fast-import.con commit index-pack: start learning to emulate "verify-pack -v" (38a4556)
   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 uintmax_t big_file_threshold = 512 * 1024 * 1024;
 288static int force_update;
 289static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 290static int pack_compression_seen;
 291
 292/* Stats and misc. counters */
 293static uintmax_t alloc_count;
 294static uintmax_t marks_set_count;
 295static uintmax_t object_count_by_type[1 << TYPE_BITS];
 296static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 297static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 298static unsigned long object_count;
 299static unsigned long branch_count;
 300static unsigned long branch_load_count;
 301static int failure;
 302static FILE *pack_edges;
 303static unsigned int show_stats = 1;
 304static int global_argc;
 305static const char **global_argv;
 306
 307/* Memory pools */
 308static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 309static size_t total_allocd;
 310static struct mem_pool *mem_pool;
 311
 312/* Atom management */
 313static unsigned int atom_table_sz = 4451;
 314static unsigned int atom_cnt;
 315static struct atom_str **atom_table;
 316
 317/* The .pack file being generated */
 318static struct pack_idx_option pack_idx_opts;
 319static unsigned int pack_id;
 320static struct sha1file *pack_file;
 321static struct packed_git *pack_data;
 322static struct packed_git **all_packs;
 323static off_t pack_size;
 324
 325/* Table of objects we've written. */
 326static unsigned int object_entry_alloc = 5000;
 327static struct object_entry_pool *blocks;
 328static struct object_entry *object_table[1 << 16];
 329static struct mark_set *marks;
 330static const char *export_marks_file;
 331static const char *import_marks_file;
 332static int import_marks_file_from_stream;
 333static int import_marks_file_ignore_missing;
 334static int relative_marks_paths;
 335
 336/* Our last blob */
 337static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 338
 339/* Tree management */
 340static unsigned int tree_entry_alloc = 1000;
 341static void *avail_tree_entry;
 342static unsigned int avail_tree_table_sz = 100;
 343static struct avail_tree_content **avail_tree_table;
 344static struct strbuf old_tree = STRBUF_INIT;
 345static struct strbuf new_tree = STRBUF_INIT;
 346
 347/* Branch data */
 348static unsigned long max_active_branches = 5;
 349static unsigned long cur_active_branches;
 350static unsigned long branch_table_sz = 1039;
 351static struct branch **branch_table;
 352static struct branch *active_branches;
 353
 354/* Tag data */
 355static struct tag *first_tag;
 356static struct tag *last_tag;
 357
 358/* Input stream parsing */
 359static whenspec_type whenspec = WHENSPEC_RAW;
 360static struct strbuf command_buf = STRBUF_INIT;
 361static int unread_command_buf;
 362static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 363static struct recent_command *cmd_tail = &cmd_hist;
 364static struct recent_command *rc_free;
 365static unsigned int cmd_save = 100;
 366static uintmax_t next_mark;
 367static struct strbuf new_data = STRBUF_INIT;
 368static int seen_data_command;
 369
 370/* Signal handling */
 371static volatile sig_atomic_t checkpoint_requested;
 372
 373/* Where to write output of cat-blob commands */
 374static int cat_blob_fd = STDOUT_FILENO;
 375
 376static void parse_argv(void);
 377static void parse_cat_blob(void);
 378
 379static void write_branch_report(FILE *rpt, struct branch *b)
 380{
 381        fprintf(rpt, "%s:\n", b->name);
 382
 383        fprintf(rpt, "  status      :");
 384        if (b->active)
 385                fputs(" active", rpt);
 386        if (b->branch_tree.tree)
 387                fputs(" loaded", rpt);
 388        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 389                fputs(" dirty", rpt);
 390        fputc('\n', rpt);
 391
 392        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 393        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 394        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 395        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 396
 397        fputs("  last pack   : ", rpt);
 398        if (b->pack_id < MAX_PACK_ID)
 399                fprintf(rpt, "%u", b->pack_id);
 400        fputc('\n', rpt);
 401
 402        fputc('\n', rpt);
 403}
 404
 405static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 406
 407static void write_crash_report(const char *err)
 408{
 409        char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 410        FILE *rpt = fopen(loc, "w");
 411        struct branch *b;
 412        unsigned long lu;
 413        struct recent_command *rc;
 414
 415        if (!rpt) {
 416                error("can't write crash report %s: %s", loc, strerror(errno));
 417                return;
 418        }
 419
 420        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 421
 422        fprintf(rpt, "fast-import crash report:\n");
 423        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 424        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 425        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 426        fputc('\n', rpt);
 427
 428        fputs("fatal: ", rpt);
 429        fputs(err, rpt);
 430        fputc('\n', rpt);
 431
 432        fputc('\n', rpt);
 433        fputs("Most Recent Commands Before Crash\n", rpt);
 434        fputs("---------------------------------\n", rpt);
 435        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 436                if (rc->next == &cmd_hist)
 437                        fputs("* ", rpt);
 438                else
 439                        fputs("  ", rpt);
 440                fputs(rc->buf, rpt);
 441                fputc('\n', rpt);
 442        }
 443
 444        fputc('\n', rpt);
 445        fputs("Active Branch LRU\n", rpt);
 446        fputs("-----------------\n", rpt);
 447        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 448                cur_active_branches,
 449                max_active_branches);
 450        fputc('\n', rpt);
 451        fputs("  pos  clock name\n", rpt);
 452        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 453        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 454                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 455                        ++lu, b->last_commit, b->name);
 456
 457        fputc('\n', rpt);
 458        fputs("Inactive Branches\n", rpt);
 459        fputs("-----------------\n", rpt);
 460        for (lu = 0; lu < branch_table_sz; lu++) {
 461                for (b = branch_table[lu]; b; b = b->table_next_branch)
 462                        write_branch_report(rpt, b);
 463        }
 464
 465        if (first_tag) {
 466                struct tag *tg;
 467                fputc('\n', rpt);
 468                fputs("Annotated Tags\n", rpt);
 469                fputs("--------------\n", rpt);
 470                for (tg = first_tag; tg; tg = tg->next_tag) {
 471                        fputs(sha1_to_hex(tg->sha1), rpt);
 472                        fputc(' ', rpt);
 473                        fputs(tg->name, rpt);
 474                        fputc('\n', rpt);
 475                }
 476        }
 477
 478        fputc('\n', rpt);
 479        fputs("Marks\n", rpt);
 480        fputs("-----\n", rpt);
 481        if (export_marks_file)
 482                fprintf(rpt, "  exported to %s\n", export_marks_file);
 483        else
 484                dump_marks_helper(rpt, 0, marks);
 485
 486        fputc('\n', rpt);
 487        fputs("-------------------\n", rpt);
 488        fputs("END OF CRASH REPORT\n", rpt);
 489        fclose(rpt);
 490}
 491
 492static void end_packfile(void);
 493static void unkeep_all_packs(void);
 494static void dump_marks(void);
 495
 496static NORETURN void die_nicely(const char *err, va_list params)
 497{
 498        static int zombie;
 499        char message[2 * PATH_MAX];
 500
 501        vsnprintf(message, sizeof(message), err, params);
 502        fputs("fatal: ", stderr);
 503        fputs(message, stderr);
 504        fputc('\n', stderr);
 505
 506        if (!zombie) {
 507                zombie = 1;
 508                write_crash_report(message);
 509                end_packfile();
 510                unkeep_all_packs();
 511                dump_marks();
 512        }
 513        exit(128);
 514}
 515
 516#ifndef SIGUSR1 /* Windows, for example */
 517
 518static void set_checkpoint_signal(void)
 519{
 520}
 521
 522#else
 523
 524static void checkpoint_signal(int signo)
 525{
 526        checkpoint_requested = 1;
 527}
 528
 529static void set_checkpoint_signal(void)
 530{
 531        struct sigaction sa;
 532
 533        memset(&sa, 0, sizeof(sa));
 534        sa.sa_handler = checkpoint_signal;
 535        sigemptyset(&sa.sa_mask);
 536        sa.sa_flags = SA_RESTART;
 537        sigaction(SIGUSR1, &sa, NULL);
 538}
 539
 540#endif
 541
 542static void alloc_objects(unsigned int cnt)
 543{
 544        struct object_entry_pool *b;
 545
 546        b = xmalloc(sizeof(struct object_entry_pool)
 547                + cnt * sizeof(struct object_entry));
 548        b->next_pool = blocks;
 549        b->next_free = b->entries;
 550        b->end = b->entries + cnt;
 551        blocks = b;
 552        alloc_count += cnt;
 553}
 554
 555static struct object_entry *new_object(unsigned char *sha1)
 556{
 557        struct object_entry *e;
 558
 559        if (blocks->next_free == blocks->end)
 560                alloc_objects(object_entry_alloc);
 561
 562        e = blocks->next_free++;
 563        hashcpy(e->idx.sha1, sha1);
 564        return e;
 565}
 566
 567static struct object_entry *find_object(unsigned char *sha1)
 568{
 569        unsigned int h = sha1[0] << 8 | sha1[1];
 570        struct object_entry *e;
 571        for (e = object_table[h]; e; e = e->next)
 572                if (!hashcmp(sha1, e->idx.sha1))
 573                        return e;
 574        return NULL;
 575}
 576
 577static struct object_entry *insert_object(unsigned char *sha1)
 578{
 579        unsigned int h = sha1[0] << 8 | sha1[1];
 580        struct object_entry *e = object_table[h];
 581
 582        while (e) {
 583                if (!hashcmp(sha1, e->idx.sha1))
 584                        return e;
 585                e = e->next;
 586        }
 587
 588        e = new_object(sha1);
 589        e->next = object_table[h];
 590        e->idx.offset = 0;
 591        object_table[h] = e;
 592        return e;
 593}
 594
 595static unsigned int hc_str(const char *s, size_t len)
 596{
 597        unsigned int r = 0;
 598        while (len-- > 0)
 599                r = r * 31 + *s++;
 600        return r;
 601}
 602
 603static void *pool_alloc(size_t len)
 604{
 605        struct mem_pool *p;
 606        void *r;
 607
 608        /* round up to a 'uintmax_t' alignment */
 609        if (len & (sizeof(uintmax_t) - 1))
 610                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 611
 612        for (p = mem_pool; p; p = p->next_pool)
 613                if ((p->end - p->next_free >= len))
 614                        break;
 615
 616        if (!p) {
 617                if (len >= (mem_pool_alloc/2)) {
 618                        total_allocd += len;
 619                        return xmalloc(len);
 620                }
 621                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 622                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 623                p->next_pool = mem_pool;
 624                p->next_free = (char *) p->space;
 625                p->end = p->next_free + mem_pool_alloc;
 626                mem_pool = p;
 627        }
 628
 629        r = p->next_free;
 630        p->next_free += len;
 631        return r;
 632}
 633
 634static void *pool_calloc(size_t count, size_t size)
 635{
 636        size_t len = count * size;
 637        void *r = pool_alloc(len);
 638        memset(r, 0, len);
 639        return r;
 640}
 641
 642static char *pool_strdup(const char *s)
 643{
 644        char *r = pool_alloc(strlen(s) + 1);
 645        strcpy(r, s);
 646        return r;
 647}
 648
 649static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 650{
 651        struct mark_set *s = marks;
 652        while ((idnum >> s->shift) >= 1024) {
 653                s = pool_calloc(1, sizeof(struct mark_set));
