;
1088                unsigned pos = sizeof(hdr) - 1;
1089
1090                delta_count_by_type[type]++;
1091                e->depth = last->depth + 1;
1092
1093                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1094                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1095                pack_size += hdrlen;
1096
1097                hdr[pos] = ofs & 127;
1098                while (ofs >>= 7)
1099                        hdr[--pos] = 128 | (--ofs & 127);
1100                write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1101                pack_size += sizeof(hdr) - pos;
1102        } else {
1103                e->depth = 0;
1104                hdrlen = encode_header(type, dat->len, hdr);
1105                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1106                pack_size += hdrlen;
1107        }
1108
1109        write_or_die(pack_data->pack_fd, out, s.total_out);
1110        pack_size += s.total_out;
1111
1112        free(out);
1113        free(delta);
1114        if (last) {
1115                if (last->no_swap) {
1116                        last->data = *dat;
1117                } else {
1118                        strbuf_swap(&last->data, dat);
1119                }
1120                last->offset = e->offset;
1121                last->depth = e->depth;
1122        }
1123        return 0;
1124}
1125
1126static void *gfi_unpack_entry(
1127        struct object_entry *oe,
1128        unsigned long *sizep)
1129{
1130        enum object_type type;
1131        struct packed_git *p = all_packs[oe->pack_id];
1132        if (p == pack_data)
1133                p->pack_size = pack_size + 20;
1134        return unpack_entry(p, oe->offset, &type, sizep);
1135}
1136
1137static const char *get_mode(const char *str, uint16_t *modep)
1138{
1139        unsigned char c;
1140        uint16_t mode = 0;
1141
1142        while ((c = *str++) != ' ') {
1143                if (c < '0' || c > '7')
1144                        return NULL;
1145                mode = (mode << 3) + (c - '0');
1146        }
1147        *modep = mode;
1148        return str;
1149}
1150
1151static void load_tree(struct tree_entry *root)
1152{
1153        unsigned char* sha1 = root->versions[1].sha1;
1154        struct object_entry *myoe;
1155        struct tree_content *t;
1156        unsigned long size;
1157        char *buf;
1158        const char *c;
1159
1160        root->tree = t = new_tree_content(8);
1161        if (is_null_sha1(sha1))
1162                return;
1163
1164        myoe = find_object(sha1);
1165        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1166                if (myoe->type != OBJ_TREE)
1167                        die("Not a tree: %s", sha1_to_hex(sha1));
1168                t->delta_depth = myoe->depth;
1169                buf = gfi_unpack_entry(myoe, &size);
1170        } else {
1171                enum object_type type;
1172                buf = read_sha1_file(sha1, &type, &size);
1173                if (!buf || type != OBJ_TREE)
1174                        die("Can't load tree %s", sha1_to_hex(sha1));
1175        }
1176
1177        c = buf;
1178        while (c != (buf + size)) {
1179                struct tree_entry *e = new_tree_entry();
1180
1181                if (t->entry_count == t->entry_capacity)
1182                        root->tree = t = grow_tree_content(t, t->entry_count);
1183                t->entries[t->entry_count++] = e;
1184
1185                e->tree = NULL;
1186                c = get_mode(c, &e->versions[1].mode);
1187                if (!c)
1188                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1189                e->versions[0].mode = e->versions[1].mode;
1190                e->name = to_atom(c, strlen(c));
1191                c += e->name->str_len + 1;
1192                hashcpy(e->versions[0].sha1, (unsigned char*)c);
1193                hashcpy(e->versions[1].sha1, (unsigned char*)c);
1194                c += 20;
1195        }
1196        free(buf);
1197}
1198
1199static int tecmp0 (const void *_a, const void *_b)
1200{
1201        struct tree_entry *a = *((struct tree_entry**)_a);
1202        struct tree_entry *b = *((struct tree_entry**)_b);
1203        return base_name_compare(
1204                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1205                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1206}
1207
1208static int tecmp1 (const void *_a, const void *_b)
1209{
1210        struct tree_entry *a = *((struct tree_entry**)_a);
1211        struct tree_entry *b = *((struct tree_entry**)_b);
1212        return base_name_compare(
1213                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1214                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1215}
1216
1217static void mktree(struct tree_content *t, int v, struct strbuf *b)
1218{
1219        size_t maxlen = 0;
1220        unsigned int i;
1221
1222        if (!v)
1223                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1224        else
1225                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1226
1227        for (i = 0; i < t->entry_count; i++) {
1228                if (t->entries[i]->versions[v].mode)
1229                        maxlen += t->entries[i]->name->str_len + 34;
1230        }
1231
1232        strbuf_reset(b);
1233        strbuf_grow(b, maxlen);
1234        for (i = 0; i < t->entry_count; i++) {
1235                struct tree_entry *e = t->entries[i];
1236                if (!e->versions[v].mode)
1237                        continue;
1238                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1239                                        e->name->str_dat, '\0');
1240                strbuf_add(b, e->versions[v].sha1, 20);
1241        }
1242}
1243
1244static void store_tree(struct tree_entry *root)
1245{
1246        struct tree_content *t = root->tree;
1247        unsigned int i, j, del;
