fast-import.con commit Teach core.bigfilethreashold to pack-objects (1536628)
   1/*
   2(See Documentation/git-fast-import.txt for maintained documentation.)
   3Format of STDIN stream:
   4
   5  stream ::= cmd*;
   6
   7  cmd ::= new_blob
   8        | new_commit
   9        | new_tag
  10        | reset_branch
  11        | checkpoint
  12        | progress
  13        ;
  14
  15  new_blob ::= 'blob' lf
  16    mark?
  17    file_content;
  18  file_content ::= data;
  19
  20  new_commit ::= 'commit' sp ref_str lf
  21    mark?
  22    ('author' (sp name)? sp '<' email '>' sp when lf)?
  23    'committer' (sp name)? sp '<' email '>' sp when lf
  24    commit_msg
  25    ('from' sp committish lf)?
  26    ('merge' sp committish lf)*
  27    file_change*
  28    lf?;
  29  commit_msg ::= data;
  30
  31  file_change ::= file_clr
  32    | file_del
  33    | file_rnm
  34    | file_cpy
  35    | file_obm
  36    | file_inm;
  37  file_clr ::= 'deleteall' lf;
  38  file_del ::= 'D' sp path_str lf;
  39  file_rnm ::= 'R' sp path_str sp path_str lf;
  40  file_cpy ::= 'C' sp path_str sp path_str lf;
  41  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  42  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  43    data;
  44  note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
  45  note_inm ::= 'N' sp 'inline' sp committish lf
  46    data;
  47
  48  new_tag ::= 'tag' sp tag_str lf
  49    'from' sp committish lf
  50    ('tagger' (sp name)? sp '<' email '>' sp when lf)?
  51    tag_msg;
  52  tag_msg ::= data;
  53
  54  reset_branch ::= 'reset' sp ref_str lf
  55    ('from' sp committish lf)?
  56    lf?;
  57
  58  checkpoint ::= 'checkpoint' lf
  59    lf?;
  60
  61  progress ::= 'progress' sp not_lf* lf
  62    lf?;
  63
  64     # note: the first idnum in a stream should be 1 and subsequent
  65     # idnums should not have gaps between values as this will cause
  66     # the stream parser to reserve space for the gapped values.  An
  67     # idnum can be updated in the future to a new object by issuing
  68     # a new mark directive with the old idnum.
  69     #
  70  mark ::= 'mark' sp idnum lf;
  71  data ::= (delimited_data | exact_data)
  72    lf?;
  73
  74    # note: delim may be any string but must not contain lf.
  75    # data_line may contain any data but must not be exactly
  76    # delim.
  77  delimited_data ::= 'data' sp '<<' delim lf
  78    (data_line lf)*
  79    delim lf;
  80
  81     # note: declen indicates the length of binary_data in bytes.
  82     # declen does not include the lf preceding the binary data.
  83     #
  84  exact_data ::= 'data' sp declen lf
  85    binary_data;
  86
  87     # note: quoted strings are C-style quoting supporting \c for
  88     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  89     # is the signed byte value in octal.  Note that the only
  90     # characters which must actually be escaped to protect the
  91     # stream formatting is: \, " and LF.  Otherwise these values
  92     # are UTF8.
  93     #
  94  committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
  95  ref_str     ::= ref;
  96  sha1exp_str ::= sha1exp;
  97  tag_str     ::= tag;
  98  path_str    ::= path    | '"' quoted(path)    '"' ;
  99  mode        ::= '100644' | '644'
 100                | '100755' | '755'
 101                | '120000'
 102                ;
 103
 104  declen ::= # unsigned 32 bit value, ascii base10 notation;
 105  bigint ::= # unsigned integer value, ascii base10 notation;
 106  binary_data ::= # file content, not interpreted;
 107
 108  when         ::= raw_when | rfc2822_when;
 109  raw_when     ::= ts sp tz;
 110  rfc2822_when ::= # Valid RFC 2822 date and time;
 111
 112  sp ::= # ASCII space character;
 113  lf ::= # ASCII newline (LF) character;
 114
 115     # note: a colon (':') must precede the numerical value assigned to
 116     # an idnum.  This is to distinguish it from a ref or tag name as
 117     # GIT does not permit ':' in ref or tag strings.
 118     #
 119  idnum   ::= ':' bigint;
 120  path    ::= # GIT style file path, e.g. "a/b/c";
 121  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 122  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 123  sha1exp ::= # Any valid GIT SHA1 expression;
 124  hexsha1 ::= # SHA1 in hexadecimal format;
 125
 126     # note: name and email are UTF8 strings, however name must not
 127     # contain '<' or lf and email must not contain any of the
 128     # following: '<', '>', lf.
 129     #
 130  name  ::= # valid GIT author/committer name;
 131  email ::= # valid GIT author/committer email;
 132  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 133  tz    ::= # GIT style timezone;
 134
 135     # note: comments and cat requests may appear anywhere
 136     # in the input, except within a data command.  Any form
 137     # of the data command always escapes the related input
 138     # from comment processing.
 139     #
 140     # In case it is not clear, the '#' that starts the comment
 141     # must be the first character on that line (an lf
 142     # preceded it).
 143     #
 144  cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
 145
 146  comment ::= '#' not_lf* lf;
 147  not_lf  ::= # Any byte that is not ASCII newline (LF);
 148*/
 149
 150#include "builtin.h"
 151#include "cache.h"
 152#include "object.h"
 153#include "blob.h"
 154#include "tree.h"
 155#include "commit.h"
 156#include "delta.h"
 157#include "pack.h"
 158#include "refs.h"
 159#include "csum-file.h"
 160#include "quote.h"
 161#include "exec_cmd.h"
 162#include "dir.h"
 163
 164#define PACK_ID_BITS 16
 165#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 166#define DEPTH_BITS 13
 167#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 168
 169struct object_entry
 170{
 171        struct pack_idx_entry idx;
 172        struct object_entry *next;
 173        uint32_t type : TYPE_BITS,
 174                pack_id : PACK_ID_BITS,
 175                depth : DEPTH_BITS;
 176};
 177
 178struct object_entry_pool
 179{
 180        struct object_entry_pool *next_pool;
 181        struct object_entry *next_free;
 182        struct object_entry *end;
 183        struct object_entry entries[FLEX_ARRAY]; /* more */
 184};
 185
 186struct mark_set
 187{
 188        union {
 189                struct object_entry *marked[1024];
 190                struct mark_set *sets[1024];
 191        } data;
 192        unsigned int shift;
 193};
 194
 195struct last_object
 196{
 197        struct strbuf data;
 198        off_t offset;
 199        unsigned int depth;
 200        unsigned no_swap : 1;
 201};
 202
 203struct mem_pool
 204{
 205        struct mem_pool *next_pool;
 206        char *next_free;
 207        char *end;
 208        uintmax_t space[FLEX_ARRAY]; /* more */
 209};
 210
 211struct atom_str
 212{
 213        struct atom_str *next_atom;
 214        unsigned short str_len;
 215        char str_dat[FLEX_ARRAY]; /* more */
 216};
 217
 218struct tree_content;
 219struct tree_entry
 220{
 221        struct tree_content *tree;
 222        struct atom_str *name;
 223        struct tree_entry_ms
 224        {
 225                uint16_t mode;
 226                unsigned char sha1[20];
 227        } versions[2];
 228};
 229
 230struct tree_content
 231{
 232        unsigned int entry_capacity; /* must match avail_tree_content */
 233        unsigned int entry_count;
 234        unsigned int delta_depth;
 235        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 236};
 237
 238struct avail_tree_content
 239{
 240        unsigned int entry_capacity; /* must match tree_content */
 241        struct avail_tree_content *next_avail;
 242};
 243
 244struct branch
 245{
 246        struct branch *table_next_branch;
 247        struct branch *active_next_branch;
 248        const char *name;
 249        struct tree_entry branch_tree;
 250        uintmax_t last_commit;
 251        uintmax_t num_notes;
 252        unsigned active : 1;
 253        unsigned pack_id : PACK_ID_BITS;
 254        unsigned char sha1[20];
 255};
 256
 257struct tag
 258{
 259        struct tag *next_tag;
 260        const char *name;
 261        unsigned int pack_id;
 262        unsigned char sha1[20];
 263};
 264
 265struct hash_list
 266{
 267        struct hash_list *next;
 268        unsigned char sha1[20];
 269};
 270
 271typedef enum {
 272        WHENSPEC_RAW = 1,
 273        WHENSPEC_RFC2822,
 274        WHENSPEC_NOW
 275} whenspec_type;
 276
 277struct recent_command
 278{
 279        struct recent_command *prev;
 280        struct recent_command *next;
 281        char *buf;
 282};
 283
 284/* Configured limits on output */
 285static unsigned long max_depth = 10;
 286static off_t max_packsize;
 287static int force_update;
 288static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 289static int pack_compression_seen;
 290
 291/* Stats and misc. counters */
 292static uintmax_t alloc_count;
 293static uintmax_t marks_set_count;
 294static uintmax_t object_count_by_type[1 << TYPE_BITS];
 295static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 296static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 297static unsigned long object_count;
 298static unsigned long branch_count;
 299static unsigned long branch_load_count;
 300static int failure;
 301static FILE *pack_edges;
 302static unsigned int show_stats = 1;
 303static int global_argc;
 304static const char **global_argv;
 305
 306/* Memory pools */
 307static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 308static size_t total_allocd;
 309static struct mem_pool *mem_pool;
 310
 311/* Atom management */
 312static unsigned int atom_table_sz = 4451;
 313static unsigned int atom_cnt;
 314static struct atom_str **atom_table;
 315
 316/* The .pack file being generated */
 317static unsigned int pack_id;
 318static struct sha1file *pack_file;
 319static struct packed_git *pack_data;
 320static struct packed_git **all_packs;
 321static off_t pack_size;
 322
 323/* Table of objects we've written. */
 324static unsigned int object_entry_alloc = 5000;
 325static struct object_entry_pool *blocks;
 326static struct object_entry *object_table[1 << 16];
 327static struct mark_set *marks;
 328static const char *export_marks_file;
 329static const char *import_marks_file;
 330static int import_marks_file_from_stream;
 331static int relative_marks_paths;
 332
 333/* Our last blob */
 334static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 335
 336/* Tree management */
 337static unsigned int tree_entry_alloc = 1000;
 338static void *avail_tree_entry;
 339static unsigned int avail_tree_table_sz = 100;
 340static struct avail_tree_content **avail_tree_table;
 341static struct strbuf old_tree = STRBUF_INIT;
 342static struct strbuf new_tree = STRBUF_INIT;
 343
 344/* Branch data */
 345static unsigned long max_active_branches = 5;
 346static unsigned long cur_active_branches;
 347static unsigned long branch_table_sz = 1039;
 348static struct branch **branch_table;
 349static struct branch *active_branches;
 350
 351/* Tag data */
 352static struct tag *first_tag;
 353static struct tag *last_tag;
 354
 355/* Input stream parsing */
 356static whenspec_type whenspec = WHENSPEC_RAW;
 357static struct strbuf command_buf = STRBUF_INIT;
 358static int unread_command_buf;
 359static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 360static struct recent_command *cmd_tail = &cmd_hist;
 361static struct recent_command *rc_free;
 362static unsigned int cmd_save = 100;
 363static uintmax_t next_mark;
 364static struct strbuf new_data = STRBUF_INIT;
 365static int seen_data_command;
 366
 367/* Signal handling */
 368static volatile sig_atomic_t checkpoint_requested;
 369
 370/* Where to write output of cat-blob commands */
 371static int cat_blob_fd = STDOUT_FILENO;
 372
 373static void parse_argv(void);
 374static void parse_cat_blob(void);
 375
 376static void write_branch_report(FILE *rpt, struct branch *b)
 377{
 378        fprintf(rpt, "%s:\n", b->name);
 379
 380        fprintf(rpt, "  status      :");
 381        if (b->active)
 382                fputs(" active", rpt);
 383        if (b->branch_tree.tree)
 384                fputs(" loaded", rpt);
 385        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 386                fputs(" dirty", rpt);
 387        fputc('\n', rpt);
 388
 389        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 390        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 391        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 392        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 393
 394        fputs("  last pack   : ", rpt);
 395        if (b->pack_id < MAX_PACK_ID)
 396                fprintf(rpt, "%u", b->pack_id);
 397        fputc('\n', rpt);
 398
 399        fputc('\n', rpt);
 400}
 401
 402static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 403
 404static void write_crash_report(const char *err)
 405{
 406        char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 407        FILE *rpt = fopen(loc, "w");
 408        struct branch *b;
 409        unsigned long lu;
 410        struct recent_command *rc;
 411
 412        if (!rpt) {
 413                error("can't write crash report %s: %s", loc, strerror(errno));
 414                return;
 415        }
 416
 417        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 418
 419        fprintf(rpt, "fast-import crash report:\n");
 420        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 421        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 422        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 423        fputc('\n', rpt);
 424
 425        fputs("fatal: ", rpt);
 426        fputs(err, rpt);
 427        fputc('\n', rpt);
 428
 429        fputc('\n', rpt);
 430        fputs("Most Recent Commands Before Crash\n", rpt);
 431        fputs("---------------------------------\n", rpt);
 432        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 433                if (rc->next == &cmd_hist)
 434                        fputs("* ", rpt);
 435                else
 436                        fputs("  ", rpt);
 437                fputs(rc->buf, rpt);
 438                fputc('\n', rpt);
 439        }
 440
 441        fputc('\n', rpt);
 442        fputs("Active Branch LRU\n", rpt);
 443        fputs("-----------------\n", rpt);
 444        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 445                cur_active_branches,
 446                max_active_branches);
 447        fputc('\n', rpt);
 448        fputs("  pos  clock name\n", rpt);