 654                s->shift = marks->shift + 10;
 655                s->data.sets[0] = marks;
 656                marks = s;
 657        }
 658        while (s->shift) {
 659                uintmax_t i = idnum >> s->shift;
 660                idnum -= i << s->shift;
 661                if (!s->data.sets[i]) {
 662                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 663                        s->data.sets[i]->shift = s->shift - 10;
 664                }
 665                s = s->data.sets[i];
 666        }
 667        if (!s->data.marked[idnum])
 668                marks_set_count++;
 669        s->data.marked[idnum] = oe;
 670}
 671
 672static struct object_entry *find_mark(uintmax_t idnum)
 673{
 674        uintmax_t orig_idnum = idnum;
 675        struct mark_set *s = marks;
 676        struct object_entry *oe = NULL;
 677        if ((idnum >> s->shift) < 1024) {
 678                while (s && s->shift) {
 679                        uintmax_t i = idnum >> s->shift;
 680                        idnum -= i << s->shift;
 681                        s = s->data.sets[i];
 682                }
 683                if (s)
 684                        oe = s->data.marked[idnum];
 685        }
 686        if (!oe)
 687                die("mark :%" PRIuMAX " not declared", orig_idnum);
 688        return oe;
 689}
 690
 691static struct atom_str *to_atom(const char *s, unsigned short len)
 692{
 693        unsigned int hc = hc_str(s, len) % atom_table_sz;
 694        struct atom_str *c;
 695
 696        for (c = atom_table[hc]; c; c = c->next_atom)
 697                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 698                        return c;
 699
 700        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 701        c->str_len = len;
 702        strncpy(c->str_dat, s, len);
 703        c->str_dat[len] = 0;
 704        c->next_atom = atom_table[hc];
 705        atom_table[hc] = c;
 706        atom_cnt++;
 707        return c;
 708}
 709
 710static struct branch *lookup_branch(const char *name)
 711{
 712        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 713        struct branch *b;
 714
 715        for (b = branch_table[hc]; b; b = b->table_next_branch)
 716                if (!strcmp(name, b->name))
 717                        return b;
 718        return NULL;
 719}
 720
 721static struct branch *new_branch(const char *name)
 722{
 723        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 724        struct branch *b = lookup_branch(name);
 725
 726        if (b)
 727                die("Invalid attempt to create duplicate branch: %s", name);
 728        switch (check_ref_format(name)) {
 729        case 0: break; /* its valid */
 730        case CHECK_REF_FORMAT_ONELEVEL:
 731                break; /* valid, but too few '/', allow anyway */
 732        default:
 733                die("Branch name doesn't conform to GIT standards: %s", name);
 734        }
 735
 736        b = pool_calloc(1, sizeof(struct branch));
 737        b->name = pool_strdup(name);
 738        b->table_next_branch = branch_table[hc];
 739        b->branch_tree.versions[0].mode = S_IFDIR;
 740        b->branch_tree.versions[1].mode = S_IFDIR;
 741        b->num_notes = 0;
 742        b->active = 0;
 743        b->pack_id = MAX_PACK_ID;
 744        branch_table[hc] = b;
 745        branch_count++;
 746        return b;
 747}
 748
 749static unsigned int hc_entries(unsigned int cnt)
 750{
 751        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 752        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 753}
 754
 755static struct tree_content *new_tree_content(unsigned int cnt)
 756{
 757        struct avail_tree_content *f, *l = NULL;
 758        struct tree_content *t;
 759        unsigned int hc = hc_entries(cnt);
 760
 761        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 762                if (f->entry_capacity >= cnt)
 763                        break;
 764
 765        if (f) {
 766                if (l)
 767                        l->next_avail = f->next_avail;
 768                else
 769                        avail_tree_table[hc] = f->next_avail;
 770        } else {
 771                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 772                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 773                f->entry_capacity = cnt;
 774        }
 775
 776        t = (struct tree_content*)f;
 777        t->entry_count = 0;
 778        t->delta_depth = 0;
 779        return t;
 780}
 781
 782static void release_tree_entry(struct tree_entry *e);
 783static void release_tree_content(struct tree_content *t)
 784{
 785        struct avail_tree_content *f = (struct avail_tree_content*)t;
 786        unsigned int hc = hc_entries(f->entry_capacity);
 787        f->next_avail = avail_tree_table[hc];
 788        avail_tree_table[hc] = f;
 789}
 790
 791static void release_tree_content_recursive(struct tree_content *t)
 792{
 793        unsigned int i;
 794        for (i = 0; i < t->entry_count; i++)
 795                release_tree_entry(t->entries[i]);
 796        release_tree_content(t);
 797}
 798
 799static struct tree_content *grow_tree_content(
 800        struct tree_content *t,
 801        int amt)
 802{
 803        struct tree_content *r = new_tree_content(t->entry_count + amt);
 804        r->entry_count = t->entry_count;
 805        r->delta_depth = t->delta_depth;
 806        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 807        release_tree_content(t);
 808        return r;
 809}
 810
 811static struct tree_entry *new_tree_entry(void)
 812{
 813        struct tree_entry *e;
 814
 815        if (!avail_tree_entry) {
 816                unsigned int n = tree_entry_alloc;
 817                total_allocd += n * sizeof(struct tree_entry);
 818                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 819                while (n-- > 1) {
 820                        *((void**)e) = e + 1;
 821                        e++;
 822                }
 823                *((void**)e) = NULL;
 824        }
 825
 826        e = avail_tree_entry;
 827        avail_tree_entry = *((void**)e);
 828        return e;
 829}
 830
 831static void release_tree_entry(struct tree_entry *e)
 832{
 833        if (e->tree)
 834                release_tree_content_recursive(e->tree);
 835        *((void**)e) = avail_tree_entry;
 836        avail_tree_entry = e;
 837}
 838
 839static struct tree_content *dup_tree_content(struct tree_content *s)
 840{
 841        struct tree_content *d;
 842        struct tree_entry *a, *b;
 843        unsigned int i;
 844
 845        if (!s)
 846                return NULL;
 847        d = new_tree_content(s->entry_count);
 848        for (i = 0; i < s->entry_count; i++) {
 849                a = s->entries[i];
 850                b = new_tree_entry();
 851                memcpy(b, a, sizeof(*a));
 852                if (a->tree && is_null_sha1(b->versions[1].sha1))
 853                        b->tree = dup_tree_content(a->tree);
 854                else
 855                        b->tree = NULL;
 856                d->entries[i] = b;
 857        }
 858        d->entry_count = s->entry_count;
 859        d->delta_depth = s->delta_depth;
 860
 861        return d;
 862}
 863
 864static void start_packfile(void)
 865{
 866        static char tmpfile[PATH_MAX];
 867        struct packed_git *p;
 868        struct pack_header hdr;
 869        int pack_fd;
 870
 871        pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 872                              "pack/tmp_pack_XXXXXX");
 873        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 874        strcpy(p->pack_name, tmpfile);
 875        p->pack_fd = pack_fd;
 876        pack_file = sha1fd(pack_fd, p->pack_name);
 877
 878        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 879        hdr.hdr_version = htonl(2);
 880        hdr.hdr_entries = 0;
 881        sha1write(pack_file, &hdr, sizeof(hdr));
 882
 883        pack_data = p;
 884        pack_size = sizeof(hdr);
 885        object_count = 0;
 886
 887        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 888        all_packs[pack_id] = p;
 889}
 890
 891static const char *create_index(void)
 892{
 893        const char *tmpfile;
 894        struct pack_idx_entry **idx, **c, **last;
 895        struct object_entry *e;
 896        struct object_entry_pool *o;
 897
 898        /* Build the table of object IDs. */
 899        idx = xmalloc(object_count * sizeof(*idx));
 900        c = idx;
 901        for (o = blocks; o; o = o->next_pool)
 902                for (e = o->next_free; e-- != o->entries;)
 903                        if (pack_id == e->pack_id)
 904                                *c++ = &e->idx;
 905        last = idx + object_count;
 906        if (c != last)
 907                die("internal consistency error creating the index");
 908
 909        tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
 910        free(idx);
 911        return tmpfile;
 912}
 913
 914static char *keep_pack(const char *curr_index_name)
 915{
 916        static char name[PATH_MAX];
 917        static const char *keep_msg = "fast-import";
 918        int keep_fd;
 919
 920        keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
 921        if (keep_fd < 0)
 922                die_errno("cannot create keep file");
 923        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 924        if (close(keep_fd))
 925                die_errno("failed to write keep file");
 926
 927        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 928                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 929        if (move_temp_to_file(pack_data->pack_name, name))
 930                die("cannot store pack file");
 931
 932        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 933                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 934        if (move_temp_to_file(curr_index_name, name))
 935                die("cannot store index file");
 936        free((void *)curr_index_name);
 937        return name;
 938}
 939
 940static void unkeep_all_packs(void)
 941{
 942        static char name[PATH_MAX];
 943        int k;
 944
 945        for (k = 0; k < pack_id; k++) {
 946                struct packed_git *p = all_packs[k];
 947                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 948                         get_object_directory(), sha1_to_hex(p->sha1));
 949                unlink_or_warn(name);
 950        }
 951}
 952
 953static void end_packfile(void)
 954{
 955        struct packed_git *old_p = pack_data, *new_p;
 956
 957        clear_delta_base_cache();
 958        if (object_count) {
 959                unsigned char cur_pack_sha1[20];
 960                char *idx_name;
 961                int i;
 962                struct branch *b;
 963                struct tag *t;
 964
 965                close_pack_windows(pack_data);
 966                sha1close(pack_file, cur_pack_sha1, 0);
 967                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 968                                    pack_data->pack_name, object_count,
 969                                    cur_pack_sha1, pack_size);
 970                close(pack_data->pack_fd);
 971                idx_name = keep_pack(create_index());
 972
 973                /* Register the packfile with core git's machinery. */
 974                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 975                if (!new_p)
 976                        die("core git rejected index %s", idx_name);
 977                all_packs[pack_id] = new_p;
 978                install_packed_git(new_p);
 979
 980                /* Print the boundary */
 981                if (pack_edges) {
 982                        fprintf(pack_edges, "%s:", new_p->pack_name);
 983                        for (i = 0; i < branch_table_sz; i++) {
 984                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 985                                        if (b->pack_id == pack_id)
 986                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 987                                }
 988                        }
 989                        for (t = first_tag; t; t = t->next_tag) {
 990                                if (t->pack_id == pack_id)
 991                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 992                        }