1248        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1249        struct object_entry *le;
1250
1251        if (!is_null_sha1(root->versions[1].sha1))
1252                return;
1253
1254        for (i = 0; i < t->entry_count; i++) {
1255                if (t->entries[i]->tree)
1256                        store_tree(t->entries[i]);
1257        }
1258
1259        le = find_object(root->versions[0].sha1);
1260        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1261                mktree(t, 0, &old_tree);
1262                lo.data = old_tree;
1263                lo.offset = le->offset;
1264                lo.depth = t->delta_depth;
1265        }
1266
1267        mktree(t, 1, &new_tree);
1268        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1269
1270        t->delta_depth = lo.depth;
1271        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1272                struct tree_entry *e = t->entries[i];
1273                if (e->versions[1].mode) {
1274                        e->versions[0].mode = e->versions[1].mode;
1275                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1276                        t->entries[j++] = e;
1277                } else {
1278                        release_tree_entry(e);
1279                        del++;
1280                }
1281        }
1282        t->entry_count -= del;
1283}
1284
1285static int tree_content_set(
1286        struct tree_entry *root,
1287        const char *p,
1288        const unsigned char *sha1,
1289        const uint16_t mode,
1290        struct tree_content *subtree)
1291{
1292        struct tree_content *t = root->tree;
1293        const char *slash1;
1294        unsigned int i, n;
1295        struct tree_entry *e;
1296
1297        slash1 = strchr(p, '/');
1298        if (slash1)
1299                n = slash1 - p;
1300        else
1301                n = strlen(p);
1302        if (!n)
1303                die("Empty path component found in input");
1304        if (!slash1 && !S_ISDIR(mode) && subtree)
1305                die("Non-directories cannot have subtrees");
1306
1307        for (i = 0; i < t->entry_count; i++) {
1308                e = t->entries[i];
1309                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1310                        if (!slash1) {
1311                                if (!S_ISDIR(mode)
1312                                                && e->versions[1].mode == mode
1313                                                && !hashcmp(e->versions[1].sha1, sha1))
1314                                        return 0;
1315                                e->versions[1].mode = mode;
1316                                hashcpy(e->versions[1].sha1, sha1);
1317                                if (e->tree)
1318                                        release_tree_content_recursive(e->tree);
1319                                e->tree = subtree;
1320                                hashclr(root->versions[1].sha1);
1321                                return 1;
1322                        }
1323                        if (!S_ISDIR(e->versions[1].mode)) {
1324                                e->tree = new_tree_content(8);
1325                                e->versions[1].mode = S_IFDIR;
1326                        }
1327                        if (!e->tree)
1328                                load_tree(e);
1329                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1330                                hashclr(root->versions[1].sha1);
1331                                return 1;
1332                        }
1333                        return 0;
1334                }
1335        }
1336
1337        if (t->entry_count == t->entry_capacity)
1338                root->tree = t = grow_tree_content(t, t->entry_count);
1339        e = new_tree_entry();
1340        e->name = to_atom(p, n);
1341        e->versions[0].mode = 0;
1342        hashclr(e->versions[0].sha1);
1343        t->entries[t->entry_count++] = e;
1344        if (slash1) {
1345                e->tree = new_tree_content(8);
1346                e->versions[1].mode = S_IFDIR;
1347                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1348        } else {
1349                e->tree = subtree;
1350                e->versions[1].mode = mode;
1351                hashcpy(e->versions[1].sha1, sha1);
1352        }
1353        hashclr(root->versions[1].sha1);
1354        return 1;
1355}
1356
1357static int tree_content_remove(
1358        struct tree_entry *root,
1359        const char *p,
1360        struct tree_entry *backup_leaf)
1361{
1362        struct tree_content *t = root->tree;
1363        const char *slash1;
1364        unsigned int i, n;
1365        struct tree_entry *e;
1366
1367        slash1 = strchr(p, '/');
1368        if (slash1)
1369                n = slash1 - p;
1370        else
1371                n = strlen(p);
1372
1373        for (i = 0; i < t->entry_count; i++) {
1374                e = t->entries[i];
1375                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1376                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1377                                goto del_entry;
1378                        if (!e->tree)
1379                                load_tree(e);
1380                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1381                                for (n = 0; n < e->tree->entry_count; n++) {
1382                                        if (e->tree->entries[n]->versions[1].mode) {
1383                                                hashclr(root->versions[1].sha1);
1384                                                return 1;
1385                                        }
1386                                }
1387                                backup_leaf = NULL;
1388                                goto del_entry;
1389                        }
1390                        return 0;
1391                }
1392        }
1393        return 0;
1394
1395del_entry:
1396        if (backup_leaf)