 449        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 450        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 451                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 452                        ++lu, b->last_commit, b->name);
 453
 454        fputc('\n', rpt);
 455        fputs("Inactive Branches\n", rpt);
 456        fputs("-----------------\n", rpt);
 457        for (lu = 0; lu < branch_table_sz; lu++) {
 458                for (b = branch_table[lu]; b; b = b->table_next_branch)
 459                        write_branch_report(rpt, b);
 460        }
 461
 462        if (first_tag) {
 463                struct tag *tg;
 464                fputc('\n', rpt);
 465                fputs("Annotated Tags\n", rpt);
 466                fputs("--------------\n", rpt);
 467                for (tg = first_tag; tg; tg = tg->next_tag) {
 468                        fputs(sha1_to_hex(tg->sha1), rpt);
 469                        fputc(' ', rpt);
 470                        fputs(tg->name, rpt);
 471                        fputc('\n', rpt);
 472                }
 473        }
 474
 475        fputc('\n', rpt);
 476        fputs("Marks\n", rpt);
 477        fputs("-----\n", rpt);
 478        if (export_marks_file)
 479                fprintf(rpt, "  exported to %s\n", export_marks_file);
 480        else
 481                dump_marks_helper(rpt, 0, marks);
 482
 483        fputc('\n', rpt);
 484        fputs("-------------------\n", rpt);
 485        fputs("END OF CRASH REPORT\n", rpt);
 486        fclose(rpt);
 487}
 488
 489static void end_packfile(void);
 490static void unkeep_all_packs(void);
 491static void dump_marks(void);
 492
 493static NORETURN void die_nicely(const char *err, va_list params)
 494{
 495        static int zombie;
 496        char message[2 * PATH_MAX];
 497
 498        vsnprintf(message, sizeof(message), err, params);
 499        fputs("fatal: ", stderr);
 500        fputs(message, stderr);
 501        fputc('\n', stderr);
 502
 503        if (!zombie) {
 504                zombie = 1;
 505                write_crash_report(message);
 506                end_packfile();
 507                unkeep_all_packs();
 508                dump_marks();
 509        }
 510        exit(128);
 511}
 512
 513#ifndef SIGUSR1 /* Windows, for example */
 514
 515static void set_checkpoint_signal(void)
 516{
 517}
 518
 519#else
 520
 521static void checkpoint_signal(int signo)
 522{
 523        checkpoint_requested = 1;
 524}
 525
 526static void set_checkpoint_signal(void)
 527{
 528        struct sigaction sa;
 529
 530        memset(&sa, 0, sizeof(sa));
 531        sa.sa_handler = checkpoint_signal;
 532        sigemptyset(&sa.sa_mask);
 533        sa.sa_flags = SA_RESTART;
 534        sigaction(SIGUSR1, &sa, NULL);
 535}
 536
 537#endif
 538
 539static void alloc_objects(unsigned int cnt)
 540{
 541        struct object_entry_pool *b;
 542
 543        b = xmalloc(sizeof(struct object_entry_pool)
 544                + cnt * sizeof(struct object_entry));
 545        b->next_pool = blocks;
 546        b->next_free = b->entries;
 547        b->end = b->entries + cnt;
 548        blocks = b;
 549        alloc_count += cnt;
 550}
 551
 552static struct object_entry *new_object(unsigned char *sha1)
 553{
 554        struct object_entry *e;
 555
 556        if (blocks->next_free == blocks->end)
 557                alloc_objects(object_entry_alloc);
 558
 559        e = blocks->next_free++;
 560        hashcpy(e->idx.sha1, sha1);
 561        return e;
 562}
 563
 564static struct object_entry *find_object(unsigned char *sha1)
 565{
 566        unsigned int h = sha1[0] << 8 | sha1[1];
 567        struct object_entry *e;
 568        for (e = object_table[h]; e; e = e->next)
 569                if (!hashcmp(sha1, e->idx.sha1))
 570                        return e;
 571        return NULL;
 572}
 573
 574static struct object_entry *insert_object(unsigned char *sha1)
 575{
 576        unsigned int h = sha1[0] << 8 | sha1[1];
 577        struct object_entry *e = object_table[h];
 578
 579        while (e) {
 580                if (!hashcmp(sha1, e->idx.sha1))
 581                        return e;
 582                e = e->next;
 583        }
 584
 585        e = new_object(sha1);
 586        e->next = object_table[h];
 587        e->idx.offset = 0;
 588        object_table[h] = e;
 589        return e;
 590}
 591
 592static unsigned int hc_str(const char *s, size_t len)
 593{
 594        unsigned int r = 0;
 595        while (len-- > 0)
 596                r = r * 31 + *s++;
 597        return r;
 598}
 599
 600static void *pool_alloc(size_t len)
 601{
 602        struct mem_pool *p;
 603        void *r;
 604
 605        /* round up to a 'uintmax_t' alignment */
 606        if (len & (sizeof(uintmax_t) - 1))
 607                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 608
 609        for (p = mem_pool; p; p = p->next_pool)
 610                if ((p->end - p->next_free >= len))
 611                        break;
 612
 613        if (!p) {
 614                if (len >= (mem_pool_alloc/2)) {
 615                        total_allocd += len;
 616                        return xmalloc(len);
 617                }
 618                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 619                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 620                p->next_pool = mem_pool;
 621                p->next_free = (char *) p->space;
 622                p->end = p->next_free + mem_pool_alloc;
 623                mem_pool = p;
 624        }
 625
 626        r = p->next_free;
 627        p->next_free += len;
 628        return r;
 629}
 630
 631static void *pool_calloc(size_t count, size_t size)
 632{
 633        size_t len = count * size;
 634        void *r = pool_alloc(len);
 635        memset(r, 0, len);
 636        return r;
 637}
 638
 639static char *pool_strdup(const char *s)
 640{
 641        char *r = pool_alloc(strlen(s) + 1);
 642        strcpy(r, s);
 643        return r;
 644}
 645
 646static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 647{
 648        struct mark_set *s = marks;
 649        while ((idnum >> s->shift) >= 1024) {
 650                s = pool_calloc(1, sizeof(struct mark_set));
 651                s->shift = marks->shift + 10;
 652                s->data.sets[0] = marks;
 653                marks = s;
 654        }
 655        while (s->shift) {
 656                uintmax_t i = idnum >> s->shift;
 657                idnum -= i << s->shift;
 658                if (!s->data.sets[i]) {
 659                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 660                        s->data.sets[i]->shift = s->shift - 10;
 661                }
 662                s = s->data.sets[i];
 663        }
 664        if (!s->data.marked[idnum])
 665                marks_set_count++;
 666        s->data.marked[idnum] = oe;
 667}
 668
 669static struct object_entry *find_mark(uintmax_t idnum)
 670{
 671        uintmax_t orig_idnum = idnum;
 672        struct mark_set *s = marks;
 673        struct object_entry *oe = NULL;
 674        if ((idnum >> s->shift) < 1024) {
 675                while (s && s->shift) {
 676                        uintmax_t i = idnum >> s->shift;
 677                        idnum -= i << s->shift;
 678                        s = s->data.sets[i];
 679                }
 680                if (s)
 681                        oe = s->data.marked[idnum];
 682        }
 683        if (!oe)
 684                die("mark :%" PRIuMAX " not declared", orig_idnum);
 685        return oe;
 686}
 687
 688static struct atom_str *to_atom(const char *s, unsigned short len)
 689{
 690        unsigned int hc = hc_str(s, len) % atom_table_sz;
 691        struct atom_str *c;
 692
 693        for (c = atom_table[hc]; c; c = c->next_atom)
 694                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 695                        return c;
 696
 697        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 698        c->str_len = len;
 699        strncpy(c->str_dat, s, len);
 700        c->str_dat[len] = 0;
 701        c->next_atom = atom_table[hc];
 702        atom_table[hc] = c;
 703        atom_cnt++;
 704        return c;
 705}
 706
 707static struct branch *lookup_branch(const char *name)
 708{
 709        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 710        struct branch *b;
 711
 712        for (b = branch_table[hc]; b; b = b->table_next_branch)
 713                if (!strcmp(name, b->name))
 714                        return b;
 715        return NULL;
 716}
 717
 718static struct branch *new_branch(const char *name)
 719{
 720        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 721        struct branch *b = lookup_branch(name);
 722
 723        if (b)
 724                die("Invalid attempt to create duplicate branch: %s", name);
 725        switch (check_ref_format(name)) {
 726        case 0: break; /* its valid */
 727        case CHECK_REF_FORMAT_ONELEVEL:
 728                break; /* valid, but too few '/', allow anyway */
 729        default:
 730                die("Branch name doesn't conform to GIT standards: %s", name);
 731        }
 732
 733        b = pool_calloc(1, sizeof(struct branch));
 734        b->name = pool_strdup(name);
 735        b->table_next_branch = branch_table[hc];
 736        b->branch_tree.versions[0].mode = S_IFDIR;
 737        b->branch_tree.versions[1].mode = S_IFDIR;
 738        b->num_notes = 0;
 739        b->active = 0;
 740        b->pack_id = MAX_PACK_ID;
 741        branch_table[hc] = b;
 742        branch_count++;
 743        return b;
 744}
 745
 746static unsigned int hc_entries(unsigned int cnt)
 747{
 748        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 749        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 750}
 751
 752static struct tree_content *new_tree_content(unsigned int cnt)
 753{
 754        struct avail_tree_content *f, *l = NULL;
 755        struct tree_content *t;
 756        unsigned int hc = hc_entries(cnt);
 757
 758        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 759                if (f->entry_capacity >= cnt)
 760                        break;
 761
 762        if (f) {
 763                if (l)
 764                        l->next_avail = f->next_avail;
 765                else
 766                        avail_tree_table[hc] = f->next_avail;
 767        } else {
 768                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 769                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 770                f->entry_capacity = cnt;
 771        }
 772
 773        t = (struct tree_content*)f;
 774        t->entry_count = 0;
 775        t->delta_depth = 0;
 776        return t;
 777}
 778
 779static void release_tree_entry(struct tree_entry *e);
 780static void release_tree_content(struct tree_content *t)
 781{
 782        struct avail_tree_content *f = (struct avail_tree_content*)t;
 783        unsigned int hc = hc_entries(f->entry_capacity);
 784        f->next_avail = avail_tree_table[hc];
 785        avail_tree_table[hc] = f;
 786}
 787
 788static void release_tree_content_recursive(struct tree_content *t)
 789{
 790        unsigned int i;
 791        for (i = 0; i < t->entry_count; i++)
 792                release_tree_entry(t->entries[i]);
 793        release_tree_content(t);
 794}
 795
 796static struct tree_content *grow_tree_content(
 797        struct tree_content *t,
 798        int amt)
 799{
 800        struct tree_content *r = new_tree_content(t->entry_count + amt);
 801        r->entry_count = t->entry_count;
 802        r->delta_depth = t->delta_depth;
 803        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 804        release_tree_content(t);
 805        return r;
 806}
 807
 808static struct tree_entry *new_tree_entry(void)
 809{
 810        struct tree_entry *e;
 811
 812        if (!avail_tree_entry) {
 813                unsigned int n = tree_entry_alloc;
 814                total_allocd += n * sizeof(struct tree_entry);
 815                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 816                while (n-- > 1) {
 817                        *((void**)e) = e + 1;
 818                        e++;
 819                }
 820                *((void**)e) = NULL;
 821        }
 822
 823        e = avail_tree_entry;
 824        avail_tree_entry = *((void**)e);
 825        return e;
 826}
 827
 828static void release_tree_entry(struct tree_entry *e)
 829{
 830        if (e->tree)
 831                release_tree_content_recursive(e->tree);
 832        *((void**)e) = avail_tree_entry;
 833        avail_tree_entry = e;
 834}
 835
 836static struct tree_content *dup_tree_content(struct tree_content *s)
 837{
 838        struct tree_content *d;
 839        struct tree_entry *a, *b;
 840        unsigned int i;
 841
 842        if (!s)
 843                return NULL;
 844        d = new_tree_content(s->entry_count);
 845        for (i = 0; i < s->entry_count; i++) {
 846                a = s->entries[i];
 847                b = new_tree_entry();
 848                memcpy(b, a, sizeof(*a));
 849                if (a->tree && is_null_sha1(b->versions[1].sha1))
 850                        b->tree = dup_tree_content(a->tree);
 851                else
 852                        b->tree = NULL;
 853                d->entries[i] = b;
 854        }
 855        d->entry_count = s->entry_count;
 856        d->delta_depth = s->delta_depth;
 857
 858        return d;
 859}
 860
 861static void start_packfile(void)
 862{
 863        static char tmpfile[PATH_MAX];
 864        struct packed_git *p;
 865        struct pack_header hdr;
 866        int pack_fd;
 867
 868        pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 869                              "pack/tmp_pack_XXXXXX");
 870        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 871        strcpy(p->pack_name, tmpfile);
 872        p->pack_fd = pack_fd;
 873        pack_file = sha1fd(pack_fd, p->pack_name);
 874
 875        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 876        hdr.hdr_version = htonl(2);
 877        hdr.hdr_entries = 0;
 878        sha1write(pack_file, &hdr, sizeof(hdr));
 879
 880        pack_data = p;
 881        pack_size = sizeof(hdr);
 882        object_count = 0;
 883
 884        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 885        all_packs[pack_id] = p;
 886}
 887
 888static const char *create_index(void)
 889{
 890        const char *tmpfile;
 891        struct pack_idx_entry **idx, **c, **last;