 993                        fputc('\n', pack_edges);
 994                        fflush(pack_edges);
 995                }
 996
 997                pack_id++;
 998        }
 999        else {
1000                close(old_p->pack_fd);
1001                unlink_or_warn(old_p->pack_name);
1002        }
1003        free(old_p);
1004
1005        /* We can't carry a delta across packfiles. */
1006        strbuf_release(&last_blob.data);
1007        last_blob.offset = 0;
1008        last_blob.depth = 0;
1009}
1010
1011static void cycle_packfile(void)
1012{
1013        end_packfile();
1014        start_packfile();
1015}
1016
1017static int store_object(
1018        enum object_type type,
1019        struct strbuf *dat,
1020        struct last_object *last,
1021        unsigned char *sha1out,
1022        uintmax_t mark)
1023{
1024        void *out, *delta;
1025        struct object_entry *e;
1026        unsigned char hdr[96];
1027        unsigned char sha1[20];
1028        unsigned long hdrlen, deltalen;
1029        git_SHA_CTX c;
1030        z_stream s;
1031
1032        hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1033                (unsigned long)dat->len) + 1;
1034        git_SHA1_Init(&c);
1035        git_SHA1_Update(&c, hdr, hdrlen);
1036        git_SHA1_Update(&c, dat->buf, dat->len);
1037        git_SHA1_Final(sha1, &c);
1038        if (sha1out)
1039                hashcpy(sha1out, sha1);
1040
1041        e = insert_object(sha1);
1042        if (mark)
1043                insert_mark(mark, e);
1044        if (e->idx.offset) {
1045                duplicate_count_by_type[type]++;
1046                return 1;
1047        } else if (find_sha1_pack(sha1, packed_git)) {
1048                e->type = type;
1049                e->pack_id = MAX_PACK_ID;
1050                e->idx.offset = 1; /* just not zero! */
1051                duplicate_count_by_type[type]++;
1052                return 1;
1053        }
1054
1055        if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1056                delta = diff_delta(last->data.buf, last->data.len,
1057                        dat->buf, dat->len,
1058                        &deltalen, dat->len - 20);
1059        } else
1060                delta = NULL;
1061
1062        memset(&s, 0, sizeof(s));
1063        deflateInit(&s, pack_compression_level);
1064        if (delta) {
1065                s.next_in = delta;
1066                s.avail_in = deltalen;
1067        } else {
1068                s.next_in = (void *)dat->buf;
1069                s.avail_in = dat->len;
1070        }
1071        s.avail_out = deflateBound(&s, s.avail_in);
1072        s.next_out = out = xmalloc(s.avail_out);
1073        while (deflate(&s, Z_FINISH) == Z_OK)
1074                /* nothing */;
1075        deflateEnd(&s);
1076
1077        /* Determine if we should auto-checkpoint. */
1078        if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1079                || (pack_size + 60 + s.total_out) < pack_size) {
1080
1081                /* This new object needs to *not* have the current pack_id. */
1082                e->pack_id = pack_id + 1;
1083                cycle_packfile();
1084
1085                /* We cannot carry a delta into the new pack. */
1086                if (delta) {
1087                        free(delta);
1088                        delta = NULL;
1089
1090                        memset(&s, 0, sizeof(s));
1091                        deflateInit(&s, pack_compression_level);
1092                        s.next_in = (void *)dat->buf;
1093                        s.avail_in = dat->len;
1094                        s.avail_out = deflateBound(&s, s.avail_in);
1095                        s.next_out = out = xrealloc(out, s.avail_out);
1096                        while (deflate(&s, Z_FINISH) == Z_OK)
1097                                /* nothing */;
1098                        deflateEnd(&s);
1099                }
1100        }
1101
1102        e->type = type;
1103        e->pack_id = pack_id;
1104        e->idx.offset = pack_size;
1105        object_count++;
1106        object_count_by_type[type]++;
1107
1108        crc32_begin(pack_file);
1109
1110        if (delta) {
1111                off_t ofs = e->idx.offset - last->offset;
1112                unsigned pos = sizeof(hdr) - 1;
1113
1114                delta_count_by_type[type]++;
1115                e->depth = last->depth + 1;
1116
1117                hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1118                sha1write(pack_file, hdr, hdrlen);
1119                pack_size += hdrlen;
1120
1121                hdr[pos] = ofs & 127;
1122                while (ofs >>= 7)
1123                        hdr[--pos] = 128 | (--ofs & 127);
1124                sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1125                pack_size += sizeof(hdr) - pos;
1126        } else {
1127                e->depth = 0;
1128                hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1129                sha1write(pack_file, hdr, hdrlen);
1130                pack_size += hdrlen;
1131        }
1132
1133        sha1write(pack_file, out, s.total_out);
1134        pack_size += s.total_out;
1135
1136        e->idx.crc32 = crc32_end(pack_file);
1137
1138        free(out);
1139        free(delta);
1140        if (last) {
1141                if (last->no_swap) {
1142                        last->data = *dat;
1143                } else {
1144                        strbuf_swap(&last->data, dat);
1145                }
1146                last->offset = e->idx.offset;
1147                last->depth = e->depth;
1148        }
1149        return 0;
1150}
1151
1152static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1153{
1154        if (ftruncate(pack_data->pack_fd, to)
1155         || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1156                die_errno("cannot truncate pack to skip duplicate");
1157        pack_size = to;
1158
1159        /* yes this is a layering violation */
1160        pack_file->total = to;
1161        pack_file->offset = 0;
1162        pack_file->ctx = *ctx;
1163}
1164
1165static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1166{
1167        size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1168        unsigned char *in_buf = xmalloc(in_sz);
1169        unsigned char *out_buf = xmalloc(out_sz);
1170        struct object_entry *e;
1171        unsigned char sha1[20];
1172        unsigned long hdrlen;
1173        off_t offset;
1174        git_SHA_CTX c;
1175        git_SHA_CTX pack_file_ctx;
1176        z_stream s;
1177        int status = Z_OK;
1178
1179        /* Determine if we should auto-checkpoint. */
1180        if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1181                || (pack_size + 60 + len) < pack_size)
1182                cycle_packfile();
1183
1184        offset = pack_size;
1185
1186        /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1187        sha1flush(pack_file);
1188        pack_file_ctx = pack_file->ctx;
1189
1190        hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1191        if (out_sz <= hdrlen)
1192                die("impossibly large object header");
1193
1194        git_SHA1_Init(&c);
1195        git_SHA1_Update(&c, out_buf, hdrlen);
1196
1197        crc32_begin(pack_file);
1198
1199        memset(&s, 0, sizeof(s));
1200        deflateInit(&s, pack_compression_level);
1201
1202        hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1203        if (out_sz <= hdrlen)
1204                die("impossibly large object header");
1205
1206        s.next_out = out_buf + hdrlen;
1207        s.avail_out = out_sz - hdrlen;
1208
1209        while (status != Z_STREAM_END) {
1210                if (0 < len && !s.avail_in) {
1211                        size_t cnt = in_sz < len ? in_sz : (size_t)len;
1212                        size_t n = fread(in_buf, 1, cnt, stdin);
1213                        if (!n && feof(stdin))
1214                                die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1215
1216                        git_SHA1_Update(&c, in_buf, n);
1217                        s.next_in = in_buf;
1218                        s.avail_in = n;
1219                        len -= n;
1220                }
1221
1222                status = deflate(&s, len ? 0 : Z_FINISH);
1223
1224                if (!s.avail_out || status == Z_STREAM_END) {
1225                        size_t n = s.next_out - out_buf;
1226                        sha1write(pack_file, out_buf, n);
1227                        pack_size += n;
1228                        s.next_out = out_buf;
1229                        s.avail_out = out_sz;
1230                }
1231
1232                switch (status) {
1233                case Z_OK:
1234                case Z_BUF_ERROR:
1235                case Z_STREAM_END:
1236                        continue;
1237                default:
1238                        die("unexpected deflate failure: %d", status);
1239                }
1240        }
1241        deflateEnd(&s);
1242        git_SHA1_Final(sha1, &c);
1243
1244        if (sha1out)
1245                hashcpy(sha1out, sha1);
1246
1247        e = insert_object(sha1);
1248
1249        if (mark)
1250                insert_mark(mark, e);
1251
1252        if (e->idx.offset) {
1253                duplicate_count_by_type[OBJ_BLOB]++;
1254                truncate_pack(offset, &pack_file_ctx);
1255
1256        } else if (find_sha1_pack(sha1, packed_git)) {
1257                e->type = OBJ_BLOB;
1258                e->pack_id = MAX_PACK_ID;
1259                e->idx.offset = 1; /* just not zero! */
1260                duplicate_count_by_type[OBJ_BLOB]++;
1261                truncate_pack(offset, &pack_file_ctx);
1262
1263        } else {
1264                e->depth = 0;
1265                e->type = OBJ_BLOB;
1266                e->pack_id = pack_id;
1267                e->idx.offset = offset;
1268                e->idx.crc32 = crc32_end(pack_file);
1269                object_count++;
1270                object_count_by_type[OBJ_BLOB]++;
1271        }
1272
1273        free(in_buf);
1274        free(out_buf);
1275}
1276
1277/* All calls must be guarded by find_object() or find_mark() to
1278 * ensure the 'struct object_entry' passed was written by this
1279 * process instance.  We unpack the entry by the offset, avoiding
1280 * the need for the corresponding .idx file.  This unpacking rule
1281 * works because we only use OBJ_REF_DELTA within the packfiles
1282 * created by fast-import.
1283 *
1284 * oe must not be NULL.  Such an oe usually comes from giving
1285 * an unknown SHA-1 to find_object() or an undefined mark to
1286 * find_mark().  Callers must test for this condition and use
1287 * the standard read_sha1_file() when it happens.
1288 *
1289 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1290 * find_mark(), where the mark was reloaded from an existing marks
1291 * file and is referencing an object that this fast-import process
1292 * instance did not write out to a packfile.  Callers must test for
1293 * this condition and use read_sha1_file() instead.
1294 */
1295static void *gfi_unpack_entry(
1296        struct object_entry *oe,
1297        unsigned long *sizep)
1298{
1299        enum object_type type;
1300        struct packed_git *p = all_packs[oe->pack_id];
1301        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1302                /* The object is stored in the packfile we are writing to
1303                 * and we have modified it since the last time we scanned
1304                 * back to read a previously written object.  If an old
1305                 * window covered [p->pack_size, p->pack_size + 20) its
1306                 * data is stale and is not valid.  Closing all windows
1307                 * and updating the packfile length ensures we can read
1308                 * the newly written data.
1309                 */
1310                close_pack_windows(p);
1311                sha1flush(pack_file);
1312
1313                /* We have to offer 20 bytes additional on the end of
1314                 * the packfile as the core unpacker code assumes the
1315                 * footer is present at the file end and must promise
1316                 * at least 20 bytes within any window it maps.  But
1317                 * we don't actually create the footer here.