1397                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1398        else if (e->tree)
1399                release_tree_content_recursive(e->tree);
1400        e->tree = NULL;
1401        e->versions[1].mode = 0;
1402        hashclr(e->versions[1].sha1);
1403        hashclr(root->versions[1].sha1);
1404        return 1;
1405}
1406
1407static int tree_content_get(
1408        struct tree_entry *root,
1409        const char *p,
1410        struct tree_entry *leaf)
1411{
1412        struct tree_content *t = root->tree;
1413        const char *slash1;
1414        unsigned int i, n;
1415        struct tree_entry *e;
1416
1417        slash1 = strchr(p, '/');
1418        if (slash1)
1419                n = slash1 - p;
1420        else
1421                n = strlen(p);
1422
1423        for (i = 0; i < t->entry_count; i++) {
1424                e = t->entries[i];
1425                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1426                        if (!slash1) {
1427                                memcpy(leaf, e, sizeof(*leaf));
1428                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1429                                        leaf->tree = dup_tree_content(e->tree);
1430                                else
1431                                        leaf->tree = NULL;
1432                                return 1;
1433                        }
1434                        if (!S_ISDIR(e->versions[1].mode))
1435                                return 0;
1436                        if (!e->tree)
1437                                load_tree(e);
1438                        return tree_content_get(e, slash1 + 1, leaf);
1439                }
1440        }
1441        return 0;
1442}
1443
1444static int update_branch(struct branch *b)
1445{
1446        static const char *msg = "fast-import";
1447        struct ref_lock *lock;
1448        unsigned char old_sha1[20];
1449
1450        if (read_ref(b->name, old_sha1))
1451                hashclr(old_sha1);
1452        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1453        if (!lock)
1454                return error("Unable to lock %s", b->name);
1455        if (!force_update && !is_null_sha1(old_sha1)) {
1456                struct commit *old_cmit, *new_cmit;
1457
1458                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1459                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1460                if (!old_cmit || !new_cmit) {
1461                        unlock_ref(lock);
1462                        return error("Branch %s is missing commits.", b->name);
1463                }
1464
1465                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1466                        unlock_ref(lock);
1467                        warning("Not updating %s"
1468                                " (new tip %s does not contain %s)",
1469                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1470                        return -1;
1471                }
1472        }
1473        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1474                return error("Unable to update %s", b->name);
1475        return 0;
1476}
1477
1478static void dump_branches(void)
1479{
1480        unsigned int i;
1481        struct branch *b;
1482
1483        for (i = 0; i < branch_table_sz; i++) {
1484                for (b = branch_table[i]; b; b = b->table_next_branch)
1485                        failure |= update_branch(b);
1486        }
1487}
1488
1489static void dump_tags(void)
1490{
1491        static const char *msg = "fast-import";
1492        struct tag *t;
1493        struct ref_lock *lock;
1494        char ref_name[PATH_MAX];
1495
1496        for (t = first_tag; t; t = t->next_tag) {
1497                sprintf(ref_name, "tags/%s", t->name);
1498                lock = lock_ref_sha1(ref_name, NULL);
1499                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1500                        failure |= error("Unable to update %s", ref_name);
1501        }
1502}
1503
1504static void dump_marks_helper(FILE *f,
1505        uintmax_t base,
1506        struct mark_set *m)
1507{
1508        uintmax_t k;
1509        if (m->shift) {
1510                for (k = 0; k < 1024; k++) {
1511                        if (m->data.sets[k])
1512                                dump_marks_helper(f, (base + k) << m->shift,
1513                                        m->data.sets[k]);
1514                }
1515        } else {
1516                for (k = 0; k < 1024; k++) {
1517                        if (m->data.marked[k])
1518                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1519                                        sha1_to_hex(m->data.marked[k]->sha1));
1520                }
1521        }
1522}
1523
1524static void dump_marks(void)
1525{
1526        static struct lock_file mark_lock;
1527        int mark_fd;
1528        FILE *f;
1529
1530        if (!mark_file)
1531                return;
1532
1533        mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1534        if (mark_fd < 0) {
1535                failure |= error("Unable to write marks file %s: %s",
1536                        mark_file, strerror(errno));
1537                return;
1538        }
1539
1540        f = fdopen(mark_fd, "w");
1541        if (!f) {
1542                rollback_lock_file(&mark_lock);
1543                failure |= error("Unable to write marks file %s: %s",
1544                        mark_file, strerror(errno));
1545                return;
1546        }
1547
1548        dump_marks_helper(f, 0, marks);
1549        if (ferror(f) || fclose(f))
1550                failure |= error("Unable to write marks file %s: %s",
1551                        mark_file, strerror(errno));
1552        /*
1553         * Since the lock file was fdopen()'ed and then fclose()'ed above,
1554         * assign -1 to the lock file descriptor so that commit_lock_file()
1555         * won't try to close() it.