 892        struct object_entry *e;
 893        struct object_entry_pool *o;
 894
 895        /* Build the table of object IDs. */
 896        idx = xmalloc(object_count * sizeof(*idx));
 897        c = idx;
 898        for (o = blocks; o; o = o->next_pool)
 899                for (e = o->next_free; e-- != o->entries;)
 900                        if (pack_id == e->pack_id)
 901                                *c++ = &e->idx;
 902        last = idx + object_count;
 903        if (c != last)
 904                die("internal consistency error creating the index");
 905
 906        tmpfile = write_idx_file(NULL, idx, object_count, pack_data->sha1);
 907        free(idx);
 908        return tmpfile;
 909}
 910
 911static char *keep_pack(const char *curr_index_name)
 912{
 913        static char name[PATH_MAX];
 914        static const char *keep_msg = "fast-import";
 915        int keep_fd;
 916
 917        keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
 918        if (keep_fd < 0)
 919                die_errno("cannot create keep file");
 920        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 921        if (close(keep_fd))
 922                die_errno("failed to write keep file");
 923
 924        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 925                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 926        if (move_temp_to_file(pack_data->pack_name, name))
 927                die("cannot store pack file");
 928
 929        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 930                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 931        if (move_temp_to_file(curr_index_name, name))
 932                die("cannot store index file");
 933        free((void *)curr_index_name);
 934        return name;
 935}
 936
 937static void unkeep_all_packs(void)
 938{
 939        static char name[PATH_MAX];
 940        int k;
 941
 942        for (k = 0; k < pack_id; k++) {
 943                struct packed_git *p = all_packs[k];
 944                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 945                         get_object_directory(), sha1_to_hex(p->sha1));
 946                unlink_or_warn(name);
 947        }
 948}
 949
 950static void end_packfile(void)
 951{
 952        struct packed_git *old_p = pack_data, *new_p;
 953
 954        clear_delta_base_cache();
 955        if (object_count) {
 956                unsigned char cur_pack_sha1[20];
 957                char *idx_name;
 958                int i;
 959                struct branch *b;
 960                struct tag *t;
 961
 962                close_pack_windows(pack_data);
 963                sha1close(pack_file, cur_pack_sha1, 0);
 964                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 965                                    pack_data->pack_name, object_count,
 966                                    cur_pack_sha1, pack_size);
 967                close(pack_data->pack_fd);
 968                idx_name = keep_pack(create_index());
 969
 970                /* Register the packfile with core git's machinery. */
 971                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 972                if (!new_p)
 973                        die("core git rejected index %s", idx_name);
 974                all_packs[pack_id] = new_p;
 975                install_packed_git(new_p);
 976
 977                /* Print the boundary */
 978                if (pack_edges) {
 979                        fprintf(pack_edges, "%s:", new_p->pack_name);
 980                        for (i = 0; i < branch_table_sz; i++) {
 981                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 982                                        if (b->pack_id == pack_id)
 983                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 984                                }
 985                        }
 986                        for (t = first_tag; t; t = t->next_tag) {
 987                                if (t->pack_id == pack_id)
 988                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 989                        }
 990                        fputc('\n', pack_edges);
 991                        fflush(pack_edges);
 992                }
 993
 994                pack_id++;
 995        }
 996        else {
 997                close(old_p->pack_fd);
 998                unlink_or_warn(old_p->pack_name);
 999        }
1000        free(old_p);
1001
1002        /* We can't carry a delta across packfiles. */
1003        strbuf_release(&last_blob.data);
1004        last_blob.offset = 0;
1005        last_blob.depth = 0;
1006}
1007
1008static void cycle_packfile(void)
1009{
1010        end_packfile();
1011        start_packfile();
1012}
1013
1014static int store_object(
1015        enum object_type type,
1016        struct strbuf *dat,
1017        struct last_object *last,
1018        unsigned char *sha1out,
1019        uintmax_t mark)
1020{
1021        void *out, *delta;
1022        struct object_entry *e;
1023        unsigned char hdr[96];
1024        unsigned char sha1[20];
1025        unsigned long hdrlen, deltalen;
1026        git_SHA_CTX c;
1027        z_stream s;
1028
1029        hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1030                (unsigned long)dat->len) + 1;
1031        git_SHA1_Init(&c);
1032        git_SHA1_Update(&c, hdr, hdrlen);
1033        git_SHA1_Update(&c, dat->buf, dat->len);
1034        git_SHA1_Final(sha1, &c);
1035        if (sha1out)
1036                hashcpy(sha1out, sha1);
1037
1038        e = insert_object(sha1);
1039        if (mark)
1040                insert_mark(mark, e);
1041        if (e->idx.offset) {
1042                duplicate_count_by_type[type]++;
1043                return 1;
1044        } else if (find_sha1_pack(sha1, packed_git)) {
1045                e->type = type;
1046                e->pack_id = MAX_PACK_ID;
1047                e->idx.offset = 1; /* just not zero! */
1048                duplicate_count_by_type[type]++;
1049                return 1;
1050        }
1051
1052        if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1053                delta = diff_delta(last->data.buf, last->data.len,
1054                        dat->buf, dat->len,
1055                        &deltalen, dat->len - 20);
1056        } else
1057                delta = NULL;
1058
1059        memset(&s, 0, sizeof(s));
1060        deflateInit(&s, pack_compression_level);
1061        if (delta) {
1062                s.next_in = delta;
1063                s.avail_in = deltalen;
1064        } else {
1065                s.next_in = (void *)dat->buf;
1066                s.avail_in = dat->len;
1067        }
1068        s.avail_out = deflateBound(&s, s.avail_in);
1069        s.next_out = out = xmalloc(s.avail_out);
1070        while (deflate(&s, Z_FINISH) == Z_OK)
1071                /* nothing */;
1072        deflateEnd(&s);
1073
1074        /* Determine if we should auto-checkpoint. */
1075        if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1076                || (pack_size + 60 + s.total_out) < pack_size) {
1077
1078                /* This new object needs to *not* have the current pack_id. */
1079                e->pack_id = pack_id + 1;
1080                cycle_packfile();
1081
1082                /* We cannot carry a delta into the new pack. */
1083                if (delta) {
1084                        free(delta);
1085                        delta = NULL;
1086
1087                        memset(&s, 0, sizeof(s));
1088                        deflateInit(&s, pack_compression_level);
1089                        s.next_in = (void *)dat->buf;
1090                        s.avail_in = dat->len;
1091                        s.avail_out = deflateBound(&s, s.avail_in);
1092                        s.next_out = out = xrealloc(out, s.avail_out);
1093                        while (deflate(&s, Z_FINISH) == Z_OK)
1094                                /* nothing */;
1095                        deflateEnd(&s);
1096                }
1097        }
1098
1099        e->type = type;
1100        e->pack_id = pack_id;
1101        e->idx.offset = pack_size;
1102        object_count++;
1103        object_count_by_type[type]++;
1104
1105        crc32_begin(pack_file);
1106
1107        if (delta) {
1108                off_t ofs = e->idx.offset - last->offset;
1109                unsigned pos = sizeof(hdr) - 1;
1110
1111                delta_count_by_type[type]++;
1112                e->depth = last->depth + 1;
1113
1114                hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1115                sha1write(pack_file, hdr, hdrlen);
1116                pack_size += hdrlen;
1117
1118                hdr[pos] = ofs & 127;
1119                while (ofs >>= 7)
1120                        hdr[--pos] = 128 | (--ofs & 127);
1121                sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1122                pack_size += sizeof(hdr) - pos;
1123        } else {
1124                e->depth = 0;
1125                hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1126                sha1write(pack_file, hdr, hdrlen);
1127                pack_size += hdrlen;
1128        }
1129
1130        sha1write(pack_file, out, s.total_out);
1131        pack_size += s.total_out;
1132
1133        e->idx.crc32 = crc32_end(pack_file);
1134
1135        free(out);
1136        free(delta);
1137        if (last) {
1138                if (last->no_swap) {
1139                        last->data = *dat;
1140                } else {
1141                        strbuf_swap(&last->data, dat);
1142                }
1143                last->offset = e->idx.offset;
1144                last->depth = e->depth;
1145        }
1146        return 0;
1147}
1148
1149static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1150{
1151        if (ftruncate(pack_data->pack_fd, to)
1152         || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1153                die_errno("cannot truncate pack to skip duplicate");
1154        pack_size = to;
1155
1156        /* yes this is a layering violation */
1157        pack_file->total = to;
1158        pack_file->offset = 0;
1159        pack_file->ctx = *ctx;
1160}
1161
1162static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1163{
1164        size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1165        unsigned char *in_buf = xmalloc(in_sz);
1166        unsigned char *out_buf = xmalloc(out_sz);
1167        struct object_entry *e;
1168        unsigned char sha1[20];
1169        unsigned long hdrlen;
1170        off_t offset;
1171        git_SHA_CTX c;
1172        git_SHA_CTX pack_file_ctx;
1173        z_stream s;
1174        int status = Z_OK;
1175
1176        /* Determine if we should auto-checkpoint. */
1177        if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1178                || (pack_size + 60 + len) < pack_size)
1179                cycle_packfile();
1180
1181        offset = pack_size;
1182
1183        /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1184        sha1flush(pack_file);
1185        pack_file_ctx = pack_file->ctx;
1186
1187        hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1188        if (out_sz <= hdrlen)
1189                die("impossibly large object header");
1190
1191        git_SHA1_Init(&c);
1192        git_SHA1_Update(&c, out_buf, hdrlen);
1193
1194        crc32_begin(pack_file);
1195
1196        memset(&s, 0, sizeof(s));
1197        deflateInit(&s, pack_compression_level);
1198
1199        hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1200        if (out_sz <= hdrlen)
1201                die("impossibly large object header");
1202
1203        s.next_out = out_buf + hdrlen;
1204        s.avail_out = out_sz - hdrlen;
1205
1206        while (status != Z_STREAM_END) {
1207                if (0 < len && !s.avail_in) {
1208                        size_t cnt = in_sz < len ? in_sz : (size_t)len;
1209                        size_t n = fread(in_buf, 1, cnt, stdin);
1210                        if (!n && feof(stdin))
1211                                die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1212
1213                        git_SHA1_Update(&c, in_buf, n);
1214                        s.next_in = in_buf;
1215                        s.avail_in = n;
1216                        len -= n;
1217                }
1218
1219                status = deflate(&s, len ? 0 : Z_FINISH);
1220
1221                if (!s.avail_out || status == Z_STREAM_END) {
1222                        size_t n = s.next_out - out_buf;
1223                        sha1write(pack_file, out_buf, n);
1224                        pack_size += n;
1225                        s.next_out = out_buf;
1226                        s.avail_out = out_sz;
1227                }
1228
1229                switch (status) {
1230                case Z_OK:
1231                case Z_BUF_ERROR:
1232                case Z_STREAM_END:
1233                        continue;
1234                default:
1235                        die("unexpected deflate failure: %d", status);
1236                }
1237        }
1238        deflateEnd(&s);
1239        git_SHA1_Final(sha1, &c);
1240
1241        if (sha1out)
1242                hashcpy(sha1out, sha1);
1243
1244        e = insert_object(sha1);
1245
1246        if (mark)
1247                insert_mark(mark, e);
1248
1249        if (e->idx.offset) {
1250                duplicate_count_by_type[OBJ_BLOB]++;
1251                truncate_pack(offset, &pack_file_ctx);
1252
1253        } else if (find_sha1_pack(sha1, packed_git)) {
1254                e->type = OBJ_BLOB;
1255                e->pack_id = MAX_PACK_ID;
1256                e->idx.offset = 1; /* just not zero! */
1257                duplicate_count_by_type[OBJ_BLOB]++;
1258                truncate_pack(offset, &pack_file_ctx);
1259
1260        } else {
1261                e->depth = 0;
1262                e->type = OBJ_BLOB;
1263                e->pack_id = pack_id;
1264                e->idx.offset = offset;
1265                e->idx.crc32 = crc32_end(pack_file);
1266                object_count++;
1267                object_count_by_type[OBJ_BLOB]++;
1268        }
1269
1270        free(in_buf);
1271        free(out_buf);
1272}
1273
1274/* All calls must be guarded by find_object() or find_mark() to
1275 * ensure the 'struct object_entry' passed was written by this
1276 * process instance.  We unpack the entry by the offset, avoiding
1277 * the need for the corresponding .idx file.  This unpacking rule
1278 * works because we only use OBJ_REF_DELTA within the packfiles
1279 * created by fast-import.