1318                 */
1319                p->pack_size = pack_size + 20;
1320        }
1321        return unpack_entry(p, oe->idx.offset, &type, sizep);
1322}
1323
1324static const char *get_mode(const char *str, uint16_t *modep)
1325{
1326        unsigned char c;
1327        uint16_t mode = 0;
1328
1329        while ((c = *str++) != ' ') {
1330                if (c < '0' || c > '7')
1331                        return NULL;
1332                mode = (mode << 3) + (c - '0');
1333        }
1334        *modep = mode;
1335        return str;
1336}
1337
1338static void load_tree(struct tree_entry *root)
1339{
1340        unsigned char *sha1 = root->versions[1].sha1;
1341        struct object_entry *myoe;
1342        struct tree_content *t;
1343        unsigned long size;
1344        char *buf;
1345        const char *c;
1346
1347        root->tree = t = new_tree_content(8);
1348        if (is_null_sha1(sha1))
1349                return;
1350
1351        myoe = find_object(sha1);
1352        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1353                if (myoe->type != OBJ_TREE)
1354                        die("Not a tree: %s", sha1_to_hex(sha1));
1355                t->delta_depth = myoe->depth;
1356                buf = gfi_unpack_entry(myoe, &size);
1357                if (!buf)
1358                        die("Can't load tree %s", sha1_to_hex(sha1));
1359        } else {
1360                enum object_type type;
1361                buf = read_sha1_file(sha1, &type, &size);
1362                if (!buf || type != OBJ_TREE)
1363                        die("Can't load tree %s", sha1_to_hex(sha1));
1364        }
1365
1366        c = buf;
1367        while (c != (buf + size)) {
1368                struct tree_entry *e = new_tree_entry();
1369
1370                if (t->entry_count == t->entry_capacity)
1371                        root->tree = t = grow_tree_content(t, t->entry_count);
1372                t->entries[t->entry_count++] = e;
1373
1374                e->tree = NULL;
1375                c = get_mode(c, &e->versions[1].mode);
1376                if (!c)
1377                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1378                e->versions[0].mode = e->versions[1].mode;
1379                e->name = to_atom(c, strlen(c));
1380                c += e->name->str_len + 1;
1381                hashcpy(e->versions[0].sha1, (unsigned char *)c);
1382                hashcpy(e->versions[1].sha1, (unsigned char *)c);
1383                c += 20;
1384        }
1385        free(buf);
1386}
1387
1388static int tecmp0 (const void *_a, const void *_b)
1389{
1390        struct tree_entry *a = *((struct tree_entry**)_a);
1391        struct tree_entry *b = *((struct tree_entry**)_b);
1392        return base_name_compare(
1393                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1394                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1395}
1396
1397static int tecmp1 (const void *_a, const void *_b)
1398{
1399        struct tree_entry *a = *((struct tree_entry**)_a);
1400        struct tree_entry *b = *((struct tree_entry**)_b);
1401        return base_name_compare(
1402                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1403                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1404}
1405
1406static void mktree(struct tree_content *t, int v, struct strbuf *b)
1407{
1408        size_t maxlen = 0;
1409        unsigned int i;
1410
1411        if (!v)
1412                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1413        else
1414                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1415
1416        for (i = 0; i < t->entry_count; i++) {
1417                if (t->entries[i]->versions[v].mode)
1418                        maxlen += t->entries[i]->name->str_len + 34;
1419        }
1420
1421        strbuf_reset(b);
1422        strbuf_grow(b, maxlen);
1423        for (i = 0; i < t->entry_count; i++) {
1424                struct tree_entry *e = t->entries[i];
1425                if (!e->versions[v].mode)
1426                        continue;
1427                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1428                                        e->name->str_dat, '\0');
1429                strbuf_add(b, e->versions[v].sha1, 20);
1430        }
1431}
1432
1433static void store_tree(struct tree_entry *root)
1434{
1435        struct tree_content *t = root->tree;
1436        unsigned int i, j, del;
1437        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1438        struct object_entry *le;
1439
1440        if (!is_null_sha1(root->versions[1].sha1))
1441                return;
1442
1443        for (i = 0; i < t->entry_count; i++) {
1444                if (t->entries[i]->tree)
1445                        store_tree(t->entries[i]);
1446        }
1447
1448        le = find_object(root->versions[0].sha1);
1449        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1450                mktree(t, 0, &old_tree);
1451                lo.data = old_tree;
1452                lo.offset = le->idx.offset;
1453                lo.depth = t->delta_depth;
1454        }
1455
1456        mktree(t, 1, &new_tree);
1457        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1458
1459        t->delta_depth = lo.depth;
1460        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1461                struct tree_entry *e = t->entries[i];
1462                if (e->versions[1].mode) {
1463                        e->versions[0].mode = e->versions[1].mode;
1464                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1465                        t->entries[j++] = e;
1466                } else {
1467                        release_tree_entry(e);
1468                        del++;
1469                }
1470        }
1471        t->entry_count -= del;
1472}
1473
1474static void tree_content_replace(
1475        struct tree_entry *root,
1476        const unsigned char *sha1,
1477        const uint16_t mode,
1478        struct tree_content *newtree)
1479{
1480        if (!S_ISDIR(mode))
1481                die("Root cannot be a non-directory");
1482        hashcpy(root->versions[1].sha1, sha1);
1483        if (root->tree)
1484                release_tree_content_recursive(root->tree);
1485        root->tree = newtree;
1486}
1487
1488static int tree_content_set(
1489        struct tree_entry *root,
1490        const char *p,
1491        const unsigned char *sha1,
1492        const uint16_t mode,
1493        struct tree_content *subtree)
1494{
1495        struct tree_content *t;
1496        const char *slash1;
1497        unsigned int i, n;
1498        struct tree_entry *e;
1499
1500        slash1 = strchr(p, '/');
1501        if (slash1)
1502                n = slash1 - p;
1503        else
1504                n = strlen(p);
1505        if (!n)
1506                die("Empty path component found in input");
1507        if (!slash1 && !S_ISDIR(mode) && subtree)
1508                die("Non-directories cannot have subtrees");
1509
1510        if (!root->tree)
1511                load_tree(root);
1512        t = root->tree;
1513        for (i = 0; i < t->entry_count; i++) {
1514                e = t->entries[i];
1515                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1516                        if (!slash1) {
1517                                if (!S_ISDIR(mode)
1518                                                && e->versions[1].mode == mode
1519                                                && !hashcmp(e->versions[1].sha1, sha1))
1520                                        return 0;
1521                                e->versions[1].mode = mode;
1522                                hashcpy(e->versions[1].sha1, sha1);
1523                                if (e->tree)
1524                                        release_tree_content_recursive(e->tree);
1525                                e->tree = subtree;
1526                                hashclr(root->versions[1].sha1);
1527                                return 1;
1528                        }
1529                        if (!S_ISDIR(e->versions[1].mode)) {
1530                                e->tree = new_tree_content(8);
1531                                e->versions[1].mode = S_IFDIR;
1532                        }
1533                        if (!e->tree)
1534                                load_tree(e);
1535                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1536                                hashclr(root->versions[1].sha1);
1537                                return 1;
1538                        }
1539                        return 0;
1540                }
1541        }
1542
1543        if (t->entry_count == t->entry_capacity)
1544                root->tree = t = grow_tree_content(t, t->entry_count);
1545        e = new_tree_entry();
1546        e->name = to_atom(p, n);
1547        e->versions[0].mode = 0;
1548        hashclr(e->versions[0].sha1);
1549        t->entries[t->entry_count++] = e;
1550        if (slash1) {
1551                e->tree = new_tree_content(8);
1552                e->versions[1].mode = S_IFDIR;
1553                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1554        } else {
1555                e->tree = subtree;
1556                e->versions[1].mode = mode;
1557                hashcpy(e->versions[1].sha1, sha1);
1558        }
1559        hashclr(root->versions[1].sha1);
1560        return 1;
1561}
1562
1563static int tree_content_remove(
1564        struct tree_entry *root,
1565        const char *p,
1566        struct tree_entry *backup_leaf)
1567{
1568        struct tree_content *t;
1569        const char *slash1;
1570        unsigned int i, n;
1571        struct tree_entry *e;
1572
1573        slash1 = strchr(p, '/');
1574        if (slash1)
1575                n = slash1 - p;
1576        else
1577                n = strlen(p);
1578
1579        if (!root->tree)
1580                load_tree(root);
1581        t = root->tree;
1582        for (i = 0; i < t->entry_count; i++) {
1583                e = t->entries[i];
1584                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1585                        if (slash1 && !S_ISDIR(e->versions[1].mode))
1586                                /*
1587                                 * If p names a file in some subdirectory, and a
1588                                 * file or symlink matching the name of the
1589                                 * parent directory of p exists, then p cannot
1590                                 * exist and need not be deleted.
1591                                 */
1592                                return 1;
1593                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1594                                goto del_entry;
1595                        if (!e->tree)
1596                                load_tree(e);
1597                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1598                                for (n = 0; n < e->tree->entry_count; n++) {
1599                                        if (e->tree->entries[n]->versions[1].mode) {
1600                                                hashclr(root->versions[1].sha1);
1601                                                return 1;
1602                                        }
1603                                }
1604                                backup_leaf = NULL;
1605                                goto del_entry;
1606                        }
1607                        return 0;
1608                }
1609        }
1610        return 0;
1611
1612del_entry:
1613        if (backup_leaf)
1614                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1615        else if (e->tree)
1616                release_tree_content_recursive(e->tree);
1617        e->tree = NULL;
1618        e->versions[1].mode = 0;
1619        hashclr(e->versions[1].sha1);
1620        hashclr(root->versions[1].sha1);
1621        return 1;
1622}
1623
1624static int tree_content_get(
1625        struct tree_entry *root,
1626        const char *p,
1627        struct tree_entry *leaf)
1628{
1629        struct tree_content *t;
1630        const char *slash1;
1631        unsigned int i, n;
1632        struct tree_entry *e;
1633
1634        slash1 = strchr(p, '/');
1635        if (slash1)
1636                n = slash1 - p;
1637        else
1638                n = strlen(p);
1639
1640        if (!root->tree)
1641                load_tree(root);
1642        t = root->tree;
1643        for (i = 0; i < t->entry_count; i++) {
1644                e = t->entries[i];
1645                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1646                        if (!slash1) {
1647                                memcpy(leaf, e, sizeof(*leaf));
1648                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1649                                        leaf->tree = dup_tree_content(e->tree);
1650                                else
1651                                        leaf->tree = NULL;
1652                                return 1;
1653                        }
1654                        if (!S_ISDIR(e->versions[1].mode))
1655                                return 0;
1656                        if (!e->tree)
1657                                load_tree(e);
1658                        return tree_content_get(e, slash1 + 1, leaf);
1659                }
1660        }
1661        return 0;
1662}
1663
1664static int update_branch(struct branch *b)
1665{
1666        static const char *msg = "fast-import";
1667        struct ref_lock *lock;
1668        unsigned char old_sha1[20];
1669
1670        if (is_null_sha1(b->sha1))
1671                return 0;
1672        if (read_ref(b->name, old_sha1))
1673                hashclr(old_sha1);
1674        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1675        if (!lock)
1676                return error("Unable to lock %s", b->name);
1677        if (!force_update && !is_null_sha1(old_sha1)) {
1678                struct commit *old_cmit, *new_cmit;
1679
1680                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1681                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1682                if (!old_cmit || !new_cmit) {
1683                        unlock_ref(lock);
1684                        return error("Branch %s is missing commits.", b->name);
1685                }
1686
1687                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1688                        unlock_ref(lock);
1689                        warning("Not updating %s"
1690                                " (new tip %s does not contain %s)",
1691                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1692                        return -1;
1693                }
1694        }
1695        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1696                return error("Unable to update %s", b->name);
1697        return 0;
1698}
1699
1700static void dump_branches(void)
1701{
1702        unsigned int i;
1703        struct branch *b;
1704
1705        for (i = 0; i < branch_table_sz; i++) {
1706                for (b = branch_table[i]; b; b = b->table_next_branch)
1707                        failure |= update_branch(b);
1708        }
1709}
1710
1711static void dump_tags(void)
1712{
1713        static const char *msg = "fast-import";
1714        struct tag *t;
1715        struct ref_lock *lock;
1716        char ref_name[PATH_MAX];
1717
1718        for (t = first_tag; t; t = t->next_tag) {
1719                sprintf(ref_name, "tags/%s", t->name);
1720                lock = lock_ref_sha1(ref_name, NULL);
1721                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1722                        failure |= error("Unable to update %s", ref_name);
1723        }
1724}
1725
1726static void dump_marks_helper(FILE *f,
1727        uintmax_t base,
1728        struct mark_set *m)
1729{
1730        uintmax_t k;
1731        if (m->shift) {
1732                for (k = 0; k < 1024; k++) {
1733                        if (m->data.sets[k])
1734                                dump_marks_helper(f, base + (k << m->shift),
1735                                        m->data.sets[k]);
1736                }
1737        } else {
1738                for (k = 0; k < 1024; k++) {
1739                        if (m->data.marked[k])
1740                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1741                                        sha1_to_hex(m->data.marked[k]->idx.sha1));
1742                }
1743        }
1744}
1745
1746static void dump_marks(void)
1747{
1748        static struct lock_file mark_lock;
1749        int mark_fd;
1750        FILE *f;
1751
1752        if (!export_marks_file)
1753                return;
1754
1755        mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1756        if (mark_fd < 0) {
1757                failure |= error("Unable to write marks file %s: %s",
1758                        export_marks_file, strerror(errno));
1759                return;
1760        }
1761
1762        f = fdopen(mark_fd, "w");
1763        if (!f) {
1764                int saved_errno = errno;
1765                rollback_lock_file(&mark_lock);
1766                failure |= error("Unable to write marks file %s: %s",
1767                        export_marks_file, strerror(saved_errno));
1768                return;
1769        }
1770
1771        /*
1772         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1773         * Assign -1 to the lock file descriptor so that commit_lock_file()
1774         * won't try to close() it.