1556         */
1557        mark_lock.fd = -1;
1558        if (commit_lock_file(&mark_lock))
1559                failure |= error("Unable to write commit file %s: %s",
1560                        mark_file, strerror(errno));
1561}
1562
1563static int read_next_command(void)
1564{
1565        static int stdin_eof = 0;
1566
1567        if (stdin_eof) {
1568                unread_command_buf = 0;
1569                return EOF;
1570        }
1571
1572        do {
1573                if (unread_command_buf) {
1574                        unread_command_buf = 0;
1575                } else {
1576                        struct recent_command *rc;
1577
1578                        strbuf_detach(&command_buf, NULL);
1579                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1580                        if (stdin_eof)
1581                                return EOF;
1582
1583                        rc = rc_free;
1584                        if (rc)
1585                                rc_free = rc->next;
1586                        else {
1587                                rc = cmd_hist.next;
1588                                cmd_hist.next = rc->next;
1589                                cmd_hist.next->prev = &cmd_hist;
1590                                free(rc->buf);
1591                        }
1592
1593                        rc->buf = command_buf.buf;
1594                        rc->prev = cmd_tail;
1595                        rc->next = cmd_hist.prev;
1596                        rc->prev->next = rc;
1597                        cmd_tail = rc;
1598                }
1599        } while (command_buf.buf[0] == '#');
1600
1601        return 0;
1602}
1603
1604static void skip_optional_lf(void)
1605{
1606        int term_char = fgetc(stdin);
1607        if (term_char != '\n' && term_char != EOF)
1608                ungetc(term_char, stdin);
1609}
1610
1611static void cmd_mark(void)
1612{
1613        if (!prefixcmp(command_buf.buf, "mark :")) {
1614                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1615                read_next_command();
1616        }
1617        else
1618                next_mark = 0;
1619}
1620
1621static void cmd_data(struct strbuf *sb)
1622{
1623        strbuf_reset(sb);
1624
1625        if (prefixcmp(command_buf.buf, "data "))
1626                die("Expected 'data n' command, found: %s", command_buf.buf);
1627
1628        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1629                char *term = xstrdup(command_buf.buf + 5 + 2);
1630                size_t term_len = command_buf.len - 5 - 2;
1631
1632                strbuf_detach(&command_buf, NULL);
1633                for (;;) {
1634                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1635                                die("EOF in data (terminator '%s' not found)", term);
1636                        if (term_len == command_buf.len
1637                                && !strcmp(term, command_buf.buf))
1638                                break;
1639                        strbuf_addbuf(sb, &command_buf);
1640                        strbuf_addch(sb, '\n');
1641                }
1642                free(term);
1643        }
1644        else {
1645                size_t n = 0, length;
1646
1647                length = strtoul(command_buf.buf + 5, NULL, 10);
1648
1649                while (n < length) {
1650                        size_t s = strbuf_fread(sb, length - n, stdin);
1651                        if (!s && feof(stdin))
1652                                die("EOF in data (%lu bytes remaining)",
1653                                        (unsigned long)(length - n));
1654                        n += s;
1655                }
1656        }
1657
1658        skip_optional_lf();
1659}
1660
1661static int validate_raw_date(const char *src, char *result, int maxlen)
1662{
1663        const char *orig_src = src;
1664        char *endp, sign;
1665
1666        strtoul(src, &endp, 10);
1667        if (endp == src || *endp != ' ')
1668                return -1;
1669
1670        src = endp + 1;
1671        if (*src != '-' && *src != '+')
1672                return -1;
1673        sign = *src;
1674
1675        strtoul(src + 1, &endp, 10);
1676        if (endp == src || *endp || (endp - orig_src) >= maxlen)
1677                return -1;
1678
1679        strcpy(result, orig_src);
1680        return 0;
1681}
1682
1683static char *parse_ident(const char *buf)
1684{
1685        const char *gt;
1686        size_t name_len;
1687        char *ident;
1688
1689        gt = strrchr(buf, '>');
1690        if (!gt)
1691                die("Missing > in ident string: %s", buf);
1692        gt++;
1693        if (*gt != ' ')
1694                die("Missing space after > in ident string: %s", buf);
1695        gt++;
1696        name_len = gt - buf;
1697        ident = xmalloc(name_len + 24);
1698        strncpy(ident, buf, name_len);
1699
1700        switch (whenspec) {
1701        case WHENSPEC_RAW:
1702                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1703                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1704                break;
1705        case WHENSPEC_RFC2822:
1706                if (parse_date(gt, ident + name_len, 24) < 0)
1707                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1708                break;
1709        case WHENSPEC_NOW:
1710                if (strcmp("now", gt))
1711                        die("Date in ident must be 'now': %s", buf);
1712                datestamp(ident + name_len, 24);
1713                break;
1714        }
1715
1716        return ident;
1717}
1718
1719static void cmd_new_blob(void)
1720{
1721        static struct strbuf buf = STRBUF_INIT;
1722
1723        read_next_command();
1724        cmd_mark();
1725        cmd_data(&buf);
1726        store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1727}
1728
1729static void unload_one_branch(void)
1730{
1731        while (cur_active_branches
1732                && cur_active_branches >= max_active_branches) {
1733                uintmax_t min_commit = ULONG_MAX;
1734                struct branch *e, *l = NULL, *p = NULL;
1735
1736                for (e = active_branches; e; e = e->active_next_branch) {
1737                        if (e->last_commit < min_commit) {
1738                                p = l;
1739                                min_commit = e->last_commit;
1740                        }