1280 *
1281 * oe must not be NULL.  Such an oe usually comes from giving
1282 * an unknown SHA-1 to find_object() or an undefined mark to
1283 * find_mark().  Callers must test for this condition and use
1284 * the standard read_sha1_file() when it happens.
1285 *
1286 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1287 * find_mark(), where the mark was reloaded from an existing marks
1288 * file and is referencing an object that this fast-import process
1289 * instance did not write out to a packfile.  Callers must test for
1290 * this condition and use read_sha1_file() instead.
1291 */
1292static void *gfi_unpack_entry(
1293        struct object_entry *oe,
1294        unsigned long *sizep)
1295{
1296        enum object_type type;
1297        struct packed_git *p = all_packs[oe->pack_id];
1298        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1299                /* The object is stored in the packfile we are writing to
1300                 * and we have modified it since the last time we scanned
1301                 * back to read a previously written object.  If an old
1302                 * window covered [p->pack_size, p->pack_size + 20) its
1303                 * data is stale and is not valid.  Closing all windows
1304                 * and updating the packfile length ensures we can read
1305                 * the newly written data.
1306                 */
1307                close_pack_windows(p);
1308                sha1flush(pack_file);
1309
1310                /* We have to offer 20 bytes additional on the end of
1311                 * the packfile as the core unpacker code assumes the
1312                 * footer is present at the file end and must promise
1313                 * at least 20 bytes within any window it maps.  But
1314                 * we don't actually create the footer here.
1315                 */
1316                p->pack_size = pack_size + 20;
1317        }
1318        return unpack_entry(p, oe->idx.offset, &type, sizep);
1319}
1320
1321static const char *get_mode(const char *str, uint16_t *modep)
1322{
1323        unsigned char c;
1324        uint16_t mode = 0;
1325
1326        while ((c = *str++) != ' ') {
1327                if (c < '0' || c > '7')
1328                        return NULL;
1329                mode = (mode << 3) + (c - '0');
1330        }
1331        *modep = mode;
1332        return str;
1333}
1334
1335static void load_tree(struct tree_entry *root)
1336{
1337        unsigned char *sha1 = root->versions[1].sha1;
1338        struct object_entry *myoe;
1339        struct tree_content *t;
1340        unsigned long size;
1341        char *buf;
1342        const char *c;
1343
1344        root->tree = t = new_tree_content(8);
1345        if (is_null_sha1(sha1))
1346                return;
1347
1348        myoe = find_object(sha1);
1349        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1350                if (myoe->type != OBJ_TREE)
1351                        die("Not a tree: %s", sha1_to_hex(sha1));
1352                t->delta_depth = myoe->depth;
1353                buf = gfi_unpack_entry(myoe, &size);
1354                if (!buf)
1355                        die("Can't load tree %s", sha1_to_hex(sha1));
1356        } else {
1357                enum object_type type;
1358                buf = read_sha1_file(sha1, &type, &size);
1359                if (!buf || type != OBJ_TREE)
1360                        die("Can't load tree %s", sha1_to_hex(sha1));
1361        }
1362
1363        c = buf;
1364        while (c != (buf + size)) {
1365                struct tree_entry *e = new_tree_entry();
1366
1367                if (t->entry_count == t->entry_capacity)
1368                        root->tree = t = grow_tree_content(t, t->entry_count);
1369                t->entries[t->entry_count++] = e;
1370
1371                e->tree = NULL;
1372                c = get_mode(c, &e->versions[1].mode);
1373                if (!c)
1374                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1375                e->versions[0].mode = e->versions[1].mode;
1376                e->name = to_atom(c, strlen(c));
1377                c += e->name->str_len + 1;
1378                hashcpy(e->versions[0].sha1, (unsigned char *)c);
1379                hashcpy(e->versions[1].sha1, (unsigned char *)c);
1380                c += 20;
1381        }
1382        free(buf);
1383}
1384
1385static int tecmp0 (const void *_a, const void *_b)
1386{
1387        struct tree_entry *a = *((struct tree_entry**)_a);
1388        struct tree_entry *b = *((struct tree_entry**)_b);
1389        return base_name_compare(
1390                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1391                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1392}
1393
1394static int tecmp1 (const void *_a, const void *_b)
1395{
1396        struct tree_entry *a = *((struct tree_entry**)_a);
1397        struct tree_entry *b = *((struct tree_entry**)_b);
1398        return base_name_compare(
1399                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1400                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1401}
1402
1403static void mktree(struct tree_content *t, int v, struct strbuf *b)
1404{
1405        size_t maxlen = 0;
1406        unsigned int i;
1407
1408        if (!v)
1409                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1410        else
1411                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1412
1413        for (i = 0; i < t->entry_count; i++) {
1414                if (t->entries[i]->versions[v].mode)
1415                        maxlen += t->entries[i]->name->str_len + 34;
1416        }
1417
1418        strbuf_reset(b);
1419        strbuf_grow(b, maxlen);
1420        for (i = 0; i < t->entry_count; i++) {
1421                struct tree_entry *e = t->entries[i];
1422                if (!e->versions[v].mode)
1423                        continue;
1424                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1425                                        e->name->str_dat, '\0');
1426                strbuf_add(b, e->versions[v].sha1, 20);
1427        }
1428}
1429
1430static void store_tree(struct tree_entry *root)
1431{
1432        struct tree_content *t = root->tree;
1433        unsigned int i, j, del;
1434        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1435        struct object_entry *le;
1436
1437        if (!is_null_sha1(root->versions[1].sha1))
1438                return;
1439
1440        for (i = 0; i < t->entry_count; i++) {
1441                if (t->entries[i]->tree)
1442                        store_tree(t->entries[i]);
1443        }
1444
1445        le = find_object(root->versions[0].sha1);
1446        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1447                mktree(t, 0, &old_tree);
1448                lo.data = old_tree;
1449                lo.offset = le->idx.offset;
1450                lo.depth = t->delta_depth;
1451        }
1452
1453        mktree(t, 1, &new_tree);
1454        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1455
1456        t->delta_depth = lo.depth;
1457        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1458                struct tree_entry *e = t->entries[i];
1459                if (e->versions[1].mode) {
1460                        e->versions[0].mode = e->versions[1].mode;
1461                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1462                        t->entries[j++] = e;
1463                } else {
1464                        release_tree_entry(e);
1465                        del++;
1466                }
1467        }
1468        t->entry_count -= del;
1469}
1470
1471static void tree_content_replace(
1472        struct tree_entry *root,
1473        const unsigned char *sha1,
1474        const uint16_t mode,
1475        struct tree_content *newtree)
1476{
1477        if (!S_ISDIR(mode))
1478                die("Root cannot be a non-directory");
1479        hashcpy(root->versions[1].sha1, sha1);
1480        if (root->tree)
1481                release_tree_content_recursive(root->tree);
1482        root->tree = newtree;
1483}
1484
1485static int tree_content_set(
1486        struct tree_entry *root,
1487        const char *p,
1488        const unsigned char *sha1,
1489        const uint16_t mode,
1490        struct tree_content *subtree)
1491{
1492        struct tree_content *t;
1493        const char *slash1;
1494        unsigned int i, n;
1495        struct tree_entry *e;
1496
1497        slash1 = strchr(p, '/');
1498        if (slash1)
1499                n = slash1 - p;
1500        else
1501                n = strlen(p);
1502        if (!n)
1503                die("Empty path component found in input");
1504        if (!slash1 && !S_ISDIR(mode) && subtree)
1505                die("Non-directories cannot have subtrees");
1506
1507        if (!root->tree)
1508                load_tree(root);
1509        t = root->tree;
1510        for (i = 0; i < t->entry_count; i++) {
1511                e = t->entries[i];
1512                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1513                        if (!slash1) {
1514                                if (!S_ISDIR(mode)
1515                                                && e->versions[1].mode == mode
1516                                                && !hashcmp(e->versions[1].sha1, sha1))
1517                                        return 0;
1518                                e->versions[1].mode = mode;
1519                                hashcpy(e->versions[1].sha1, sha1);
1520                                if (e->tree)
1521                                        release_tree_content_recursive(e->tree);
1522                                e->tree = subtree;
1523                                hashclr(root->versions[1].sha1);
1524                                return 1;
1525                        }
1526                        if (!S_ISDIR(e->versions[1].mode)) {
1527                                e->tree = new_tree_content(8);
1528                                e->versions[1].mode = S_IFDIR;
1529                        }
1530                        if (!e->tree)
1531                                load_tree(e);
1532                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1533                                hashclr(root->versions[1].sha1);
1534                                return 1;
1535                        }
1536                        return 0;
1537                }
1538        }
1539
1540        if (t->entry_count == t->entry_capacity)
1541                root->tree = t = grow_tree_content(t, t->entry_count);
1542        e = new_tree_entry();
1543        e->name = to_atom(p, n);
1544        e->versions[0].mode = 0;
1545        hashclr(e->versions[0].sha1);
1546        t->entries[t->entry_count++] = e;
1547        if (slash1) {
1548                e->tree = new_tree_content(8);
1549                e->versions[1].mode = S_IFDIR;
1550                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1551        } else {
1552                e->tree = subtree;
1553                e->versions[1].mode = mode;
1554                hashcpy(e->versions[1].sha1, sha1);
1555        }
1556        hashclr(root->versions[1].sha1);
1557        return 1;
1558}
1559
1560static int tree_content_remove(
1561        struct tree_entry *root,
1562        const char *p,
1563        struct tree_entry *backup_leaf)
1564{
1565        struct tree_content *t;
1566        const char *slash1;
1567        unsigned int i, n;
1568        struct tree_entry *e;
1569
1570        slash1 = strchr(p, '/');
1571        if (slash1)
1572                n = slash1 - p;
1573        else
1574                n = strlen(p);
1575
1576        if (!root->tree)
1577                load_tree(root);
1578        t = root->tree;
1579        for (i = 0; i < t->entry_count; i++) {
1580                e = t->entries[i];
1581                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1582                        if (slash1 && !S_ISDIR(e->versions[1].mode))
1583                                /*
1584                                 * If p names a file in some subdirectory, and a
1585                                 * file or symlink matching the name of the
1586                                 * parent directory of p exists, then p cannot
1587                                 * exist and need not be deleted.