1775         */
1776        mark_lock.fd = -1;
1777
1778        dump_marks_helper(f, 0, marks);
1779        if (ferror(f) || fclose(f)) {
1780                int saved_errno = errno;
1781                rollback_lock_file(&mark_lock);
1782                failure |= error("Unable to write marks file %s: %s",
1783                        export_marks_file, strerror(saved_errno));
1784                return;
1785        }
1786
1787        if (commit_lock_file(&mark_lock)) {
1788                int saved_errno = errno;
1789                rollback_lock_file(&mark_lock);
1790                failure |= error("Unable to commit marks file %s: %s",
1791                        export_marks_file, strerror(saved_errno));
1792                return;
1793        }
1794}
1795
1796static void read_marks(void)
1797{
1798        char line[512];
1799        FILE *f = fopen(import_marks_file, "r");
1800        if (f)
1801                ;
1802        else if (import_marks_file_ignore_missing && errno == ENOENT)
1803                return; /* Marks file does not exist */
1804        else
1805                die_errno("cannot read '%s'", import_marks_file);
1806        while (fgets(line, sizeof(line), f)) {
1807                uintmax_t mark;
1808                char *end;
1809                unsigned char sha1[20];
1810                struct object_entry *e;
1811
1812                end = strchr(line, '\n');
1813                if (line[0] != ':' || !end)
1814                        die("corrupt mark line: %s", line);
1815                *end = 0;
1816                mark = strtoumax(line + 1, &end, 10);
1817                if (!mark || end == line + 1
1818                        || *end != ' ' || get_sha1(end + 1, sha1))
1819                        die("corrupt mark line: %s", line);
1820                e = find_object(sha1);
1821                if (!e) {
1822                        enum object_type type = sha1_object_info(sha1, NULL);
1823                        if (type < 0)
1824                                die("object not found: %s", sha1_to_hex(sha1));
1825                        e = insert_object(sha1);
1826                        e->type = type;
1827                        e->pack_id = MAX_PACK_ID;
1828                        e->idx.offset = 1; /* just not zero! */
1829                }
1830                insert_mark(mark, e);
1831        }
1832        fclose(f);
1833}
1834
1835
1836static int read_next_command(void)
1837{
1838        static int stdin_eof = 0;
1839
1840        if (stdin_eof) {
1841                unread_command_buf = 0;
1842                return EOF;
1843        }
1844
1845        for (;;) {
1846                if (unread_command_buf) {
1847                        unread_command_buf = 0;
1848                } else {
1849                        struct recent_command *rc;
1850
1851                        strbuf_detach(&command_buf, NULL);
1852                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1853                        if (stdin_eof)
1854                                return EOF;
1855
1856                        if (!seen_data_command
1857                                && prefixcmp(command_buf.buf, "feature ")
1858                                && prefixcmp(command_buf.buf, "option ")) {
1859                                parse_argv();
1860                        }
1861
1862                        rc = rc_free;
1863                        if (rc)
1864                                rc_free = rc->next;
1865                        else {
1866                                rc = cmd_hist.next;
1867                                cmd_hist.next = rc->next;
1868                                cmd_hist.next->prev = &cmd_hist;
1869                                free(rc->buf);
1870                        }
1871
1872                        rc->buf = command_buf.buf;
1873                        rc->prev = cmd_tail;
1874                        rc->next = cmd_hist.prev;
1875                        rc->prev->next = rc;
1876                        cmd_tail = rc;
1877                }
1878                if (!prefixcmp(command_buf.buf, "cat-blob ")) {
1879                        parse_cat_blob();
1880                        continue;
1881                }
1882                if (command_buf.buf[0] == '#')
1883                        continue;
1884                return 0;
1885        }
1886}
1887
1888static void skip_optional_lf(void)
1889{
1890        int term_char = fgetc(stdin);
1891        if (term_char != '\n' && term_char != EOF)
1892                ungetc(term_char, stdin);
1893}
1894
1895static void parse_mark(void)
1896{
1897        if (!prefixcmp(command_buf.buf, "mark :")) {
1898                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1899                read_next_command();
1900        }
1901        else
1902                next_mark = 0;
1903}
1904
1905static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1906{
1907        strbuf_reset(sb);
1908
1909        if (prefixcmp(command_buf.buf, "data "))
1910                die("Expected 'data n' command, found: %s", command_buf.buf);
1911
1912        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1913                char *term = xstrdup(command_buf.buf + 5 + 2);
1914                size_t term_len = command_buf.len - 5 - 2;
1915
1916                strbuf_detach(&command_buf, NULL);
1917                for (;;) {
1918                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1919                                die("EOF in data (terminator '%s' not found)", term);
1920                        if (term_len == command_buf.len
1921                                && !strcmp(term, command_buf.buf))
1922                                break;
1923                        strbuf_addbuf(sb, &command_buf);
1924                        strbuf_addch(sb, '\n');
1925                }
1926                free(term);
1927        }
1928        else {
1929                uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1930                size_t n = 0, length = (size_t)len;
1931
1932                if (limit && limit < len) {
1933                        *len_res = len;
1934                        return 0;
1935                }
1936                if (length < len)
1937                        die("data is too large to use in this context");
1938
1939                while (n < length) {
1940                        size_t s = strbuf_fread(sb, length - n, stdin);
1941                        if (!s && feof(stdin))
1942                                die("EOF in data (%lu bytes remaining)",
1943                                        (unsigned long)(length - n));
1944                        n += s;
1945                }
1946        }
1947
1948        skip_optional_lf();
1949        return 1;
1950}
1951
1952static int validate_raw_date(const char *src, char *result, int maxlen)
1953{
1954        const char *orig_src = src;
1955        char *endp;
1956        unsigned long num;
1957
1958        errno = 0;
1959
1960        num = strtoul(src, &endp, 10);
1961        /* NEEDSWORK: perhaps check for reasonable values? */
1962        if (errno || endp == src || *endp != ' ')
1963                return -1;
1964
1965        src = endp + 1;
1966        if (*src != '-' && *src != '+')
1967                return -1;
1968
1969        num = strtoul(src + 1, &endp, 10);
1970        if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1971            1400 < num)
1972                return -1;
1973
1974        strcpy(result, orig_src);
1975        return 0;
1976}
1977
1978static char *parse_ident(const char *buf)
1979{
1980        const char *gt;
1981        size_t name_len;
1982        char *ident;
1983
1984        gt = strrchr(buf, '>');
1985        if (!gt)
1986                die("Missing > in ident string: %s", buf);
1987        gt++;
1988        if (*gt != ' ')
1989                die("Missing space after > in ident string: %s", buf);
1990        gt++;
1991        name_len = gt - buf;
1992        ident = xmalloc(name_len + 24);
1993        strncpy(ident, buf, name_len);
1994
1995        switch (whenspec) {
1996        case WHENSPEC_RAW:
1997                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1998                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1999                break;
2000        case WHENSPEC_RFC2822:
2001                if (parse_date(gt, ident + name_len, 24) < 0)
2002                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
2003                break;
2004        case WHENSPEC_NOW:
2005                if (strcmp("now", gt))
2006                        die("Date in ident must be 'now': %s", buf);
2007                datestamp(ident + name_len, 24);
2008                break;
2009        }
2010
2011        return ident;
2012}
2013
2014static void parse_and_store_blob(
2015        struct last_object *last,
2016        unsigned char *sha1out,
2017        uintmax_t mark)
2018{
2019        static struct strbuf buf = STRBUF_INIT;
2020        uintmax_t len;
2021
2022        if (parse_data(&buf, big_file_threshold, &len))
2023                store_object(OBJ_BLOB, &buf, last, sha1out, mark);
2024        else {
2025                if (last) {
2026                        strbuf_release(&last->data);
2027                        last->offset = 0;
2028                        last->depth = 0;
2029                }
2030                stream_blob(len, sha1out, mark);
2031                skip_optional_lf();
2032        }
2033}
2034
2035static void parse_new_blob(void)
2036{
2037        read_next_command();
2038        parse_mark();
2039        parse_and_store_blob(&last_blob, NULL, next_mark);
2040}
2041
2042static void unload_one_branch(void)
2043{
2044        while (cur_active_branches
2045                && cur_active_branches >= max_active_branches) {
2046                uintmax_t min_commit = ULONG_MAX;
2047                struct branch *e, *l = NULL, *p = NULL;
2048
2049                for (e = active_branches; e; e = e->active_next_branch) {
2050                        if (e->last_commit < min_commit) {
2051                                p = l;
2052                                min_commit = e->last_commit;
2053                        }
2054                        l = e;
2055                }
2056
2057                if (p) {
2058                        e = p->active_next_branch;
2059                        p->active_next_branch = e->active_next_branch;
2060                } else {
2061                        e = active_branches;
2062                        active_branches = e->active_next_branch;
2063                }
2064                e->active = 0;
2065                e->active_next_branch = NULL;
2066                if (e->branch_tree.tree) {
2067                        release_tree_content_recursive(e->branch_tree.tree);
2068                        e->branch_tree.tree = NULL;
2069                }
2070                cur_active_branches--;
2071        }
2072}
2073
2074static void load_branch(struct branch *b)
2075{
2076        load_tree(&b->branch_tree);
2077        if (!b->active) {
2078                b->active = 1;
2079                b->active_next_branch = active_branches;
2080                active_branches = b;
2081                cur_active_branches++;
2082                branch_load_count++;
2083        }
2084}
2085
2086static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2087{
2088        unsigned char fanout = 0;
2089        while ((num_notes >>= 8))
2090                fanout++;
2091        return fanout;
2092}
2093
2094static void construct_path_with_fanout(const char *hex_sha1,
2095                unsigned char fanout, char *path)
2096{
2097        unsigned int i = 0, j = 0;
2098        if (fanout >= 20)
2099                die("Too large fanout (%u)", fanout);
2100        while (fanout) {
2101                path[i++] = hex_sha1[j++];
2102                path[i++] = hex_sha1[j++];
2103                path[i++] = '/';
2104                fanout--;
2105        }
2106        memcpy(path + i, hex_sha1 + j, 40 - j);
2107        path[i + 40 - j] = '\0';
2108}
2109
2110static uintmax_t do_change_note_fanout(
2111                struct tree_entry *orig_root, struct tree_entry *root,
2112                char *hex_sha1, unsigned int hex_sha1_len,
2113                char *fullpath, unsigned int fullpath_len,
2114                unsigned char fanout)
2115{
2116        struct tree_content *t = root->tree;
2117        struct tree_entry *e, leaf;
2118        unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2119        uintmax_t num_notes = 0;
2120        unsigned char sha1[20];
2121        char realpath[60];
2122
2123        for (i = 0; t && i < t->entry_count; i++) {
2124                e = t->entries[i];
2125                tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2126                tmp_fullpath_len = fullpath_len;
2127
2128                /*
2129                 * We're interested in EITHER existing note entries (entries
2130                 * with exactly 40 hex chars in path, not including directory
2131                 * separators), OR directory entries that may contain note
2132                 * entries (with < 40 hex chars in path).
2133                 * Also, each path component in a note entry must be a multiple
2134                 * of 2 chars.