1741                        l = e;
1742                }
1743
1744                if (p) {
1745                        e = p->active_next_branch;
1746                        p->active_next_branch = e->active_next_branch;
1747                } else {
1748                        e = active_branches;
1749                        active_branches = e->active_next_branch;
1750                }
1751                e->active = 0;
1752                e->active_next_branch = NULL;
1753                if (e->branch_tree.tree) {
1754                        release_tree_content_recursive(e->branch_tree.tree);
1755                        e->branch_tree.tree = NULL;
1756                }
1757                cur_active_branches--;
1758        }
1759}
1760
1761static void load_branch(struct branch *b)
1762{
1763        load_tree(&b->branch_tree);
1764        if (!b->active) {
1765                b->active = 1;
1766                b->active_next_branch = active_branches;
1767                active_branches = b;
1768                cur_active_branches++;
1769                branch_load_count++;
1770        }
1771}
1772
1773static void file_change_m(struct branch *b)
1774{
1775        const char *p = command_buf.buf + 2;
1776        static struct strbuf uq = STRBUF_INIT;
1777        const char *endp;
1778        struct object_entry *oe = oe;
1779        unsigned char sha1[20];
1780        uint16_t mode, inline_data = 0;
1781
1782        p = get_mode(p, &mode);
1783        if (!p)
1784                die("Corrupt mode: %s", command_buf.buf);
1785        switch (mode) {
1786        case S_IFREG | 0644:
1787        case S_IFREG | 0755:
1788        case S_IFLNK:
1789        case 0644:
1790        case 0755:
1791                /* ok */
1792                break;
1793        default:
1794                die("Corrupt mode: %s", command_buf.buf);
1795        }
1796
1797        if (*p == ':') {
1798                char *x;
1799                oe = find_mark(strtoumax(p + 1, &x, 10));
1800                hashcpy(sha1, oe->sha1);
1801                p = x;
1802        } else if (!prefixcmp(p, "inline")) {
1803                inline_data = 1;
1804                p += 6;
1805        } else {
1806                if (get_sha1_hex(p, sha1))
1807                        die("Invalid SHA1: %s", command_buf.buf);
1808                oe = find_object(sha1);
1809                p += 40;
1810        }
1811        if (*p++ != ' ')
1812                die("Missing space after SHA1: %s", command_buf.buf);
1813
1814        strbuf_reset(&uq);
1815        if (!unquote_c_style(&uq, p, &endp)) {
1816                if (*endp)
1817                        die("Garbage after path in: %s", command_buf.buf);
1818                p = uq.buf;
1819        }
1820
1821        if (inline_data) {
1822                static struct strbuf buf = STRBUF_INIT;
1823
1824                if (p != uq.buf) {
1825                        strbuf_addstr(&uq, p);
1826                        p = uq.buf;
1827                }
1828                read_next_command();
1829                cmd_data(&buf);
1830                store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1831        } else if (oe) {
1832                if (oe->type != OBJ_BLOB)
1833                        die("Not a blob (actually a %s): %s",
1834                                typename(oe->type), command_buf.buf);
1835        } else {
1836                enum object_type type = sha1_object_info(sha1, NULL);
1837                if (type < 0)
1838                        die("Blob not found: %s", command_buf.buf);
1839                if (type != OBJ_BLOB)
1840                        die("Not a blob (actually a %s): %s",
1841                            typename(type), command_buf.buf);
1842        }
1843
1844        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode, NULL);
1845}
1846
1847static void file_change_d(struct branch *b)
1848{
1849        const char *p = command_buf.buf + 2;
1850        static struct strbuf uq = STRBUF_INIT;
1851        const char *endp;
1852
1853        strbuf_reset(&uq);
1854        if (!unquote_c_style(&uq, p, &endp)) {
1855                if (*endp)
1856                        die("Garbage after path in: %s", command_buf.buf);
1857                p = uq.buf;
1858        }
1859        tree_content_remove(&b->branch_tree, p, NULL);
1860}
1861
1862static void file_change_cr(struct branch *b, int rename)
1863{
1864        const char *s, *d;
1865        static struct strbuf s_uq = STRBUF_INIT;
1866        static struct strbuf d_uq = STRBUF_INIT;
1867        const char *endp;
1868        struct tree_entry leaf;
1869
1870        s = command_buf.buf + 2;
1871        strbuf_reset(&s_uq);
1872        if (!unquote_c_style(&s_uq, s, &endp)) {
1873                if (*endp != ' ')
1874                        die("Missing space after source: %s", command_buf.buf);
1875        } else {
1876                endp = strchr(s, ' ');
1877                if (!endp)
1878                        die("Missing space after source: %s", command_buf.buf);
1879                strbuf_add(&s_uq, s, endp - s);
1880        }
1881        s = s_uq.buf;
1882
1883        endp++;
1884        if (!*endp)
1885                die("Missing dest: %s", command_buf.buf);
1886
1887        d = endp;
1888        strbuf_reset(&d_uq);
1889        if (!unquote_c_style(&d_uq, d, &endp)) {
1890                if (*endp)
1891                        die("Garbage after dest in: %s", command_buf.buf);
1892                d = d_uq.buf;
1893        }
1894
1895        memset(&leaf, 0, sizeof(leaf));
1896        if (rename)
1897                tree_content_remove(&b->branch_tree, s, &leaf);
1898        else
1899                tree_content_get(&b->branch_tree, s, &leaf);
1900        if (!leaf.versions[1].mode)
1901                die("Path %s not in branch", s);
1902        tree_content_set(&b->branch_tree, d,
1903                leaf.versions[1].sha1,
1904                leaf.versions[1].mode,
1905                leaf.tree);
1906}
1907
1908static void file_change_deleteall(struct branch *b)
1909{
1910        release_tree_content_recursive(b->branch_tree.tree);
1911        hashclr(b->branch_tree.versions[0].sha1);
1912        hashclr(b->branch_tree.versions[1].sha1);
1913        load_tree(&b->branch_tree);
1914}
1915
1916static void cmd_from_commit(struct branch *b, char *buf, unsigned long size)
1917{
1918        if (!buf || size < 46)
1919                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