1588                                 */
1589                                return 1;
1590                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1591                                goto del_entry;
1592                        if (!e->tree)
1593                                load_tree(e);
1594                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1595                                for (n = 0; n < e->tree->entry_count; n++) {
1596                                        if (e->tree->entries[n]->versions[1].mode) {
1597                                                hashclr(root->versions[1].sha1);
1598                                                return 1;
1599                                        }
1600                                }
1601                                backup_leaf = NULL;
1602                                goto del_entry;
1603                        }
1604                        return 0;
1605                }
1606        }
1607        return 0;
1608
1609del_entry:
1610        if (backup_leaf)
1611                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1612        else if (e->tree)
1613                release_tree_content_recursive(e->tree);
1614        e->tree = NULL;
1615        e->versions[1].mode = 0;
1616        hashclr(e->versions[1].sha1);
1617        hashclr(root->versions[1].sha1);
1618        return 1;
1619}
1620
1621static int tree_content_get(
1622        struct tree_entry *root,
1623        const char *p,
1624        struct tree_entry *leaf)
1625{
1626        struct tree_content *t;
1627        const char *slash1;
1628        unsigned int i, n;
1629        struct tree_entry *e;
1630
1631        slash1 = strchr(p, '/');
1632        if (slash1)
1633                n = slash1 - p;
1634        else
1635                n = strlen(p);
1636
1637        if (!root->tree)
1638                load_tree(root);
1639        t = root->tree;
1640        for (i = 0; i < t->entry_count; i++) {
1641                e = t->entries[i];
1642                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1643                        if (!slash1) {
1644                                memcpy(leaf, e, sizeof(*leaf));
1645                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1646                                        leaf->tree = dup_tree_content(e->tree);
1647                                else
1648                                        leaf->tree = NULL;
1649                                return 1;
1650                        }
1651                        if (!S_ISDIR(e->versions[1].mode))
1652                                return 0;
1653                        if (!e->tree)
1654                                load_tree(e);
1655                        return tree_content_get(e, slash1 + 1, leaf);
1656                }
1657        }
1658        return 0;
1659}
1660
1661static int update_branch(struct branch *b)
1662{
1663        static const char *msg = "fast-import";
1664        struct ref_lock *lock;
1665        unsigned char old_sha1[20];
1666
1667        if (is_null_sha1(b->sha1))
1668                return 0;
1669        if (read_ref(b->name, old_sha1))
1670                hashclr(old_sha1);
1671        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1672        if (!lock)
1673                return error("Unable to lock %s", b->name);
1674        if (!force_update && !is_null_sha1(old_sha1)) {
1675                struct commit *old_cmit, *new_cmit;
1676
1677                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1678                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1679                if (!old_cmit || !new_cmit) {
1680                        unlock_ref(lock);
1681                        return error("Branch %s is missing commits.", b->name);
1682                }
1683
1684                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1685                        unlock_ref(lock);
1686                        warning("Not updating %s"
1687                                " (new tip %s does not contain %s)",
1688                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1689                        return -1;
1690                }
1691        }
1692        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1693                return error("Unable to update %s", b->name);
1694        return 0;
1695}
1696
1697static void dump_branches(void)
1698{
1699        unsigned int i;
1700        struct branch *b;
1701
1702        for (i = 0; i < branch_table_sz; i++) {
1703                for (b = branch_table[i]; b; b = b->table_next_branch)
1704                        failure |= update_branch(b);
1705        }
1706}
1707
1708static void dump_tags(void)
1709{
1710        static const char *msg = "fast-import";
1711        struct tag *t;
1712        struct ref_lock *lock;
1713        char ref_name[PATH_MAX];
1714
1715        for (t = first_tag; t; t = t->next_tag) {
1716                sprintf(ref_name, "tags/%s", t->name);
1717                lock = lock_ref_sha1(ref_name, NULL);
1718                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1719                        failure |= error("Unable to update %s", ref_name);
1720        }
1721}
1722
1723static void dump_marks_helper(FILE *f,
1724        uintmax_t base,
1725        struct mark_set *m)
1726{
1727        uintmax_t k;
1728        if (m->shift) {
1729                for (k = 0; k < 1024; k++) {
1730                        if (m->data.sets[k])
1731                                dump_marks_helper(f, base + (k << m->shift),
1732                                        m->data.sets[k]);
1733                }
1734        } else {
1735                for (k = 0; k < 1024; k++) {
1736                        if (m->data.marked[k])
1737                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1738                                        sha1_to_hex(m->data.marked[k]->idx.sha1));
1739                }
1740        }
1741}
1742
1743static void dump_marks(void)
1744{
1745        static struct lock_file mark_lock;
1746        int mark_fd;
1747        FILE *f;
1748
1749        if (!export_marks_file)
1750                return;
1751
1752        mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1753        if (mark_fd < 0) {
1754                failure |= error("Unable to write marks file %s: %s",
1755                        export_marks_file, strerror(errno));
1756                return;
1757        }
1758
1759        f = fdopen(mark_fd, "w");
1760        if (!f) {
1761                int saved_errno = errno;
1762                rollback_lock_file(&mark_lock);
1763                failure |= error("Unable to write marks file %s: %s",
1764                        export_marks_file, strerror(saved_errno));
1765                return;
1766        }
1767
1768        /*
1769         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1770         * Assign -1 to the lock file descriptor so that commit_lock_file()
1771         * won't try to close() it.
1772         */
1773        mark_lock.fd = -1;
1774
1775        dump_marks_helper(f, 0, marks);
1776        if (ferror(f) || fclose(f)) {
1777                int saved_errno = errno;
1778                rollback_lock_file(&mark_lock);
1779                failure |= error("Unable to write marks file %s: %s",
1780                        export_marks_file, strerror(saved_errno));
1781                return;
1782        }
1783
1784        if (commit_lock_file(&mark_lock)) {
1785                int saved_errno = errno;
1786                rollback_lock_file(&mark_lock);
1787                failure |= error("Unable to commit marks file %s: %s",
1788                        export_marks_file, strerror(saved_errno));
1789                return;
1790        }
1791}
1792
1793static void read_marks(void)
1794{
1795        char line[512];
1796        FILE *f = fopen(import_marks_file, "r");
1797        if (!f)
1798                die_errno("cannot read '%s'", import_marks_file);
1799        while (fgets(line, sizeof(line), f)) {
1800                uintmax_t mark;
1801                char *end;
1802                unsigned char sha1[20];
1803                struct object_entry *e;
1804
1805                end = strchr(line, '\n');
1806                if (line[0] != ':' || !end)
1807                        die("corrupt mark line: %s", line);
1808                *end = 0;
1809                mark = strtoumax(line + 1, &end, 10);
1810                if (!mark || end == line + 1
1811                        || *end != ' ' || get_sha1(end + 1, sha1))
1812                        die("corrupt mark line: %s", line);
1813                e = find_object(sha1);
1814                if (!e) {
1815                        enum object_type type = sha1_object_info(sha1, NULL);
1816                        if (type < 0)
1817                                die("object not found: %s", sha1_to_hex(sha1));
1818                        e = insert_object(sha1);
1819                        e->type = type;
1820                        e->pack_id = MAX_PACK_ID;
1821                        e->idx.offset = 1; /* just not zero! */
1822                }
1823                insert_mark(mark, e);
1824        }
1825        fclose(f);
1826}
1827
1828
1829static int read_next_command(void)
1830{
1831        static int stdin_eof = 0;
1832
1833        if (stdin_eof) {
1834                unread_command_buf = 0;
1835                return EOF;
1836        }
1837
1838        for (;;) {
1839                if (unread_command_buf) {
1840                        unread_command_buf = 0;
1841                } else {
1842                        struct recent_command *rc;
1843
1844                        strbuf_detach(&command_buf, NULL);
1845                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1846                        if (stdin_eof)
1847                                return EOF;
1848
1849                        if (!seen_data_command
1850                                && prefixcmp(command_buf.buf, "feature ")
1851                                && prefixcmp(command_buf.buf, "option ")) {
1852                                parse_argv();
1853                        }
1854
1855                        rc = rc_free;
1856                        if (rc)
1857                                rc_free = rc->next;
1858                        else {
1859                                rc = cmd_hist.next;
1860                                cmd_hist.next = rc->next;
1861                                cmd_hist.next->prev = &cmd_hist;
1862                                free(rc->buf);
1863                        }
1864
1865                        rc->buf = command_buf.buf;
1866                        rc->prev = cmd_tail;
1867                        rc->next = cmd_hist.prev;
1868                        rc->prev->next = rc;
1869                        cmd_tail = rc;
1870                }
1871                if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1872                        parse_cat_blob();
1873                        continue;
1874                }
1875                if (command_buf.buf[0] == '#')
1876                        continue;
1877                return 0;
1878        }
1879}
1880
1881static void skip_optional_lf(void)
1882{
1883        int term_char = fgetc(stdin);
1884        if (term_char != '\n' && term_char != EOF)
1885                ungetc(term_char, stdin);
1886}
1887
1888static void parse_mark(void)
1889{
1890        if (!prefixcmp(command_buf.buf, "mark :")) {
1891                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1892                read_next_command();
1893        }
1894        else
1895                next_mark = 0;
1896}
1897
1898static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1899{
1900        strbuf_reset(sb);
1901
1902        if (prefixcmp(command_buf.buf, "data "))
1903                die("Expected 'data n' command, found: %s", command_buf.buf);
1904
1905        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1906                char *term = xstrdup(command_buf.buf + 5 + 2);
1907                size_t term_len = command_buf.len - 5 - 2;
1908
1909                strbuf_detach(&command_buf, NULL);
1910                for (;;) {
1911                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1912                                die("EOF in data (terminator '%s' not found)", term);
1913                        if (term_len == command_buf.len
1914                                && !strcmp(term, command_buf.buf))
1915                                break;
1916                        strbuf_addbuf(sb, &command_buf);
1917                        strbuf_addch(sb, '\n');
1918                }
1919                free(term);
1920        }
1921        else {
1922                uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1923                size_t n = 0, length = (size_t)len;
1924
1925                if (limit && limit < len) {
1926                        *len_res = len;
1927                        return 0;
1928                }
1929                if (length < len)
1930                        die("data is too large to use in this context");
1931
1932                while (n < length) {
1933                        size_t s = strbuf_fread(sb, length - n, stdin);
1934                        if (!s && feof(stdin))
1935                                die("EOF in data (%lu bytes remaining)",
1936                                        (unsigned long)(length - n));
1937                        n += s;
1938                }
1939        }
1940
1941        skip_optional_lf();
1942        return 1;
1943}
1944
1945static int validate_raw_date(const char *src, char *result, int maxlen)
1946{
1947        const char *orig_src = src;
1948        char *endp;
1949        unsigned long num;
1950
1951        errno = 0;
1952
1953        num = strtoul(src, &endp, 10);
1954        /* NEEDSWORK: perhaps check for reasonable values? */
1955        if (errno || endp == src || *endp != ' ')
1956                return -1;
1957
1958        src = endp + 1;
1959        if (*src != '-' && *src != '+')
1960                return -1;
1961
1962        num = strtoul(src + 1, &endp, 10);
1963        if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1964            1400 < num)
1965                return -1;
1966
1967        strcpy(result, orig_src);
1968        return 0;
1969}
1970
1971static char *parse_ident(const char *buf)
1972{
1973        const char *gt;
1974        size_t name_len;
1975        char *ident;
1976
1977        gt = strrchr(buf, '>');
1978        if (!gt)
1979                die("Missing > in ident string: %s", buf);
1980        gt++;
1981        if (*gt != ' ')
1982                die("Missing space after > in ident string: %s", buf);
1983        gt++;
1984        name_len = gt - buf;
1985        ident = xmalloc(name_len + 24);
1986        strncpy(ident, buf, name_len);
1987
1988        switch (whenspec) {
1989        case WHENSPEC_RAW:
1990                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1991                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1992                break;
1993        case WHENSPEC_RFC2822:
1994                if (parse_date(gt, ident + name_len, 24) < 0)
1995                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1996                break;
1997        case WHENSPEC_NOW:
1998                if (strcmp("now", gt))
1999                        die("Date in ident must be 'now': %s", buf);
2000                datestamp(ident + name_len, 24);
2001                break;
2002        }
2003
2004        return ident;
2005}
2006
2007static void parse_and_store_blob(
2008        struct last_object *last,
2009        unsigned char *sha1out,
2010        uintmax_t mark)
2011{
2012        static struct strbuf buf = STRBUF_INIT;
2013        uintmax_t len;
2014
2015        if (parse_data(&buf, big_file_threshold, &len))
2016                store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2017        else {
2018                if (last) {
2019                        strbuf_release(&last->data);
2020                        last->offset = 0;
2021                        last->depth = 0;
2022                }
2023                stream_blob(len, sha1out, mark);
2024                skip_optional_lf();
2025        }
2026}
2027
2028static void parse_new_blob(void)
2029{
2030        read_next_command();
2031        parse_mark();
2032        parse_and_store_blob(&last_blob, NULL, next_mark);
2033}
2034
2035static void unload_one_branch(void)
2036{
2037        while (cur_active_branches
2038                && cur_active_branches >= max_active_branches) {
2039                uintmax_t min_commit = ULONG_MAX;
2040                struct branch *e, *l = NULL, *p = NULL;
2041
2042                for (e = active_branches; e; e = e->active_next_branch) {
2043                        if (e->last_commit < min_commit) {
2044                                p = l;
2045                                min_commit = e->last_commit;
2046                        }
2047                        l = e;
2048                }
2049
2050                if (p) {
2051                        e = p->active_next_branch;
2052                        p->active_next_branch = e->active_next_branch;
2053                } else {
2054                        e = active_branches;
2055                        active_branches = e->active_next_branch;
2056                }
2057                e->active = 0;
2058                e->active_next_branch = NULL;
2059                if (e->branch_tree.tree) {
2060                        release_tree_content_recursive(e->branch_tree.tree);
2061                        e->branch_tree.tree = NULL;
2062                }
2063                cur_active_branches--;
2064        }
2065}
2066
2067static void load_branch(struct branch *b)
2068{
2069        load_tree(&b->branch_tree);
2070        if (!b->active) {
2071                b->active = 1;
2072                b->active_next_branch = active_branches;
2073                active_branches = b;
2074                cur_active_branches++;
2075                branch_load_count++;
2076        }
2077}
2078
2079static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2080{
2081        unsigned char fanout = 0;
2082        while ((num_notes >>= 8))
2083                fanout++;
2084        return fanout;
2085}
2086
2087static void construct_path_with_fanout(const char *hex_sha1,
2088                unsigned char fanout, char *path)
2089{
2090        unsigned int i = 0, j = 0;
2091        if (fanout >= 20)
2092                die("Too large fanout (%u)", fanout);
2093        while (fanout) {
2094                path[i++] = hex_sha1[j++];
2095                path[i++] = hex_sha1[j++];
2096                path[i++] = '/';
2097                fanout--;
2098        }
2099        memcpy(path + i, hex_sha1 + j, 40 - j);
2100        path[i + 40 - j] = '\0';
2101}
2102
2103static uintmax_t do_change_note_fanout(
2104                struct tree_entry *orig_root, struct tree_entry *root,
2105                char *hex_sha1, unsigned int hex_sha1_len,
2106                char *fullpath, unsigned int fullpath_len,
2107                unsigned char fanout)
2108{
2109        struct tree_content *t = root->tree;
2110        struct tree_entry *e, leaf;
2111        unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2112        uintmax_t num_notes = 0;
2113        unsigned char sha1[20];
2114        char realpath[60];
2115
2116        for (i = 0; t && i < t->entry_count; i++) {
2117                e = t->entries[i];
2118                tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2119                tmp_fullpath_len = fullpath_len;
2120
2121                /*
2122                 * We're interested in EITHER existing note entries (entries
2123                 * with exactly 40 hex chars in path, not including directory
2124                 * separators), OR directory entries that may contain note
2125                 * entries (with < 40 hex chars in path).