2135                 */
2136                if (!e->versions[1].mode ||
2137                    tmp_hex_sha1_len > 40 ||
2138                    e->name->str_len % 2)
2139                        continue;
2140
2141                /* This _may_ be a note entry, or a subdir containing notes */
2142                memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2143                       e->name->str_len);
2144                if (tmp_fullpath_len)
2145                        fullpath[tmp_fullpath_len++] = '/';
2146                memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2147                       e->name->str_len);
2148                tmp_fullpath_len += e->name->str_len;
2149                fullpath[tmp_fullpath_len] = '\0';
2150
2151                if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2152                        /* This is a note entry */
2153                        construct_path_with_fanout(hex_sha1, fanout, realpath);
2154                        if (!strcmp(fullpath, realpath)) {
2155                                /* Note entry is in correct location */
2156                                num_notes++;
2157                                continue;
2158                        }
2159
2160                        /* Rename fullpath to realpath */
2161                        if (!tree_content_remove(orig_root, fullpath, &leaf))
2162                                die("Failed to remove path %s", fullpath);
2163                        tree_content_set(orig_root, realpath,
2164                                leaf.versions[1].sha1,
2165                                leaf.versions[1].mode,
2166                                leaf.tree);
2167                } else if (S_ISDIR(e->versions[1].mode)) {
2168                        /* This is a subdir that may contain note entries */
2169                        if (!e->tree)
2170                                load_tree(e);
2171                        num_notes += do_change_note_fanout(orig_root, e,
2172                                hex_sha1, tmp_hex_sha1_len,
2173                                fullpath, tmp_fullpath_len, fanout);
2174                }
2175
2176                /* The above may have reallocated the current tree_content */
2177                t = root->tree;
2178        }
2179        return num_notes;
2180}
2181
2182static uintmax_t change_note_fanout(struct tree_entry *root,
2183                unsigned char fanout)
2184{
2185        char hex_sha1[40], path[60];
2186        return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2187}
2188
2189static void file_change_m(struct branch *b)
2190{
2191        const char *p = command_buf.buf + 2;
2192        static struct strbuf uq = STRBUF_INIT;
2193        const char *endp;
2194        struct object_entry *oe = oe;
2195        unsigned char sha1[20];
2196        uint16_t mode, inline_data = 0;
2197
2198        p = get_mode(p, &mode);
2199        if (!p)
2200                die("Corrupt mode: %s", command_buf.buf);
2201        switch (mode) {
2202        case 0644:
2203        case 0755:
2204                mode |= S_IFREG;
2205        case S_IFREG | 0644:
2206        case S_IFREG | 0755:
2207        case S_IFLNK:
2208        case S_IFDIR:
2209        case S_IFGITLINK:
2210                /* ok */
2211                break;
2212        default:
2213                die("Corrupt mode: %s", command_buf.buf);
2214        }
2215
2216        if (*p == ':') {
2217                char *x;
2218                oe = find_mark(strtoumax(p + 1, &x, 10));
2219                hashcpy(sha1, oe->idx.sha1);
2220                p = x;
2221        } else if (!prefixcmp(p, "inline")) {
2222                inline_data = 1;
2223                p += 6;
2224        } else {
2225                if (get_sha1_hex(p, sha1))
2226                        die("Invalid SHA1: %s", command_buf.buf);
2227                oe = find_object(sha1);
2228                p += 40;
2229        }
2230        if (*p++ != ' ')
2231                die("Missing space after SHA1: %s", command_buf.buf);
2232
2233        strbuf_reset(&uq);
2234        if (!unquote_c_style(&uq, p, &endp)) {
2235                if (*endp)
2236                        die("Garbage after path in: %s", command_buf.buf);
2237                p = uq.buf;
2238        }
2239
2240        /* Git does not track empty, non-toplevel directories. */
2241        if (S_ISDIR(mode) && !memcmp(sha1, EMPTY_TREE_SHA1_BIN, 20) && *p) {
2242                tree_content_remove(&b->branch_tree, p, NULL);
2243                return;
2244        }
2245
2246        if (S_ISGITLINK(mode)) {
2247                if (inline_data)
2248                        die("Git links cannot be specified 'inline': %s",
2249                                command_buf.buf);
2250                else if (oe) {
2251                        if (oe->type != OBJ_COMMIT)
2252                                die("Not a commit (actually a %s): %s",
2253                                        typename(oe->type), command_buf.buf);
2254                }
2255                /*
2256                 * Accept the sha1 without checking; it expected to be in
2257                 * another repository.
2258                 */
2259        } else if (inline_data) {
2260                if (S_ISDIR(mode))
2261                        die("Directories cannot be specified 'inline': %s",
2262                                command_buf.buf);
2263                if (p != uq.buf) {
2264                        strbuf_addstr(&uq, p);
2265                        p = uq.buf;
2266                }
2267                read_next_command();
2268                parse_and_store_blob(&last_blob, sha1, 0);
2269        } else {
2270                enum object_type expected = S_ISDIR(mode) ?
2271                                                OBJ_TREE: OBJ_BLOB;
2272                enum object_type type = oe ? oe->type :
2273                                        sha1_object_info(sha1, NULL);
2274                if (type < 0)
2275                        die("%s not found: %s",
2276                                        S_ISDIR(mode) ?  "Tree" : "Blob",
2277                                        command_buf.buf);
2278                if (type != expected)
2279                        die("Not a %s (actually a %s): %s",
2280                                typename(expected), typename(type),
2281                                command_buf.buf);
2282        }
2283
2284        if (!*p) {
2285                tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2286                return;
2287        }
2288        tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2289}
2290
2291static void file_change_d(struct branch *b)
2292{
2293        const char *p = command_buf.buf + 2;
2294        static struct strbuf uq = STRBUF_INIT;
2295        const char *endp;
2296
2297        strbuf_reset(&uq);
2298        if (!unquote_c_style(&uq, p, &endp)) {
2299                if (*endp)
2300                        die("Garbage after path in: %s", command_buf.buf);
2301                p = uq.buf;
2302        }
2303        tree_content_remove(&b->branch_tree, p, NULL);
2304}
2305
2306static void file_change_cr(struct branch *b, int rename)
2307{
2308        const char *s, *d;
2309        static struct strbuf s_uq = STRBUF_INIT;
2310        static struct strbuf d_uq = STRBUF_INIT;
2311        const char *endp;
2312        struct tree_entry leaf;
2313
2314        s = command_buf.buf + 2;
2315        strbuf_reset(&s_uq);
2316        if (!unquote_c_style(&s_uq, s, &endp)) {
2317                if (*endp != ' ')
2318                        die("Missing space after source: %s", command_buf.buf);
2319        } else {
2320                endp = strchr(s, ' ');
2321                if (!endp)
2322                        die("Missing space after source: %s", command_buf.buf);
2323                strbuf_add(&s_uq, s, endp - s);
2324        }
2325        s = s_uq.buf;
2326
2327        endp++;
2328        if (!*endp)
2329                die("Missing dest: %s", command_buf.buf);
2330
2331        d = endp;
2332        strbuf_reset(&d_uq);
2333        if (!unquote_c_style(&d_uq, d, &endp)) {
2334                if (*endp)
2335                        die("Garbage after dest in: %s", command_buf.buf);
2336                d = d_uq.buf;
2337        }
2338
2339        memset(&leaf, 0, sizeof(leaf));
2340        if (rename)
2341                tree_content_remove(&b->branch_tree, s, &leaf);
2342        else
2343                tree_content_get(&b->branch_tree, s, &leaf);
2344        if (!leaf.versions[1].mode)
2345                die("Path %s not in branch", s);
2346        if (!*d) {      /* C "path/to/subdir" "" */
2347                tree_content_replace(&b->branch_tree,
2348                        leaf.versions[1].sha1,
2349                        leaf.versions[1].mode,
2350                        leaf.tree);
2351                return;
2352        }
2353        tree_content_set(&b->branch_tree, d,
2354                leaf.versions[1].sha1,
2355                leaf.versions[1].mode,
2356                leaf.tree);
2357}
2358
2359static void note_change_n(struct branch *b, unsigned char old_fanout)
2360{
2361        const char *p = command_buf.buf + 2;
2362        static struct strbuf uq = STRBUF_INIT;
2363        struct object_entry *oe = oe;
2364        struct branch *s;
2365        unsigned char sha1[20], commit_sha1[20];
2366        char path[60];
2367        uint16_t inline_data = 0;
2368        unsigned char new_fanout;
2369
2370        /* <dataref> or 'inline' */
2371        if (*p == ':') {
2372                char *x;
2373                oe = find_mark(strtoumax(p + 1, &x, 10));
2374                hashcpy(sha1, oe->idx.sha1);
2375                p = x;
2376        } else if (!prefixcmp(p, "inline")) {
2377                inline_data = 1;
2378                p += 6;
2379        } else {
2380                if (get_sha1_hex(p, sha1))
2381                        die("Invalid SHA1: %s", command_buf.buf);
2382                oe = find_object(sha1);
2383                p += 40;
2384        }
2385        if (*p++ != ' ')
2386                die("Missing space after SHA1: %s", command_buf.buf);
2387
2388        /* <committish> */
2389        s = lookup_branch(p);
2390        if (s) {
2391                hashcpy(commit_sha1, s->sha1);
2392        } else if (*p == ':') {
2393                uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2394                struct object_entry *commit_oe = find_mark(commit_mark);
2395                if (commit_oe->type != OBJ_COMMIT)
2396                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2397                hashcpy(commit_sha1, commit_oe->idx.sha1);
2398        } else if (!get_sha1(p, commit_sha1)) {
2399                unsigned long size;
2400                char *buf = read_object_with_reference(commit_sha1,
2401                        commit_type, &size, commit_sha1);
2402                if (!buf || size < 46)
2403                        die("Not a valid commit: %s", p);
2404                free(buf);
2405        } else
2406                die("Invalid ref name or SHA1 expression: %s", p);
2407
2408        if (inline_data) {
2409                if (p != uq.buf) {
2410                        strbuf_addstr(&uq, p);
2411                        p = uq.buf;
2412                }
2413                read_next_command();
2414                parse_and_store_blob(&last_blob, sha1, 0);
2415        } else if (oe) {
2416                if (oe->type != OBJ_BLOB)
2417                        die("Not a blob (actually a %s): %s",
2418                                typename(oe->type), command_buf.buf);
2419        } else if (!is_null_sha1(sha1)) {
2420                enum object_type type = sha1_object_info(sha1, NULL);
2421                if (type < 0)
2422                        die("Blob not found: %s", command_buf.buf);
2423                if (type != OBJ_BLOB)
2424                        die("Not a blob (actually a %s): %s",
2425                            typename(type), command_buf.buf);
2426        }
2427
2428        construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2429        if (tree_content_remove(&b->branch_tree, path, NULL))
2430                b->num_notes--;
2431
2432        if (is_null_sha1(sha1))
2433                return; /* nothing to insert */
2434
2435        b->num_notes++;
2436        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2437        construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2438        tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2439}
2440
2441static void file_change_deleteall(struct branch *b)
2442{
2443        release_tree_content_recursive(b->branch_tree.tree);
2444        hashclr(b->branch_tree.versions[0].sha1);
2445        hashclr(b->branch_tree.versions[1].sha1);
2446        load_tree(&b->branch_tree);
2447        b->num_notes = 0;
2448}
2449
2450static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2451{