1920        if (memcmp("tree ", buf, 5)
1921                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1922                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1923        hashcpy(b->branch_tree.versions[0].sha1,
1924                b->branch_tree.versions[1].sha1);
1925}
1926
1927static void cmd_from_existing(struct branch *b)
1928{
1929        if (is_null_sha1(b->sha1)) {
1930                hashclr(b->branch_tree.versions[0].sha1);
1931                hashclr(b->branch_tree.versions[1].sha1);
1932        } else {
1933                unsigned long size;
1934                char *buf;
1935
1936                buf = read_object_with_reference(b->sha1,
1937                        commit_type, &size, b->sha1);
1938                cmd_from_commit(b, buf, size);
1939                free(buf);
1940        }
1941}
1942
1943static int cmd_from(struct branch *b)
1944{
1945        const char *from;
1946        struct branch *s;
1947
1948        if (prefixcmp(command_buf.buf, "from "))
1949                return 0;
1950
1951        if (b->branch_tree.tree) {
1952                release_tree_content_recursive(b->branch_tree.tree);
1953                b->branch_tree.tree = NULL;
1954        }
1955
1956        from = strchr(command_buf.buf, ' ') + 1;
1957        s = lookup_branch(from);
1958        if (b == s)
1959                die("Can't create a branch from itself: %s", b->name);
1960        else if (s) {
1961                unsigned char *t = s->branch_tree.versions[1].sha1;
1962                hashcpy(b->sha1, s->sha1);
1963                hashcpy(b->branch_tree.versions[0].sha1, t);
1964                hashcpy(b->branch_tree.versions[1].sha1, t);
1965        } else if (*from == ':') {
1966                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1967                struct object_entry *oe = find_mark(idnum);
1968                if (oe->type != OBJ_COMMIT)
1969                        die("Mark :%" PRIuMAX " not a commit", idnum);
1970                hashcpy(b->sha1, oe->sha1);
1971                if (oe->pack_id != MAX_PACK_ID) {
1972                        unsigned long size;
1973                        char *buf = gfi_unpack_entry(oe, &size);
1974                        cmd_from_commit(b, buf, size);
1975                        free(buf);
1976                } else
1977                        cmd_from_existing(b);
1978        } else if (!get_sha1(from, b->sha1))
1979                cmd_from_existing(b);
1980        else
1981                die("Invalid ref name or SHA1 expression: %s", from);
1982
1983        read_next_command();
1984        return 1;
1985}
1986
1987static struct hash_list *cmd_merge(unsigned int *count)
1988{
1989        struct hash_list *list = NULL, *n, *e = e;
1990        const char *from;
1991        struct branch *s;
1992
1993        *count = 0;
1994        while (!prefixcmp(command_buf.buf, "merge ")) {
1995                from = strchr(command_buf.buf, ' ') + 1;
1996                n = xmalloc(sizeof(*n));
1997                s = lookup_branch(from);
1998                if (s)
1999                        hashcpy(n->sha1, s->sha1);
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(n->sha1, oe->sha1);
2006                } else if (!get_sha1(from, n->sha1)) {
2007                        unsigned long size;
2008                        char *buf = read_object_with_reference(n->sha1,
2009                                commit_type, &size, n->sha1);
2010                        if (!buf || size < 46)
2011                                die("Not a valid commit: %s", from);
2012                        free(buf);
2013                } else
2014                        die("Invalid ref name or SHA1 expression: %s", from);
2015
2016                n->next = NULL;
2017                if (list)
2018                        e->next = n;
2019                else
2020                        list = n;
2021                e = n;
2022                (*count)++;
2023                read_next_command();
2024        }
2025        return list;
2026}
2027
2028static void cmd_new_commit(void)
2029{
2030        static struct strbuf msg = STRBUF_INIT;
2031        struct branch *b;
2032        char *sp;
2033        char *author = NULL;
2034        char *committer = NULL;
2035        struct hash_list *merge_list = NULL;
2036        unsigned int merge_count;
2037
2038        /* Obtain the branch name from the rest of our command */
2039        sp = strchr(command_buf.buf, ' ') + 1;
2040        b = lookup_branch(sp);
2041        if (!b)
2042                b = new_branch(sp);
2043
2044        read_next_command();
2045        cmd_mark();
2046        if (!prefixcmp(command_buf.buf, "author ")) {
2047                author = parse_ident(command_buf.buf + 7);
2048                read_next_command();
2049        }
2050        if (!prefixcmp(command_buf.buf, "committer ")) {
2051                committer = parse_ident(command_buf.buf + 10);
2052                read_next_command();
2053        }
2054        if (!committer)
2055                die("Expected committer but didn't get one");
2056        cmd_data(&msg);
2057        read_next_command();
2058        cmd_from(b);
2059        merge_list = cmd_merge(&merge_count);
2060
2061        /* ensure the branch is active/loaded */
2062        if (!b->branch_tree.tree || !max_active_branches) {
2063                unload_one_branch();
2064                load_branch(b);
2065        }
2066
2067        /* file_change* */
2068        while (command_buf.len > 0) {
2069                if (!prefixcmp(command_buf.buf, "M "))
2070                        file_change_m(b);
2071                else if (!prefixcmp(command_buf.buf, "D "))
2072                        file_change_d(b);
2073                else if (!prefixcmp(command_buf.buf, "R "))
2074                        file_change_cr(b, 1);
2075                else if (!prefixcmp(command_buf.buf, "C "))
2076                        file_change_cr(b, 0);
2077                else if (!strcmp("deleteall", command_buf.buf))
2078                        file_change_deleteall(b);
2079                else {
2080                        unread_command_buf = 1;
2081                        break;
2082                }
2083                if (read_next_command() == EOF)
2084                        break;
2085        }
2086
2087        /* build the tree and the commit */
2088        store_tree(&b->branch_tree);
2089        hashcpy(b->branch_tree.versions[0].sha1,