2126                 * Also, each path component in a note entry must be a multiple
2127                 * of 2 chars.
2128                 */
2129                if (!e->versions[1].mode ||
2130                    tmp_hex_sha1_len > 40 ||
2131                    e->name->str_len % 2)
2132                        continue;
2133
2134                /* This _may_ be a note entry, or a subdir containing notes */
2135                memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2136                       e->name->str_len);
2137                if (tmp_fullpath_len)
2138                        fullpath[tmp_fullpath_len++] = '/';
2139                memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2140                       e->name->str_len);
2141                tmp_fullpath_len += e->name->str_len;
2142                fullpath[tmp_fullpath_len] = '\0';
2143
2144                if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2145                        /* This is a note entry */
2146                        construct_path_with_fanout(hex_sha1, fanout, realpath);
2147                        if (!strcmp(fullpath, realpath)) {
2148                                /* Note entry is in correct location */
2149                                num_notes++;
2150                                continue;
2151                        }
2152
2153                        /* Rename fullpath to realpath */
2154                        if (!tree_content_remove(orig_root, fullpath, &leaf))
2155                                die("Failed to remove path %s", fullpath);
2156                        tree_content_set(orig_root, realpath,
2157                                leaf.versions[1].sha1,
2158                                leaf.versions[1].mode,
2159                                leaf.tree);
2160                } else if (S_ISDIR(e->versions[1].mode)) {
2161                        /* This is a subdir that may contain note entries */
2162                        if (!e->tree)
2163                                load_tree(e);
2164                        num_notes += do_change_note_fanout(orig_root, e,
2165                                hex_sha1, tmp_hex_sha1_len,
2166                                fullpath, tmp_fullpath_len, fanout);
2167                }
2168
2169                /* The above may have reallocated the current tree_content */
2170                t = root->tree;
2171        }
2172        return num_notes;
2173}
2174
2175static uintmax_t change_note_fanout(struct tree_entry *root,
2176                unsigned char fanout)
2177{
2178        char hex_sha1[40], path[60];
2179        return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2180}
2181
2182static void file_change_m(struct branch *b)
2183{
2184        const char *p = command_buf.buf + 2;
2185        static struct strbuf uq = STRBUF_INIT;
2186        const char *endp;
2187        struct object_entry *oe = oe;
2188        unsigned char sha1[20];
2189        uint16_t mode, inline_data = 0;
2190
2191        p = get_mode(p, &mode);
2192        if (!p)
2193                die("Corrupt mode: %s", command_buf.buf);
2194        switch (mode) {
2195        case 0644:
2196        case 0755:
2197                mode |= S_IFREG;
2198        case S_IFREG | 0644:
2199        case S_IFREG | 0755:
2200        case S_IFLNK:
2201        case S_IFDIR:
2202        case S_IFGITLINK:
2203                /* ok */
2204                break;
2205        default:
2206                die("Corrupt mode: %s", command_buf.buf);
2207        }
2208
2209        if (*p == ':') {
2210                char *x;
2211                oe = find_mark(strtoumax(p + 1, &x, 10));
2212                hashcpy(sha1, oe->idx.sha1);
2213                p = x;
2214        } else if (!prefixcmp(p, "inline")) {
2215                inline_data = 1;
2216                p += 6;
2217        } else {
2218                if (get_sha1_hex(p, sha1))
2219                        die("Invalid SHA1: %s", command_buf.buf);
2220                oe = find_object(sha1);
2221                p += 40;
2222        }
2223        if (*p++ != ' ')
2224                die("Missing space after SHA1: %s", command_buf.buf);
2225
2226        strbuf_reset(&uq);
2227        if (!unquote_c_style(&uq, p, &endp)) {
2228                if (*endp)
2229                        die("Garbage after path in: %s", command_buf.buf);
2230                p = uq.buf;
2231        }
2232
2233        /* Git does not track empty, non-toplevel directories. */
2234        if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2235                tree_content_remove(&b->branch_tree, p, NULL);
2236                return;
2237        }
2238
2239        if (S_ISGITLINK(mode)) {
2240                if (inline_data)
2241                        die("Git links cannot be specified 'inline': %s",
2242                                command_buf.buf);
2243                else if (oe) {
2244                        if (oe->type != OBJ_COMMIT)
2245                                die("Not a commit (actually a %s): %s",
2246                                        typename(oe->type), command_buf.buf);
2247                }
2248                /*
2249                 * Accept the sha1 without checking; it expected to be in
2250                 * another repository.
2251                 */
2252        } else if (inline_data) {
2253                if (S_ISDIR(mode))
2254                        die("Directories cannot be specified 'inline': %s",
2255                                command_buf.buf);
2256                if (p != uq.buf) {
2257                        strbuf_addstr(&uq, p);
2258                        p = uq.buf;
2259                }
2260                read_next_command();
2261                parse_and_store_blob(&last_blob, sha1, 0);
2262        } else {
2263                enum object_type expected = S_ISDIR(mode) ?
2264                                                OBJ_TREE: OBJ_BLOB;
2265                enum object_type type = oe ? oe->type :
2266                                        sha1_object_info(sha1, NULL);
2267                if (type < 0)
2268                        die("%s not found: %s",
2269                                        S_ISDIR(mode) ?  "Tree" : "Blob",
2270                                        command_buf.buf);
2271                if (type != expected)
2272                        die("Not a %s (actually a %s): %s",
2273                                typename(expected), typename(type),
2274                                command_buf.buf);
2275        }
2276
2277        if (!*p) {
2278                tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2279                return;
2280        }
2281        tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2282}
2283
2284static void file_change_d(struct branch *b)
2285{
2286        const char *p = command_buf.buf + 2;
2287        static struct strbuf uq = STRBUF_INIT;
2288        const char *endp;
2289
2290        strbuf_reset(&uq);
2291        if (!unquote_c_style(&uq, p, &endp)) {
2292                if (*endp)
2293                        die("Garbage after path in: %s", command_buf.buf);
2294                p = uq.buf;
2295        }
2296        tree_content_remove(&b->branch_tree, p, NULL);
2297}
2298
2299static void file_change_cr(struct branch *b, int rename)
2300{
2301        const char *s, *d;
2302        static struct strbuf s_uq = STRBUF_INIT;
2303        static struct strbuf d_uq = STRBUF_INIT;
2304        const char *endp;
2305        struct tree_entry leaf;
2306
2307        s = command_buf.buf + 2;
2308        strbuf_reset(&s_uq);
2309        if (!unquote_c_style(&s_uq, s, &endp)) {
2310                if (*endp != ' ')
2311                        die("Missing space after source: %s", command_buf.buf);
2312        } else {
2313                endp = strchr(s, ' ');
2314                if (!endp)
2315                        die("Missing space after source: %s", command_buf.buf);
2316                strbuf_add(&s_uq, s, endp - s);
2317        }
2318        s = s_uq.buf;
2319
2320        endp++;
2321        if (!*endp)
2322                die("Missing dest: %s", command_buf.buf);
2323
2324        d = endp;
2325        strbuf_reset(&d_uq);
2326        if (!unquote_c_style(&d_uq, d, &endp)) {
2327                if (*endp)
2328                        die("Garbage after dest in: %s", command_buf.buf);
2329                d = d_uq.buf;
2330        }
2331
2332        memset(&leaf, 0, sizeof(leaf));
2333        if (rename)
2334                tree_content_remove(&b->branch_tree, s, &leaf);
2335        else
2336                tree_content_get(&b->branch_tree, s, &leaf);
2337        if (!leaf.versions[1].mode)
2338                die("Path %s not in branch", s);
2339        if (!*d) {      /* C "path/to/subdir" "" */
2340                tree_content_replace(&b->branch_tree,
2341                        leaf.versions[1].sha1,
2342                        leaf.versions[1].mode,
2343                        leaf.tree);
2344                return;
2345        }
2346        tree_content_set(&b->branch_tree, d,
2347                leaf.versions[1].sha1,
2348                leaf.versions[1].mode,
2349                leaf.tree);
2350}
2351
2352static void note_change_n(struct branch *b, unsigned char old_fanout)
2353{
2354        const char *p = command_buf.buf + 2;
2355        static struct strbuf uq = STRBUF_INIT;
2356        struct object_entry *oe = oe;
2357        struct branch *s;
2358        unsigned char sha1[20], commit_sha1[20];
2359        char path[60];
2360        uint16_t inline_data = 0;
2361        unsigned char new_fanout;
2362
2363        /* <dataref> or 'inline' */
2364        if (*p == ':') {
2365                char *x;
2366                oe = find_mark(strtoumax(p + 1, &x, 10));
2367                hashcpy(sha1, oe->idx.sha1);
2368                p = x;
2369        } else if (!prefixcmp(p, "inline")) {
2370                inline_data = 1;
2371                p += 6;
2372        } else {
2373                if (get_sha1_hex(p, sha1))
2374                        die("Invalid SHA1: %s", command_buf.buf);
2375                oe = find_object(sha1);
2376                p += 40;
2377        }
2378        if (*p++ != ' ')
2379                die("Missing space after SHA1: %s", command_buf.buf);
2380
2381        /* <committish> */
2382        s = lookup_branch(p);
2383        if (s) {
2384                hashcpy(commit_sha1, s->sha1);
2385        } else if (*p == ':') {
2386                uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2387                struct object_entry *commit_oe = find_mark(commit_mark);
2388                if (commit_oe->type != OBJ_COMMIT)
2389                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2390                hashcpy(commit_sha1, commit_oe->idx.sha1);
2391        } else if (!get_sha1(p, commit_sha1)) {
2392                unsigned long size;
2393                char *buf = read_object_with_reference(commit_sha1,
2394                        commit_type, &size, commit_sha1);
2395                if (!buf || size < 46)
2396                        die("Not a valid commit: %s", p);
2397                free(buf);
2398        } else
2399                die("Invalid ref name or SHA1 expression: %s", p);
2400
2401        if (inline_data) {
2402                if (p != uq.buf) {
2403                        strbuf_addstr(&uq, p);
2404                        p = uq.buf;
2405                }
2406                read_next_command();
2407                parse_and_store_blob(&last_blob, sha1, 0);
2408        } else if (oe) {
2409                if (oe->type != OBJ_BLOB)
2410                        die("Not a blob (actually a %s): %s",
2411                                typename(oe->type), command_buf.buf);
2412        } else if (!is_null_sha1(sha1)) {
2413                enum object_type type = sha1_object_info(sha1, NULL);
2414                if (type < 0)
2415                        die("Blob not found: %s", command_buf.buf);
2416                if (type != OBJ_BLOB)
2417                        die("Not a blob (actually a %s): %s",
2418                            typename(type), command_buf.buf);
2419        }
2420
2421        construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2422        if (tree_content_remove(&b->branch_tree, path, NULL))
2423                b->num_notes--;
2424
2425        if (is_null_sha1(sha1))
2426                return; /* nothing to insert */
2427
2428        b->num_notes++;
2429        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2430        construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2431        tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2432}
2433
2434static void file_change_deleteall(struct branch *b)
2435{
2436        release_tree_content_recursive(b->branch_tree.tree);
2437        hashclr(b->branch_tree.versions[0].sha1);
2438        hashclr(b->branch_tree.versions[1].sha1);
2439        load_tree(&b->branch_tree);
2440        b->num_notes = 0;