2452        if (!buf || size < 46)
2453                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2454        if (memcmp("tree ", buf, 5)
2455                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2456                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2457        hashcpy(b->branch_tree.versions[0].sha1,
2458                b->branch_tree.versions[1].sha1);
2459}
2460
2461static void parse_from_existing(struct branch *b)
2462{
2463        if (is_null_sha1(b->sha1)) {
2464                hashclr(b->branch_tree.versions[0].sha1);
2465                hashclr(b->branch_tree.versions[1].sha1);
2466        } else {
2467                unsigned long size;
2468                char *buf;
2469
2470                buf = read_object_with_reference(b->sha1,
2471                        commit_type, &size, b->sha1);
2472                parse_from_commit(b, buf, size);
2473                free(buf);
2474        }
2475}
2476
2477static int parse_from(struct branch *b)
2478{
2479        const char *from;
2480        struct branch *s;
2481
2482        if (prefixcmp(command_buf.buf, "from "))
2483                return 0;
2484
2485        if (b->branch_tree.tree) {
2486                release_tree_content_recursive(b->branch_tree.tree);
2487                b->branch_tree.tree = NULL;
2488        }
2489
2490        from = strchr(command_buf.buf, ' ') + 1;
2491        s = lookup_branch(from);
2492        if (b == s)
2493                die("Can't create a branch from itself: %s", b->name);
2494        else if (s) {
2495                unsigned char *t = s->branch_tree.versions[1].sha1;
2496                hashcpy(b->sha1, s->sha1);
2497                hashcpy(b->branch_tree.versions[0].sha1, t);
2498                hashcpy(b->branch_tree.versions[1].sha1, t);
2499        } else if (*from == ':') {
2500                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2501                struct object_entry *oe = find_mark(idnum);
2502                if (oe->type != OBJ_COMMIT)
2503                        die("Mark :%" PRIuMAX " not a commit", idnum);
2504                hashcpy(b->sha1, oe->idx.sha1);
2505                if (oe->pack_id != MAX_PACK_ID) {
2506                        unsigned long size;
2507                        char *buf = gfi_unpack_entry(oe, &size);
2508                        parse_from_commit(b, buf, size);
2509                        free(buf);
2510                } else
2511                        parse_from_existing(b);
2512        } else if (!get_sha1(from, b->sha1))
2513                parse_from_existing(b);
2514        else
2515                die("Invalid ref name or SHA1 expression: %s", from);
2516
2517        read_next_command();
2518        return 1;
2519}
2520
2521static struct hash_list *parse_merge(unsigned int *count)
2522{
2523        struct hash_list *list = NULL, *n, *e = e;
2524        const char *from;
2525        struct branch *s;
2526
2527        *count = 0;
2528        while (!prefixcmp(command_buf.buf, "merge ")) {
2529                from = strchr(command_buf.buf, ' ') + 1;
2530                n = xmalloc(sizeof(*n));
2531                s = lookup_branch(from);
2532                if (s)
2533                        hashcpy(n->sha1, s->sha1);
2534                else if (*from == ':') {
2535                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2536                        struct object_entry *oe = find_mark(idnum);
2537                        if (oe->type != OBJ_COMMIT)
2538                                die("Mark :%" PRIuMAX " not a commit", idnum);
2539                        hashcpy(n->sha1, oe->idx.sha1);
2540                } else if (!get_sha1(from, n->sha1)) {
2541                        unsigned long size;
2542                        char *buf = read_object_with_reference(n->sha1,
2543                                commit_type, &size, n->sha1);
2544                        if (!buf || size < 46)
2545                                die("Not a valid commit: %s", from);
2546                        free(buf);
2547                } else
2548                        die("Invalid ref name or SHA1 expression: %s", from);
2549
2550                n->next = NULL;
2551                if (list)
2552                        e->next = n;
2553                else
2554                        list = n;
2555                e = n;
2556                (*count)++;
2557                read_next_command();
2558        }
2559        return list;
2560}
2561
2562static void parse_new_commit(void)
2563{
2564        static struct strbuf msg = STRBUF_INIT;
2565        struct branch *b;
2566        char *sp;
2567        char *author = NULL;
2568        char *committer = NULL;
2569        struct hash_list *merge_list = NULL;
2570        unsigned int merge_count;
2571        unsigned char prev_fanout, new_fanout;
2572
2573        /* Obtain the branch name from the rest of our command */
2574        sp = strchr(command_buf.buf, ' ') + 1;
2575        b = lookup_branch(sp);
2576        if (!b)
2577                b = new_branch(sp);
2578
2579        read_next_command();
2580        parse_mark();
2581        if (!prefixcmp(command_buf.buf, "author ")) {
2582                author = parse_ident(command_buf.buf + 7);
2583                read_next_command();
2584        }
2585        if (!prefixcmp(command_buf.buf, "committer ")) {
2586                committer = parse_ident(command_buf.buf + 10);
2587                read_next_command();
2588        }
2589        if (!committer)
2590                die("Expected committer but didn't get one");
2591        parse_data(&msg, 0, NULL);
2592        read_next_command();
2593        parse_from(b);
2594        merge_list = parse_merge(&merge_count);
2595
2596        /* ensure the branch is active/loaded */
2597        if (!b->branch_tree.tree || !max_active_branches) {
2598                unload_one_branch();
2599                load_branch(b);
2600        }
2601
2602        prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2603
2604        /* file_change* */
2605        while (command_buf.len > 0) {
2606                if (!prefixcmp(command_buf.buf, "M "))
2607                        file_change_m(b);
2608                else if (!prefixcmp(command_buf.buf, "D "))
2609                        file_change_d(b);
2610                else if (!prefixcmp(command_buf.buf, "R "))
2611                        file_change_cr(b, 1);
2612                else if (!prefixcmp(command_buf.buf, "C "))
2613                        file_change_cr(b, 0);
2614                else if (!prefixcmp(command_buf.buf, "N "))
2615                        note_change_n(b, prev_fanout);
2616                else if (!strcmp("deleteall", command_buf.buf))
2617                        file_change_deleteall(b);
2618                else {
2619                        unread_command_buf = 1;
2620                        break;
2621                }
2622                if (read_next_command() == EOF)
2623                        break;
2624        }
2625
2626        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2627        if (new_fanout != prev_fanout)
2628                b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2629
2630        /* build the tree and the commit */
2631        store_tree(&b->branch_tree);
2632        hashcpy(b->branch_tree.versions[0].sha1,
2633                b->branch_tree.versions[1].sha1);
2634
2635        strbuf_reset(&new_data);
2636        strbuf_addf(&new_data, "tree %s\n",
2637                sha1_to_hex(b->branch_tree.versions[1].sha1));
2638        if (!is_null_sha1(b->sha1))
2639                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2640        while (merge_list) {
2641                struct hash_list *next = merge_list->next;
2642                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2643                free(merge_list);
2644                merge_list = next;
2645        }
2646        strbuf_addf(&new_data,
2647                "author %s\n"
2648                "committer %s\n"
2649                "\n",
2650                author ? author : committer, committer);
2651        strbuf_addbuf(&new_data, &msg);
2652        free(author);
2653        free(committer);
2654
2655        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2656                b->pack_id = pack_id;
2657        b->last_commit = object_count_by_type[OBJ_COMMIT];
2658}
2659
2660static void parse_new_tag(void)
2661{
2662        static struct strbuf msg = STRBUF_INIT;
2663        char *sp;
2664        const char *from;
2665        char *tagger;
2666        struct branch *s;
2667        struct tag *t;
2668        uintmax_t from_mark = 0;
2669        unsigned char sha1[20];
2670        enum object_type type;
2671
2672        /* Obtain the new tag name from the rest of our command */
2673        sp = strchr(command_buf.buf, ' ') + 1;
2674        t = pool_alloc(sizeof(struct tag));
2675        t->next_tag = NULL;
2676        t->name = pool_strdup(sp);
2677        if (last_tag)
2678                last_tag->next_tag = t;
2679        else
2680                first_tag = t;
2681        last_tag = t;
2682        read_next_command();
2683
2684        /* from ... */
2685        if (prefixcmp(command_buf.buf, "from "))
2686                die("Expected from command, got %s", command_buf.buf);
2687        from = strchr(command_buf.buf, ' ') + 1;
2688        s = lookup_branch(from);
2689        if (s) {
2690                hashcpy(sha1, s->sha1);
2691                type = OBJ_COMMIT;
2692        } else if (*from == ':') {
2693                struct object_entry *oe;
2694                from_mark = strtoumax(from + 1, NULL, 10);
2695                oe = find_mark(from_mark);
2696                type = oe->type;
2697                hashcpy(sha1, oe->idx.sha1);
2698        } else if (!get_sha1(from, sha1)) {
2699                unsigned long size;
2700                char *buf;
2701
2702                buf = read_sha1_file(sha1, &type, &size);
2703                if (!buf || size < 46)
2704                        die("Not a valid commit: %s", from);
2705                free(buf);
2706        } else
2707                die("Invalid ref name or SHA1 expression: %s", from);
2708        read_next_command();
2709
2710        /* tagger ... */
2711        if (!prefixcmp(command_buf.buf, "tagger ")) {
2712                tagger = parse_ident(command_buf.buf + 7);
2713                read_next_command();
2714        } else
2715                tagger = NULL;
2716
2717        /* tag payload/message */
2718        parse_data(&msg, 0, NULL);
2719
2720        /* build the tag object */
2721        strbuf_reset(&new_data);
2722
2723        strbuf_addf(&new_data,
2724                    "object %s\n"
2725                    "type %s\n"
2726                    "tag %s\n",
2727                    sha1_to_hex(sha1), typename(type), t->name);
2728        if (tagger)
2729                strbuf_addf(&new_data,
2730                            "tagger %s\n", tagger);
2731        strbuf_addch(&new_data, '\n');
2732        strbuf_addbuf(&new_data, &msg);
2733        free(tagger);
2734
2735        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2736                t->pack_id = MAX_PACK_ID;
2737        else
2738                t->pack_id = pack_id;
2739}
2740
2741static void parse_reset_branch(void)
2742{
2743        struct branch *b;
2744        char *sp;
2745
2746        /* Obtain the branch name from the rest of our command */
2747        sp = strchr(command_buf.buf, ' ') + 1;
2748        b = lookup_branch(sp);
2749        if (b) {
2750                hashclr(b->sha1);
2751                hashclr(b->branch_tree.versions[0].sha1);
2752                hashclr(b->branch_tree.versions[1].sha1);
2753                if (b->branch_tree.tree) {
2754                        release_tree_content_recursive(b->branch_tree.tree);
2755                        b->branch_tree.tree = NULL;
2756                }
2757        }
2758        else
2759                b = new_branch(sp);
2760        read_next_command();
2761        parse_from(b);
2762        if (command_buf.len > 0)
2763                unread_command_buf = 1;
2764}
2765
2766static void cat_blob_write(const char *buf, unsigned long size)
2767{
2768        if (write_in_full(cat_blob_fd, buf, size) != size)
2769                die_errno("Write to frontend failed");
2770}
2771
2772static void cat_blob(struct object_entry *oe, unsigned char sha1[20])
2773{
2774        struct strbuf line = STRBUF_INIT;
2775        unsigned long size;
2776        enum object_type type = 0;
2777        char *buf;
2778
2779        if (!oe || oe->pack_id == MAX_PACK_ID) {
2780                buf = read_sha1_file(sha1, &type, &size);
2781        } else {
2782                type = oe->type;
2783                buf = gfi_unpack_entry(oe, &size);
2784        }
2785
2786        /*
2787         * Output based on batch_one_object() from cat-file.c.