2090                b->branch_tree.versions[1].sha1);
2091
2092        strbuf_reset(&new_data);
2093        strbuf_addf(&new_data, "tree %s\n",
2094                sha1_to_hex(b->branch_tree.versions[1].sha1));
2095        if (!is_null_sha1(b->sha1))
2096                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2097        while (merge_list) {
2098                struct hash_list *next = merge_list->next;
2099                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2100                free(merge_list);
2101                merge_list = next;
2102        }
2103        strbuf_addf(&new_data,
2104                "author %s\n"
2105                "committer %s\n"
2106                "\n",
2107                author ? author : committer, committer);
2108        strbuf_addbuf(&new_data, &msg);
2109        free(author);
2110        free(committer);
2111
2112        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2113                b->pack_id = pack_id;
2114        b->last_commit = object_count_by_type[OBJ_COMMIT];
2115}
2116
2117static void cmd_new_tag(void)
2118{
2119        static struct strbuf msg = STRBUF_INIT;
2120        char *sp;
2121        const char *from;
2122        char *tagger;
2123        struct branch *s;
2124        struct tag *t;
2125        uintmax_t from_mark = 0;
2126        unsigned char sha1[20];
2127
2128        /* Obtain the new tag name from the rest of our command */
2129        sp = strchr(command_buf.buf, ' ') + 1;
2130        t = pool_alloc(sizeof(struct tag));
2131        t->next_tag = NULL;
2132        t->name = pool_strdup(sp);
2133        if (last_tag)
2134                last_tag->next_tag = t;
2135        else
2136                first_tag = t;
2137        last_tag = t;
2138        read_next_command();
2139
2140        /* from ... */
2141        if (prefixcmp(command_buf.buf, "from "))
2142                die("Expected from command, got %s", command_buf.buf);
2143        from = strchr(command_buf.buf, ' ') + 1;
2144        s = lookup_branch(from);
2145        if (s) {
2146                hashcpy(sha1, s->sha1);
2147        } else if (*from == ':') {
2148                struct object_entry *oe;
2149                from_mark = strtoumax(from + 1, NULL, 10);
2150                oe = find_mark(from_mark);
2151                if (oe->type != OBJ_COMMIT)
2152                        die("Mark :%" PRIuMAX " not a commit", from_mark);
2153                hashcpy(sha1, oe->sha1);
2154        } else if (!get_sha1(from, sha1)) {
2155                unsigned long size;
2156                char *buf;
2157
2158                buf = read_object_with_reference(sha1,
2159                        commit_type, &size, sha1);
2160                if (!buf || size < 46)
2161                        die("Not a valid commit: %s", from);
2162                free(buf);
2163        } else
2164                die("Invalid ref name or SHA1 expression: %s", from);
2165        read_next_command();
2166
2167        /* tagger ... */
2168        if (prefixcmp(command_buf.buf, "tagger "))
2169                die("Expected tagger command, got %s", command_buf.buf);
2170        tagger = parse_ident(command_buf.buf + 7);
2171
2172        /* tag payload/message */
2173        read_next_command();
2174        cmd_data(&msg);
2175
2176        /* build the tag object */
2177        strbuf_reset(&new_data);
2178        strbuf_addf(&new_data,
2179                "object %s\n"
2180                "type %s\n"
2181                "tag %s\n"
2182                "tagger %s\n"
2183                "\n",
2184                sha1_to_hex(sha1), commit_type, t->name, tagger);
2185        strbuf_addbuf(&new_data, &msg);
2186        free(tagger);
2187
2188        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2189                t->pack_id = MAX_PACK_ID;
2190        else
2191                t->pack_id = pack_id;
2192}
2193
2194static void cmd_reset_branch(void)
2195{
2196        struct branch *b;
2197        char *sp;
2198
2199        /* Obtain the branch name from the rest of our command */
2200        sp = strchr(command_buf.buf, ' ') + 1;
2201        b = lookup_branch(sp);
2202        if (b) {
2203                hashclr(b->sha1);
2204                hashclr(b->branch_tree.versions[0].sha1);
2205                hashclr(b->branch_tree.versions[1].sha1);
2206                if (b->branch_tree.tree) {
2207                        release_tree_content_recursive(b->branch_tree.tree);
2208                        b->branch_tree.tree = NULL;
2209                }
2210        }
2211        else
2212                b = new_branch(sp);
2213        read_next_command();
2214        if (!cmd_from(b) && command_buf.len > 0)
2215                unread_command_buf = 1;
2216}
2217
2218static void cmd_checkpoint(void)
2219{
2220        if (object_count) {
2221                cycle_packfile();
2222                dump_branches();
2223                dump_tags();
2224                dump_marks();
2225        }
2226        skip_optional_lf();
2227}
2228
2229static void cmd_progress(void)
2230{
2231        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2232        fputc('\n', stdout);
2233        fflush(stdout);
2234        skip_optional_lf();
2235}
2236
2237static void import_marks(const char *input_file)
2238{
2239        char line[512];
2240        FILE *f = fopen(input_file, "r");
2241        if (!f)
2242                die("cannot read %s: %s", input_file, strerror(errno));
2243        while (fgets(line, sizeof(line), f)) {
2244                uintmax_t mark;
2245                char *end;
2246                unsigned char sha1[20];
2247                struct object_entry *e;
2248
2249                end = strchr(line, '\n');
2250                if (line[0] != ':' || !end)
2251                        die("corrupt mark line: %s", line);
2252                *end = 0;
2253                mark = strtoumax(line + 1, &end, 10);
2254                if (!mark || end == line + 1
2255                        || *end != ' ' || get_sha1(end + 1, sha1))
2256                        die("corrupt mark line: %s", line);
2257                e = find_object(sha1);
2258                if (!e) {
2259                        enum object_type type = sha1_object_info(sha1, NULL);
2260                        if (type < 0)
2261                                die("object not found: %s", sha1_to_hex(sha1));
2262                        e = insert_object(sha1);
2263                        e->type = type;