2441}
2442
2443static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2444{
2445        if (!buf || size < 46)
2446                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2447        if (memcmp("tree ", buf, 5)
2448                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2449                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2450        hashcpy(b->branch_tree.versions[0].sha1,
2451                b->branch_tree.versions[1].sha1);
2452}
2453
2454static void parse_from_existing(struct branch *b)
2455{
2456        if (is_null_sha1(b->sha1)) {
2457                hashclr(b->branch_tree.versions[0].sha1);
2458                hashclr(b->branch_tree.versions[1].sha1);
2459        } else {
2460                unsigned long size;
2461                char *buf;
2462
2463                buf = read_object_with_reference(b->sha1,
2464                        commit_type, &size, b->sha1);
2465                parse_from_commit(b, buf, size);
2466                free(buf);
2467        }
2468}
2469
2470static int parse_from(struct branch *b)
2471{
2472        const char *from;
2473        struct branch *s;
2474
2475        if (prefixcmp(command_buf.buf, "from "))
2476                return 0;
2477
2478        if (b->branch_tree.tree) {
2479                release_tree_content_recursive(b->branch_tree.tree);
2480                b->branch_tree.tree = NULL;
2481        }
2482
2483        from = strchr(command_buf.buf, ' ') + 1;
2484        s = lookup_branch(from);
2485        if (b == s)
2486                die("Can't create a branch from itself: %s", b->name);
2487        else if (s) {
2488                unsigned char *t = s->branch_tree.versions[1].sha1;
2489                hashcpy(b->sha1, s->sha1);
2490                hashcpy(b->branch_tree.versions[0].sha1, t);
2491                hashcpy(b->branch_tree.versions[1].sha1, t);
2492        } else if (*from == ':') {
2493                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2494                struct object_entry *oe = find_mark(idnum);
2495                if (oe->type != OBJ_COMMIT)
2496                        die("Mark :%" PRIuMAX " not a commit", idnum);
2497                hashcpy(b->sha1, oe->idx.sha1);
2498                if (oe->pack_id != MAX_PACK_ID) {
2499                        unsigned long size;
2500                        char *buf = gfi_unpack_entry(oe, &size);
2501                        parse_from_commit(b, buf, size);
2502                        free(buf);
2503                } else
2504                        parse_from_existing(b);
2505        } else if (!get_sha1(from, b->sha1))
2506                parse_from_existing(b);
2507        else
2508                die("Invalid ref name or SHA1 expression: %s", from);
2509
2510        read_next_command();
2511        return 1;
2512}
2513
2514static struct hash_list *parse_merge(unsigned int *count)
2515{
2516        struct hash_list *list = NULL, *n, *e = e;
2517        const char *from;
2518        struct branch *s;
2519
2520        *count = 0;
2521        while (!prefixcmp(command_buf.buf, "merge ")) {
2522                from = strchr(command_buf.buf, ' ') + 1;
2523                n = xmalloc(sizeof(*n));
2524                s = lookup_branch(from);
2525                if (s)
2526                        hashcpy(n->sha1, s->sha1);
2527                else if (*from == ':') {
2528                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2529                        struct object_entry *oe = find_mark(idnum);
2530                        if (oe->type != OBJ_COMMIT)
2531                                die("Mark :%" PRIuMAX " not a commit", idnum);
2532                        hashcpy(n->sha1, oe->idx.sha1);
2533                } else if (!get_sha1(from, n->sha1)) {
2534                        unsigned long size;
2535                        char *buf = read_object_with_reference(n->sha1,
2536                                commit_type, &size, n->sha1);
2537                        if (!buf || size < 46)
2538                                die("Not a valid commit: %s", from);
2539                        free(buf);
2540                } else
2541                        die("Invalid ref name or SHA1 expression: %s", from);
2542
2543                n->next = NULL;
2544                if (list)
2545                        e->next = n;
2546                else
2547                        list = n;
2548                e = n;
2549                (*count)++;
2550                read_next_command();
2551        }
2552        return list;
2553}
2554
2555static void parse_new_commit(void)
2556{
2557        static struct strbuf msg = STRBUF_INIT;
2558        struct branch *b;
2559        char *sp;
2560        char *author = NULL;
2561        char *committer = NULL;
2562        struct hash_list *merge_list = NULL;
2563        unsigned int merge_count;
2564        unsigned char prev_fanout, new_fanout;
2565
2566        /* Obtain the branch name from the rest of our command */
2567        sp = strchr(command_buf.buf, ' ') + 1;
2568        b = lookup_branch(sp);
2569        if (!b)
2570                b = new_branch(sp);
2571
2572        read_next_command();
2573        parse_mark();
2574        if (!prefixcmp(command_buf.buf, "author ")) {
2575                author = parse_ident(command_buf.buf + 7);
2576                read_next_command();
2577        }
2578        if (!prefixcmp(command_buf.buf, "committer ")) {
2579                committer = parse_ident(command_buf.buf + 10);
2580                read_next_command();
2581        }
2582        if (!committer)
2583                die("Expected committer but didn't get one");
2584        parse_data(&msg, 0, NULL);
2585        read_next_command();
2586        parse_from(b);
2587        merge_list = parse_merge(&merge_count);
2588
2589        /* ensure the branch is active/loaded */
2590        if (!b->branch_tree.tree || !max_active_branches) {
2591                unload_one_branch();
2592                load_branch(b);
2593        }
2594
2595        prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2596
2597        /* file_change* */
2598        while (command_buf.len > 0) {
2599                if (!prefixcmp(command_buf.buf, "M "))
2600                        file_change_m(b);
2601                else if (!prefixcmp(command_buf.buf, "D "))
2602                        file_change_d(b);
2603                else if (!prefixcmp(command_buf.buf, "R "))
2604                        file_change_cr(b, 1);
2605                else if (!prefixcmp(command_buf.buf, "C "))
2606                        file_change_cr(b, 0);
2607                else if (!prefixcmp(command_buf.buf, "N "))
2608                        note_change_n(b, prev_fanout);
2609                else if (!strcmp("deleteall", command_buf.buf))
2610                        file_change_deleteall(b);
2611                else {
2612                        unread_command_buf = 1;
2613                        break;
2614                }
2615                if (read_next_command() == EOF)
2616                        break;
2617        }
2618
2619        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2620        if (new_fanout != prev_fanout)
2621                b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2622
2623        /* build the tree and the commit */
2624        store_tree(&b->branch_tree);
2625        hashcpy(b->branch_tree.versions[0].sha1,
2626                b->branch_tree.versions[1].sha1);
2627
2628        strbuf_reset(&new_data);
2629        strbuf_addf(&new_data, "tree %s\n",
2630                sha1_to_hex(b->branch_tree.versions[1].sha1));
2631        if (!is_null_sha1(b->sha1))
2632                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2633        while (merge_list) {
2634                struct hash_list *next = merge_list->next;
2635                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2636                free(merge_list);
2637                merge_list = next;
2638        }
2639        strbuf_addf(&new_data,
2640                "author %s\n"
2641                "committer %s\n"
2642                "\n",
2643                author ? author : committer, committer);
2644        strbuf_addbuf(&new_data, &msg);
2645        free(author);
2646        free(committer);
2647
2648        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2649                b->pack_id = pack_id;
2650        b->last_commit = object_count_by_type[OBJ_COMMIT];
2651}
2652
2653static void parse_new_tag(void)
2654{
2655        static struct strbuf msg = STRBUF_INIT;
2656        char *sp;
2657        const char *from;
2658        char *tagger;
2659        struct branch *s;
2660        struct tag *t;
2661        uintmax_t from_mark = 0;
2662        unsigned char sha1[20];
2663        enum object_type type;
2664
2665        /* Obtain the new tag name from the rest of our command */
2666        sp = strchr(command_buf.buf, ' ') + 1;
2667        t = pool_alloc(sizeof(struct tag));
2668        t->next_tag = NULL;
2669        t->name = pool_strdup(sp);
2670        if (last_tag)
2671                last_tag->next_tag = t;
2672        else
2673                first_tag = t;
2674        last_tag = t;
2675        read_next_command();
2676
2677        /* from ... */
2678        if (prefixcmp(command_buf.buf, "from "))
2679                die("Expected from command, got %s", command_buf.buf);
2680        from = strchr(command_buf.buf, ' ') + 1;
2681        s = lookup_branch(from);
2682        if (s) {
2683                hashcpy(sha1, s->sha1);
2684                type = OBJ_COMMIT;
2685        } else if (*from == ':') {
2686                struct object_entry *oe;
2687                from_mark = strtoumax(from + 1, NULL, 10);
2688                oe = find_mark(from_mark);
2689                type = oe->type;
2690                hashcpy(sha1, oe->idx.sha1);
2691        } else if (!get_sha1(from, sha1)) {
2692                unsigned long size;
2693                char *buf;
2694
2695                buf = read_sha1_file(sha1, &type, &size);
2696                if (!buf || size < 46)
2697                        die("Not a valid commit: %s", from);
2698                free(buf);
2699        } else
2700                die("Invalid ref name or SHA1 expression: %s", from);
2701        read_next_command();
2702
2703        /* tagger ... */
2704        if (!prefixcmp(command_buf.buf, "tagger ")) {
2705                tagger = parse_ident(command_buf.buf + 7);
2706                read_next_command();
2707        } else
2708                tagger = NULL;
2709
2710        /* tag payload/message */
2711        parse_data(&msg, 0, NULL);
2712
2713        /* build the tag object */
2714        strbuf_reset(&new_data);
2715
2716        strbuf_addf(&new_data,
2717                    "object %s\n"
2718                    "type %s\n"
2719                    "tag %s\n",
2720                    sha1_to_hex(sha1), typename(type), t->name);
2721        if (tagger)
2722                strbuf_addf(&new_data,
2723                            "tagger %s\n", tagger);
2724        strbuf_addch(&new_data, '\n');
2725        strbuf_addbuf(&new_data, &msg);
2726        free(tagger);
2727
2728        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2729                t->pack_id = MAX_PACK_ID;
2730        else
2731                t->pack_id = pack_id;
2732}
2733
2734static void parse_reset_branch(void)
2735{
2736        struct branch *b;
2737        char *sp;
2738
2739        /* Obtain the branch name from the rest of our command */
2740        sp = strchr(command_buf.buf, ' ') + 1;
2741        b = lookup_branch(sp);
2742        if (b) {
2743                hashclr(b->sha1);
2744                hashclr(b->branch_tree.versions[0].sha1);
2745                hashclr(b->branch_tree.versions[1].sha1);
2746                if (b->branch_tree.tree) {
2747                        release_tree_content_recursive(b->branch_tree.tree);
2748                        b->branch_tree.tree = NULL;
2749                }
2750        }
2751        else
2752                b = new_branch(sp);
2753        read_next_command();
2754        parse_from(b);
2755        if (command_buf.len > 0)
2756                unread_command_buf = 1;
2757}
2758
2759static void cat_blob_write(const char *buf, unsigned long size)
2760{
2761        if (write_in_full(cat_blob_fd, buf, size) != size)
2762                die_errno("Write to frontend failed");
2763}
2764
2765static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2766{
2767        struct strbuf line = STRBUF_INIT;
2768        unsigned long size;
2769        enum object_type type = 0;
2770        char *buf;
2771
2772        if (!oe || oe->pack_id == MAX_PACK_ID) {
2773                buf = read_sha1_file(sha1, &type, &size);
2774        } else {
2775                type = oe->type;
2776                buf = gfi_unpack_entry(oe, &size);
2777        }
2778
2779        /*
2780         * Output based on batch_one_object() from cat-file.c.