2788         */
2789        if (type <= 0) {
2790                strbuf_reset(&line);
2791                strbuf_addf(&line, "%s missing\n", sha1_to_hex(sha1));
2792                cat_blob_write(line.buf, line.len);
2793                strbuf_release(&line);
2794                free(buf);
2795                return;
2796        }
2797        if (!buf)
2798                die("Can't read object %s", sha1_to_hex(sha1));
2799        if (type != OBJ_BLOB)
2800                die("Object %s is a %s but a blob was expected.",
2801                    sha1_to_hex(sha1), typename(type));
2802        strbuf_reset(&line);
2803        strbuf_addf(&line, "%s %s %lu\n", sha1_to_hex(sha1),
2804                                                typename(type), size);
2805        cat_blob_write(line.buf, line.len);
2806        strbuf_release(&line);
2807        cat_blob_write(buf, size);
2808        cat_blob_write("\n", 1);
2809        free(buf);
2810}
2811
2812static void parse_cat_blob(void)
2813{
2814        const char *p;
2815        struct object_entry *oe = oe;
2816        unsigned char sha1[20];
2817
2818        /* cat-blob SP <object> LF */
2819        p = command_buf.buf + strlen("cat-blob ");
2820        if (*p == ':') {
2821                char *x;
2822                oe = find_mark(strtoumax(p + 1, &x, 10));
2823                if (x == p + 1)
2824                        die("Invalid mark: %s", command_buf.buf);
2825                if (!oe)
2826                        die("Unknown mark: %s", command_buf.buf);
2827                if (*x)
2828                        die("Garbage after mark: %s", command_buf.buf);
2829                hashcpy(sha1, oe->idx.sha1);
2830        } else {
2831                if (get_sha1_hex(p, sha1))
2832                        die("Invalid SHA1: %s", command_buf.buf);
2833                if (p[40])
2834                        die("Garbage after SHA1: %s", command_buf.buf);
2835                oe = find_object(sha1);
2836        }
2837
2838        cat_blob(oe, sha1);
2839}
2840
2841static void checkpoint(void)
2842{
2843        checkpoint_requested = 0;
2844        if (object_count) {
2845                cycle_packfile();
2846                dump_branches();
2847                dump_tags();
2848                dump_marks();
2849        }
2850}
2851
2852static void parse_checkpoint(void)
2853{
2854        checkpoint_requested = 1;
2855        skip_optional_lf();
2856}
2857
2858static void parse_progress(void)
2859{
2860        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2861        fputc('\n', stdout);
2862        fflush(stdout);
2863        skip_optional_lf();
2864}
2865
2866static char* make_fast_import_path(const char *path)
2867{
2868        struct strbuf abs_path = STRBUF_INIT;
2869
2870        if (!relative_marks_paths || is_absolute_path(path))
2871                return xstrdup(path);
2872        strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
2873        return strbuf_detach(&abs_path, NULL);
2874}
2875
2876static void option_import_marks(const char *marks,
2877                                        int from_stream, int ignore_missing)
2878{
2879        if (import_marks_file) {
2880                if (from_stream)
2881                        die("Only one import-marks command allowed per stream");
2882
2883                /* read previous mark file */
2884                if(!import_marks_file_from_stream)
2885                        read_marks();
2886        }
2887
2888        import_marks_file = make_fast_import_path(marks);
2889        safe_create_leading_directories_const(import_marks_file);
2890        import_marks_file_from_stream = from_stream;
2891        import_marks_file_ignore_missing = ignore_missing;
2892}
2893
2894static void option_date_format(const char *fmt)
2895{
2896        if (!strcmp(fmt, "raw"))
2897                whenspec = WHENSPEC_RAW;
2898        else if (!strcmp(fmt, "rfc2822"))
2899                whenspec = WHENSPEC_RFC2822;
2900        else if (!strcmp(fmt, "now"))
2901                whenspec = WHENSPEC_NOW;
2902        else
2903                die("unknown --date-format argument %s", fmt);
2904}
2905
2906static unsigned long ulong_arg(const char *option, const char *arg)
2907{
2908        char *endptr;
2909        unsigned long rv = strtoul(arg, &endptr, 0);
2910        if (strchr(arg, '-') || endptr == arg || *endptr)
2911                die("%s: argument must be a non-negative integer", option);
2912        return rv;
2913}
2914
2915static void option_depth(const char *depth)
2916{
2917        max_depth = ulong_arg("--depth", depth);
2918        if (max_depth > MAX_DEPTH)
2919                die("--depth cannot exceed %u", MAX_DEPTH);
2920}
2921
2922static void option_active_branches(const char *branches)
2923{
2924        max_active_branches = ulong_arg("--active-branches", branches);
2925}
2926
2927static void option_export_marks(const char *marks)
2928{
2929        export_marks_file = make_fast_import_path(marks);
2930        safe_create_leading_directories_const(export_marks_file);
2931}
2932
2933static void option_cat_blob_fd(const char *fd)
2934{
2935        unsigned long n = ulong_arg("--cat-blob-fd", fd);
2936        if (n > (unsigned long) INT_MAX)
2937                die("--cat-blob-fd cannot exceed %d", INT_MAX);
2938        cat_blob_fd = (int) n;
2939}
2940
2941static void option_export_pack_edges(const char *edges)
2942{
2943        if (pack_edges)
2944                fclose(pack_edges);
2945        pack_edges = fopen(edges, "a");
2946        if (!pack_edges)
2947                die_errno("Cannot open '%s'", edges);
2948}
2949
2950static int parse_one_option(const char *option)
2951{
2952        if (!prefixcmp(option, "max-pack-size=")) {
2953                unsigned long v;
2954                if (!git_parse_ulong(option + 14, &v))
2955                        return 0;
2956                if (v < 8192) {
2957                        warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
2958                        v *= 1024 * 1024;
2959                } else if (v < 1024 * 1024) {
2960                        warning("minimum max-pack-size is 1 MiB");
2961                        v = 1024 * 1024;
2962                }
2963                max_packsize = v;
2964        } else if (!prefixcmp(option, "big-file-threshold=")) {
2965                unsigned long v;
2966                if (!git_parse_ulong(option + 19, &v))
2967                        return 0;
2968                big_file_threshold = v;
2969        } else if (!prefixcmp(option, "depth=")) {
2970                option_depth(option + 6);
2971        } else if (!prefixcmp(option, "active-branches=")) {
2972                option_active_branches(option + 16);
2973        } else if (!prefixcmp(option, "export-pack-edges=")) {
2974                option_export_pack_edges(option + 18);
2975        } else if (!prefixcmp(option, "quiet")) {
2976                show_stats = 0;
2977        } else if (!prefixcmp(option, "stats")) {
2978                show_stats = 1;
2979        } else {
2980                return 0;
2981        }
2982
2983        return 1;
2984}
2985
2986static int parse_one_feature(const char *feature, int from_stream)
2987{
2988        if (!prefixcmp(feature, "date-format=")) {
2989                option_date_format(feature + 12);
2990        } else if (!prefixcmp(feature, "import-marks=")) {
2991                option_import_marks(feature + 13, from_stream, 0);
2992        } else if (!prefixcmp(feature, "import-marks-if-exists=")) {
2993                option_import_marks(feature + strlen("import-marks-if-exists="),
2994                                        from_stream, 1);
2995        } else if (!prefixcmp(feature, "export-marks=")) {
2996                option_export_marks(feature + 13);
2997        } else if (!strcmp(feature, "cat-blob")) {
2998                ; /* Don't die - this feature is supported */
2999        } else if (!prefixcmp(feature, "relative-marks")) {
3000                relative_marks_paths = 1;
3001        } else if (!prefixcmp(feature, "no-relative-marks")) {
3002                relative_marks_paths = 0;
3003        } else if (!prefixcmp(feature, "force")) {
3004                force_update = 1;
3005        } else if (!strcmp(feature, "notes")) {
3006                ; /* do nothing; we have the feature */
3007        } else {
3008                return 0;
3009        }
3010
3011        return 1;
3012}
3013
3014static void parse_feature(void)
3015{
3016        char *feature = command_buf.buf + 8;
3017
3018        if (seen_data_command)
3019                die("Got feature command '%s' after data command", feature);
3020
3021        if (parse_one_feature(feature, 1))
3022                return;
3023
3024        die("This version of fast-import does not support feature %s.", feature);
3025}
3026
3027static void parse_option(void)
3028{
3029        char *option = command_buf.buf + 11;
3030
3031        if (seen_data_command)
3032                die("Got option command '%s' after data command", option);
3033
3034        if (parse_one_option(option))
3035                return;
3036
3037        die("This version of fast-import does not support option: %s", option);
3038}
3039
3040static int git_pack_config(const char *k, const char *v, void *cb)
3041{
3042        if (!strcmp(k, "pack.depth")) {
3043                max_depth = git_config_int(k, v);
3044                if (max_depth > MAX_DEPTH)
3045                        max_depth = MAX_DEPTH;
3046                return 0;
3047        }
3048        if (!strcmp(k, "pack.compression")) {
3049                int level = git_config_int(k, v);
3050                if (level == -1)
3051                        level = Z_DEFAULT_COMPRESSION;
3052                else if (level < 0 || level > Z_BEST_COMPRESSION)
3053                        die("bad pack compression level %d", level);
3054                pack_compression_level = level;
3055                pack_compression_seen = 1;
3056                return 0;
3057        }
3058        if (!strcmp(k, "pack.indexversion")) {
3059                pack_idx_opts.version = git_config_int(k, v);
3060                if (pack_idx_opts.version > 2)
3061                        die("bad pack.indexversion=%"PRIu32,
3062                            pack_idx_opts.version);
3063                return 0;
3064        }
3065        if (!strcmp(k, "pack.packsizelimit")) {
3066                max_packsize = git_config_ulong(k, v);
3067                return 0;
3068        }
3069        if (!strcmp(k, "core.bigfilethreshold")) {
3070                long n = git_config_int(k, v);
3071                big_file_threshold = 0 < n ? n : 0;
3072        }
3073        return git_default_config(k, v, cb);
3074}
3075
3076static const char fast_import_usage[] =
3077"git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3078
3079static void parse_argv(void)
3080{
3081        unsigned int i;
3082
3083        for (i = 1; i < global_argc; i++) {
3084                const char *a = global_argv[i];
3085
3086                if (*a != '-' || !strcmp(a, "--"))
3087                        break;
3088
3089                if (parse_one_option(a + 2))
3090                        continue;
3091
3092                if (parse_one_feature(a + 2, 0))
3093                        continue;
3094
3095                if (!prefixcmp(a + 2, "cat-blob-fd=")) {
3096                        option_cat_blob_fd(a + 2 + strlen("cat-blob-fd="));
3097                        continue;
3098                }
3099
3100                die("unknown option %s", a);
3101        }
3102        if (i != global_argc)
3103                usage(fast_import_usage);
3104
3105        seen_data_command = 1;
3106        if (import_marks_file)
3107                read_marks();
3108}
3109
3110int main(int argc, const char **argv)
3111{
3112        unsigned int i;
3113
3114        git_extract_argv0_path(argv[0]);
3115
3116        if (argc == 2 && !strcmp(argv[1], "-h"))
3117                usage(fast_import_usage);
3118
3119        setup_git_directory();
3120        reset_pack_idx_option(&pack_idx_opts);
3121        git_config(git_pack_config, NULL);
3122        if (!pack_compression_seen && core_compression_seen)
3123                pack_compression_level = core_compression_level;
3124
3125        alloc_objects(object_entry_alloc);
3126        strbuf_init(&command_buf, 0);
3127        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3128        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3129        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3130        marks = pool_calloc(1, sizeof(struct mark_set));
3131
3132        global_argc = argc;
3133        global_argv = argv;
3134
3135        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
3136        for (i = 0; i < (cmd_save - 1); i++)
3137                rc_free[i].next = &rc_free[i + 1];
3138        rc_free[cmd_save - 1].next = NULL;
3139
3140        prepare_packed_git();
3141        start_packfile();
3142        set_die_routine(die_nicely);
3143        set_checkpoint_signal();
3144        while (read_next_command() != EOF) {
3145                if (!strcmp("blob", command_buf.buf))
3146                        parse_new_blob();
3147                else if (!prefixcmp(command_buf.buf, "commit "))
3148                        parse_new_commit();
3149                else if (!prefixcmp(command_buf.buf, "tag "))
3150                        parse_new_tag();
3151                else if (!prefixcmp(command_buf.buf, "reset "))
3152                        parse_reset_branch();
3153                else if (!strcmp("checkpoint", command_buf.buf))
3154                        parse_checkpoint();
3155                else if (!prefixcmp(command_buf.buf, "progress "))
3156                        parse_progress();
3157                else if (!prefixcmp(command_buf.buf, "feature "))
3158                        parse_feature();
3159                else if (!prefixcmp(command_buf.buf, "option git "))
3160                        parse_option();
3161                else if (!prefixcmp(command_buf.buf, "option "))
3162                        /* ignore non-git options*/;
3163                else
3164                        die("Unsupported command: %s", command_buf.buf);
3165
3166                if (checkpoint_requested)
3167                        checkpoint();
3168        }
3169
3170        /* argv hasn't been parsed yet, do so */
3171        if (!seen_data_command)
3172                parse_argv();
3173
3174        end_packfile();
3175
3176        dump_branches();
3177        dump_tags();
3178        unkeep_all_packs();
3179        dump_marks();
3180
3181        if (pack_edges)
3182                fclose(pack_edges);
3183
3184        if (show_stats) {
3185                uintmax_t total_count = 0, duplicate_count = 0;
3186                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3187                        total_count += object_count_by_type[i];
3188                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3189                        duplicate_count += duplicate_count_by_type[i];
3190
3191                fprintf(stderr, "%s statistics:\n", argv[0]);
3192                fprintf(stderr, "---------------------------------------------------------------------\n");
3193                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3194                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3195                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]);
3196                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]);
3197                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]);
3198                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]);
3199                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3200                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3201                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3202                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3203                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3204                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3205                fprintf(stderr, "---------------------------------------------------------------------\n");
3206                pack_report();
3207                fprintf(stderr, "---------------------------------------------------------------------\n");
3208                fprintf(stderr, "\n");
3209        }
3210
3211        return failure ? 1 : 0;
3212}