2264                        e->pack_id = MAX_PACK_ID;
2265                        e->offset = 1; /* just not zero! */
2266                }
2267                insert_mark(mark, e);
2268        }
2269        fclose(f);
2270}
2271
2272static const char fast_import_usage[] =
2273"git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2274
2275int main(int argc, const char **argv)
2276{
2277        unsigned int i, show_stats = 1;
2278
2279        git_config(git_default_config);
2280        alloc_objects(object_entry_alloc);
2281        strbuf_init(&command_buf, 0);
2282        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2283        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2284        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2285        marks = pool_calloc(1, sizeof(struct mark_set));
2286
2287        for (i = 1; i < argc; i++) {
2288                const char *a = argv[i];
2289
2290                if (*a != '-' || !strcmp(a, "--"))
2291                        break;
2292                else if (!prefixcmp(a, "--date-format=")) {
2293                        const char *fmt = a + 14;
2294                        if (!strcmp(fmt, "raw"))
2295                                whenspec = WHENSPEC_RAW;
2296                        else if (!strcmp(fmt, "rfc2822"))
2297                                whenspec = WHENSPEC_RFC2822;
2298                        else if (!strcmp(fmt, "now"))
2299                                whenspec = WHENSPEC_NOW;
2300                        else
2301                                die("unknown --date-format argument %s", fmt);
2302                }
2303                else if (!prefixcmp(a, "--max-pack-size="))
2304                        max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2305                else if (!prefixcmp(a, "--depth=")) {
2306                        max_depth = strtoul(a + 8, NULL, 0);
2307                        if (max_depth > MAX_DEPTH)
2308                                die("--depth cannot exceed %u", MAX_DEPTH);
2309                }
2310                else if (!prefixcmp(a, "--active-branches="))
2311                        max_active_branches = strtoul(a + 18, NULL, 0);
2312                else if (!prefixcmp(a, "--import-marks="))
2313                        import_marks(a + 15);
2314                else if (!prefixcmp(a, "--export-marks="))
2315                        mark_file = a + 15;
2316                else if (!prefixcmp(a, "--export-pack-edges=")) {
2317                        if (pack_edges)
2318                                fclose(pack_edges);
2319                        pack_edges = fopen(a + 20, "a");
2320                        if (!pack_edges)
2321                                die("Cannot open %s: %s", a + 20, strerror(errno));
2322                } else if (!strcmp(a, "--force"))
2323                        force_update = 1;
2324                else if (!strcmp(a, "--quiet"))
2325                        show_stats = 0;
2326                else if (!strcmp(a, "--stats"))
2327                        show_stats = 1;
2328                else
2329                        die("unknown option %s", a);
2330        }
2331        if (i != argc)
2332                usage(fast_import_usage);
2333
2334        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2335        for (i = 0; i < (cmd_save - 1); i++)
2336                rc_free[i].next = &rc_free[i + 1];
2337        rc_free[cmd_save - 1].next = NULL;
2338
2339        prepare_packed_git();
2340        start_packfile();
2341        set_die_routine(die_nicely);
2342        while (read_next_command() != EOF) {
2343                if (!strcmp("blob", command_buf.buf))
2344                        cmd_new_blob();
2345                else if (!prefixcmp(command_buf.buf, "commit "))
2346                        cmd_new_commit();
2347                else if (!prefixcmp(command_buf.buf, "tag "))
2348                        cmd_new_tag();
2349                else if (!prefixcmp(command_buf.buf, "reset "))
2350                        cmd_reset_branch();
2351                else if (!strcmp("checkpoint", command_buf.buf))
2352                        cmd_checkpoint();
2353                else if (!prefixcmp(command_buf.buf, "progress "))
2354                        cmd_progress();
2355                else
2356                        die("Unsupported command: %s", command_buf.buf);
2357        }
2358        end_packfile();
2359
2360        dump_branches();
2361        dump_tags();
2362        unkeep_all_packs();
2363        dump_marks();
2364
2365        if (pack_edges)
2366                fclose(pack_edges);
2367
2368        if (show_stats) {
2369                uintmax_t total_count = 0, duplicate_count = 0;
2370                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2371                        total_count += object_count_by_type[i];
2372                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2373                        duplicate_count += duplicate_count_by_type[i];
2374
2375                fprintf(stderr, "%s statistics:\n", argv[0]);
2376                fprintf(stderr, "---------------------------------------------------------------------\n");
2377                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2378                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2379                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]);
2380                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]);
2381                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]);
2382                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]);
2383                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2384                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2385                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2386                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2387                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2388                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2389                fprintf(stderr, "---------------------------------------------------------------------\n");
2390                pack_report();
2391                fprintf(stderr, "---------------------------------------------------------------------\n");
2392                fprintf(stderr, "\n");
2393        }
2394
2395        return failure ? 1 : 0;
2396}