2781         */
2782        if (type <= 0) {
2783                strbuf_reset(&line);
2784                strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2785                cat_blob_write(line.buf, line.len);
2786                strbuf_release(&line);
2787                free(buf);
2788                return;
2789        }
2790        if (!buf)
2791                die("Can't read object %s", sha1_to_hex(sha1));
2792        if (type != OBJ_BLOB)
2793                die("Object %s is a %s but a blob was expected.",
2794                    sha1_to_hex(sha1), typename(type));
2795        strbuf_reset(&line);
2796        strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2797                                                typename(type), size);
2798        cat_blob_write(line.buf, line.len);
2799        strbuf_release(&line);
2800        cat_blob_write(buf, size);
2801        cat_blob_write("\n", 1);
2802        free(buf);
2803}
2804
2805static void parse_cat_blob(void)
2806{
2807        const char *p;
2808        struct object_entry *oe = oe;
2809        unsigned char sha1[20];
2810
2811        /* cat-blob SP <object> LF */
2812        p = command_buf.buf + strlen("cat-blob ");
2813        if (*p == ':') {
2814                char *x;
2815                oe = find_mark(strtoumax(p + 1, &x, 10));
2816                if (x == p + 1)
2817                        die("Invalid mark: %s", command_buf.buf);
2818                if (!oe)
2819                        die("Unknown mark: %s", command_buf.buf);
2820                if (*x)
2821                        die("Garbage after mark: %s", command_buf.buf);
2822                hashcpy(sha1, oe->idx.sha1);
2823        } else {
2824                if (get_sha1_hex(p, sha1))
2825                        die("Invalid SHA1: %s", command_buf.buf);
2826                if (p[40])
2827                        die("Garbage after SHA1: %s", command_buf.buf);
2828                oe = find_object(sha1);
2829        }
2830
2831        cat_blob(oe, sha1);
2832}
2833
2834static void checkpoint(void)
2835{
2836        checkpoint_requested = 0;
2837        if (object_count) {
2838                cycle_packfile();
2839                dump_branches();
2840                dump_tags();
2841                dump_marks();
2842        }
2843}
2844
2845static void parse_checkpoint(void)
2846{
2847        checkpoint_requested = 1;
2848        skip_optional_lf();
2849}
2850
2851static void parse_progress(void)
2852{
2853        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2854        fputc('\n', stdout);
2855        fflush(stdout);
2856        skip_optional_lf();
2857}
2858
2859static char* make_fast_import_path(const char *path)
2860{
2861        struct strbuf abs_path = STRBUF_INIT;
2862
2863        if (!relative_marks_paths || is_absolute_path(path))
2864                return xstrdup(path);
2865        strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
2866        return strbuf_detach(&abs_path, NULL);
2867}
2868
2869static void option_import_marks(const char *marks, int from_stream)
2870{
2871        if (import_marks_file) {
2872                if (from_stream)
2873                        die("Only one import-marks command allowed per stream");
2874
2875                /* read previous mark file */
2876                if(!import_marks_file_from_stream)
2877                        read_marks();
2878        }
2879
2880        import_marks_file = make_fast_import_path(marks);
2881        safe_create_leading_directories_const(import_marks_file);
2882        import_marks_file_from_stream = from_stream;
2883}
2884
2885static void option_date_format(const char *fmt)
2886{
2887        if (!strcmp(fmt, "raw"))
2888                whenspec = WHENSPEC_RAW;
2889        else if (!strcmp(fmt, "rfc2822"))
2890                whenspec = WHENSPEC_RFC2822;
2891        else if (!strcmp(fmt, "now"))
2892                whenspec = WHENSPEC_NOW;
2893        else
2894                die("unknown --date-format argument %s", fmt);
2895}
2896
2897static unsigned long ulong_arg(const char *option, const char *arg)
2898{
2899        char *endptr;
2900        unsigned long rv = strtoul(arg, &endptr, 0);
2901        if (strchr(arg, '-') || endptr == arg || *endptr)
2902                die("%s: argument must be a non-negative integer", option);
2903        return rv;
2904}
2905
2906static void option_depth(const char *depth)
2907{
2908        max_depth = ulong_arg("--depth", depth);
2909        if (max_depth > MAX_DEPTH)
2910                die("--depth cannot exceed %u", MAX_DEPTH);
2911}
2912
2913static void option_active_branches(const char *branches)
2914{
2915        max_active_branches = ulong_arg("--active-branches", branches);
2916}
2917
2918static void option_export_marks(const char *marks)
2919{
2920        export_marks_file = make_fast_import_path(marks);
2921        safe_create_leading_directories_const(export_marks_file);
2922}
2923
2924static void option_cat_blob_fd(const char *fd)
2925{
2926        unsigned long n = ulong_arg("--cat-blob-fd", fd);
2927        if (n > (unsigned long) INT_MAX)
2928                die("--cat-blob-fd cannot exceed %d", INT_MAX);
2929        cat_blob_fd = (int) n;
2930}
2931
2932static void option_export_pack_edges(const char *edges)
2933{
2934        if (pack_edges)
2935                fclose(pack_edges);
2936        pack_edges = fopen(edges, "a");
2937        if (!pack_edges)
2938                die_errno("Cannot open '%s'", edges);
2939}
2940
2941static int parse_one_option(const char *option)
2942{
2943        if (!prefixcmp(option, "max-pack-size=")) {
2944                unsigned long v;
2945                if (!git_parse_ulong(option + 14, &v))
2946                        return 0;
2947                if (v < 8192) {
2948                        warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
2949                        v *= 1024 * 1024;
2950                } else if (v < 1024 * 1024) {
2951                        warning("minimum max-pack-size is 1 MiB");
2952                        v = 1024 * 1024;
2953                }
2954                max_packsize = v;
2955        } else if (!prefixcmp(option, "big-file-threshold=")) {
2956                unsigned long v;
2957                if (!git_parse_ulong(option + 19, &v))
2958                        return 0;
2959                big_file_threshold = v;
2960        } else if (!prefixcmp(option, "depth=")) {
2961                option_depth(option + 6);
2962        } else if (!prefixcmp(option, "active-branches=")) {
2963                option_active_branches(option + 16);
2964        } else if (!prefixcmp(option, "export-pack-edges=")) {
2965                option_export_pack_edges(option + 18);
2966        } else if (!prefixcmp(option, "quiet")) {
2967                show_stats = 0;
2968        } else if (!prefixcmp(option, "stats")) {
2969                show_stats = 1;
2970        } else {
2971                return 0;
2972        }
2973
2974        return 1;
2975}
2976
2977static int parse_one_feature(const char *feature, int from_stream)
2978{
2979        if (!prefixcmp(feature, "date-format=")) {
2980                option_date_format(feature + 12);
2981        } else if (!prefixcmp(feature, "import-marks=")) {
2982                option_import_marks(feature + 13, from_stream);
2983        } else if (!prefixcmp(feature, "export-marks=")) {
2984                option_export_marks(feature + 13);
2985        } else if (!strcmp(feature, "cat-blob")) {
2986                ; /* Don't die - this feature is supported */
2987        } else if (!prefixcmp(feature, "relative-marks")) {
2988                relative_marks_paths = 1;
2989        } else if (!prefixcmp(feature, "no-relative-marks")) {
2990                relative_marks_paths = 0;
2991        } else if (!prefixcmp(feature, "force")) {
2992                force_update = 1;
2993        } else {
2994                return 0;
2995        }
2996
2997        return 1;
2998}
2999
3000static void parse_feature(void)
3001{
3002        char *feature = command_buf.buf + 8;
3003
3004        if (seen_data_command)
3005                die("Got feature command '%s' after data command", feature);
3006
3007        if (parse_one_feature(feature, 1))
3008                return;
3009
3010        die("This version of fast-import does not support feature %s.", feature);
3011}
3012
3013static void parse_option(void)
3014{
3015        char *option = command_buf.buf + 11;
3016
3017        if (seen_data_command)
3018                die("Got option command '%s' after data command", option);
3019
3020        if (parse_one_option(option))
3021                return;
3022
3023        die("This version of fast-import does not support option: %s", option);
3024}
3025
3026static int git_pack_config(const char *k, const char *v, void *cb)
3027{
3028        if (!strcmp(k, "pack.depth")) {
3029                max_depth = git_config_int(k, v);
3030                if (max_depth > MAX_DEPTH)
3031                        max_depth = MAX_DEPTH;
3032                return 0;
3033        }
3034        if (!strcmp(k, "pack.compression")) {
3035                int level = git_config_int(k, v);
3036                if (level == -1)
3037                        level = Z_DEFAULT_COMPRESSION;
3038                else if (level < 0 || level > Z_BEST_COMPRESSION)
3039                        die("bad pack compression level %d", level);
3040                pack_compression_level = level;
3041                pack_compression_seen = 1;
3042                return 0;
3043        }
3044        if (!strcmp(k, "pack.indexversion")) {
3045                pack_idx_default_version = git_config_int(k, v);
3046                if (pack_idx_default_version > 2)
3047                        die("bad pack.indexversion=%"PRIu32,
3048                            pack_idx_default_version);
3049                return 0;
3050        }
3051        if (!strcmp(k, "pack.packsizelimit")) {
3052                max_packsize = git_config_ulong(k, v);
3053                return 0;
3054        }
3055        return git_default_config(k, v, cb);
3056}
3057
3058static const char fast_import_usage[] =
3059"git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3060
3061static void parse_argv(void)
3062{
3063        unsigned int i;
3064
3065        for (i = 1; i < global_argc; i++) {
3066                const char *a = global_argv[i];
3067
3068                if (*a != '-' || !strcmp(a, "--"))
3069                        break;
3070
3071                if (parse_one_option(a + 2))
3072                        continue;
3073
3074                if (parse_one_feature(a + 2, 0))
3075                        continue;
3076
3077                if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3078                        option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3079                        continue;
3080                }
3081
3082                die("unknown option %s", a);
3083        }
3084        if (i != global_argc)
3085                usage(fast_import_usage);
3086
3087        seen_data_command = 1;
3088        if (import_marks_file)
3089                read_marks();
3090}
3091
3092int main(int argc, const char **argv)
3093{
3094        unsigned int i;
3095
3096        git_extract_argv0_path(argv[0]);
3097
3098        if (argc == 2 && !strcmp(argv[1], "-h"))
3099                usage(fast_import_usage);
3100
3101        setup_git_directory();
3102        git_config(git_pack_config, NULL);
3103        if (!pack_compression_seen && core_compression_seen)
3104                pack_compression_level = core_compression_level;
3105
3106        alloc_objects(object_entry_alloc);
3107        strbuf_init(&command_buf, 0);
3108        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3109        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3110        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3111        marks = pool_calloc(1, sizeof(struct mark_set));
3112
3113        global_argc = argc;
3114        global_argv = argv;
3115
3116        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3117        for (i = 0; i < (cmd_save - 1); i++)
3118                rc_free[i].next = &rc_free[i + 1];
3119        rc_free[cmd_save - 1].next = NULL;
3120
3121        prepare_packed_git();
3122        start_packfile();
3123        set_die_routine(die_nicely);
3124        set_checkpoint_signal();
3125        while (read_next_command() != EOF) {
3126                if (!strcmp("blob", command_buf.buf))
3127                        parse_new_blob();
3128                else if (!prefixcmp(command_buf.buf, "commit "))
3129                        parse_new_commit();
3130                else if (!prefixcmp(command_buf.buf, "tag "))
3131                        parse_new_tag();
3132                else if (!prefixcmp(command_buf.buf, "reset "))
3133                        parse_reset_branch();
3134                else if (!strcmp("checkpoint", command_buf.buf))
3135                        parse_checkpoint();
3136                else if (!prefixcmp(command_buf.buf, "progress "))
3137                        parse_progress();
3138                else if (!prefixcmp(command_buf.buf, "feature "))
3139                        parse_feature();
3140                else if (!prefixcmp(command_buf.buf, "option git "))
3141                        parse_option();
3142                else if (!prefixcmp(command_buf.buf, "option "))
3143                        /* ignore non-git options*/;
3144                else
3145                        die("Unsupported command: %s", command_buf.buf);
3146
3147                if (checkpoint_requested)
3148                        checkpoint();
3149        }
3150
3151        /* argv hasn't been parsed yet, do so */
3152        if (!seen_data_command)
3153                parse_argv();
3154
3155        end_packfile();
3156
3157        dump_branches();
3158        dump_tags();
3159        unkeep_all_packs();
3160        dump_marks();
3161
3162        if (pack_edges)
3163                fclose(pack_edges);
3164
3165        if (show_stats) {
3166                uintmax_t total_count = 0, duplicate_count = 0;
3167                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3168                        total_count += object_count_by_type[i];
3169                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3170                        duplicate_count += duplicate_count_by_type[i];
3171
3172                fprintf(stderr, "%s statistics:\n", argv[0]);
3173                fprintf(stderr, "---------------------------------------------------------------------\n");
3174                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3175                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3176                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]);
3177                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]);
3178                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]);
3179                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]);
3180                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3181                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3182                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3183                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3184                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3185                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3186                fprintf(stderr, "---------------------------------------------------------------------\n");
3187                pack_report();
3188                fprintf(stderr, "---------------------------------------------------------------------\n");
3189                fprintf(stderr, "\n");
3190        }
3191
3192        return failure ? 1 : 0;
3193}