bbac58cae97b3587af5eb2fdf198f250be8005be
   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 commit-ish lf)?
  26    ('merge' sp commit-ish lf)*
  27    (file_change | ls)*
  28    lf?;
  29  commit_msg ::= data;
  30
  31  ls ::= 'ls' sp '"' quoted(path) '"' lf;
  32
  33  file_change ::= file_clr
  34    | file_del
  35    | file_rnm
  36    | file_cpy
  37    | file_obm
  38    | file_inm;
  39  file_clr ::= 'deleteall' lf;
  40  file_del ::= 'D' sp path_str lf;
  41  file_rnm ::= 'R' sp path_str sp path_str lf;
  42  file_cpy ::= 'C' sp path_str sp path_str lf;
  43  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  44  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  45    data;
  46  note_obm ::= 'N' sp (hexsha1 | idnum) sp commit-ish lf;
  47  note_inm ::= 'N' sp 'inline' sp commit-ish lf
  48    data;
  49
  50  new_tag ::= 'tag' sp tag_str lf
  51    'from' sp commit-ish lf
  52    ('tagger' (sp name)? sp '<' email '>' sp when lf)?
  53    tag_msg;
  54  tag_msg ::= data;
  55
  56  reset_branch ::= 'reset' sp ref_str lf
  57    ('from' sp commit-ish lf)?
  58    lf?;
  59
  60  checkpoint ::= 'checkpoint' lf
  61    lf?;
  62
  63  progress ::= 'progress' sp not_lf* lf
  64    lf?;
  65
  66     # note: the first idnum in a stream should be 1 and subsequent
  67     # idnums should not have gaps between values as this will cause
  68     # the stream parser to reserve space for the gapped values.  An
  69     # idnum can be updated in the future to a new object by issuing
  70     # a new mark directive with the old idnum.
  71     #
  72  mark ::= 'mark' sp idnum lf;
  73  data ::= (delimited_data | exact_data)
  74    lf?;
  75
  76    # note: delim may be any string but must not contain lf.
  77    # data_line may contain any data but must not be exactly
  78    # delim.
  79  delimited_data ::= 'data' sp '<<' delim lf
  80    (data_line lf)*
  81    delim lf;
  82
  83     # note: declen indicates the length of binary_data in bytes.
  84     # declen does not include the lf preceding the binary data.
  85     #
  86  exact_data ::= 'data' sp declen lf
  87    binary_data;
  88
  89     # note: quoted strings are C-style quoting supporting \c for
  90     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  91     # is the signed byte value in octal.  Note that the only
  92     # characters which must actually be escaped to protect the
  93     # stream formatting is: \, " and LF.  Otherwise these values
  94     # are UTF8.
  95     #
  96  commit-ish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
  97  ref_str     ::= ref;
  98  sha1exp_str ::= sha1exp;
  99  tag_str     ::= tag;
 100  path_str    ::= path    | '"' quoted(path)    '"' ;
 101  mode        ::= '100644' | '644'
 102                | '100755' | '755'
 103                | '120000'
 104                ;
 105
 106  declen ::= # unsigned 32 bit value, ascii base10 notation;
 107  bigint ::= # unsigned integer value, ascii base10 notation;
 108  binary_data ::= # file content, not interpreted;
 109
 110  when         ::= raw_when | rfc2822_when;
 111  raw_when     ::= ts sp tz;
 112  rfc2822_when ::= # Valid RFC 2822 date and time;
 113
 114  sp ::= # ASCII space character;
 115  lf ::= # ASCII newline (LF) character;
 116
 117     # note: a colon (':') must precede the numerical value assigned to
 118     # an idnum.  This is to distinguish it from a ref or tag name as
 119     # GIT does not permit ':' in ref or tag strings.
 120     #
 121  idnum   ::= ':' bigint;
 122  path    ::= # GIT style file path, e.g. "a/b/c";
 123  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 124  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 125  sha1exp ::= # Any valid GIT SHA1 expression;
 126  hexsha1 ::= # SHA1 in hexadecimal format;
 127
 128     # note: name and email are UTF8 strings, however name must not
 129     # contain '<' or lf and email must not contain any of the
 130     # following: '<', '>', lf.
 131     #
 132  name  ::= # valid GIT author/committer name;
 133  email ::= # valid GIT author/committer email;
 134  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 135  tz    ::= # GIT style timezone;
 136
 137     # note: comments, get-mark, ls-tree, and cat-blob requests may
 138     # appear anywhere in the input, except within a data command. Any
 139     # form of the data command always escapes the related input from
 140     # comment processing.
 141     #
 142     # In case it is not clear, the '#' that starts the comment
 143     # must be the first character on that line (an lf
 144     # preceded it).
 145     #
 146
 147  get_mark ::= 'get-mark' sp idnum lf;
 148  cat_blob ::= 'cat-blob' sp (hexsha1 | idnum) lf;
 149  ls_tree  ::= 'ls' sp (hexsha1 | idnum) sp path_str lf;
 150
 151  comment ::= '#' not_lf* lf;
 152  not_lf  ::= # Any byte that is not ASCII newline (LF);
 153*/
 154
 155#include "builtin.h"
 156#include "cache.h"
 157#include "config.h"
 158#include "lockfile.h"
 159#include "object.h"
 160#include "blob.h"
 161#include "tree.h"
 162#include "commit.h"
 163#include "delta.h"
 164#include "pack.h"
 165#include "refs.h"
 166#include "csum-file.h"
 167#include "quote.h"
 168#include "dir.h"
 169#include "run-command.h"
 170#include "packfile.h"
 171
 172#define PACK_ID_BITS 16
 173#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 174#define DEPTH_BITS 13
 175#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 176
 177/*
 178 * We abuse the setuid bit on directories to mean "do not delta".
 179 */
 180#define NO_DELTA S_ISUID
 181
 182struct object_entry {
 183        struct pack_idx_entry idx;
 184        struct object_entry *next;
 185        uint32_t type : TYPE_BITS,
 186                pack_id : PACK_ID_BITS,
 187                depth : DEPTH_BITS;
 188};
 189
 190struct object_entry_pool {
 191        struct object_entry_pool *next_pool;
 192        struct object_entry *next_free;
 193        struct object_entry *end;
 194        struct object_entry entries[FLEX_ARRAY]; /* more */
 195};
 196
 197struct mark_set {
 198        union {
 199                struct object_entry *marked[1024];
 200                struct mark_set *sets[1024];
 201        } data;
 202        unsigned int shift;
 203};
 204
 205struct last_object {
 206        struct strbuf data;
 207        off_t offset;
 208        unsigned int depth;
 209        unsigned no_swap : 1;
 210};
 211
 212struct mp_block {
 213        struct mp_block *next_block;
 214        char *next_free;
 215        char *end;
 216        uintmax_t space[FLEX_ARRAY]; /* more */
 217};
 218
 219struct mem_pool {
 220        struct mp_block *mp_block;
 221
 222        /*
 223         * The amount of available memory to grow the pool by.
 224         * This size does not include the overhead for the mp_block.
 225         */
 226        size_t block_alloc;
 227
 228        /* The total amount of memory allocated by the pool. */
 229        size_t pool_alloc;
 230};
 231
 232struct atom_str {
 233        struct atom_str *next_atom;
 234        unsigned short str_len;
 235        char str_dat[FLEX_ARRAY]; /* more */
 236};
 237
 238struct tree_content;
 239struct tree_entry {
 240        struct tree_content *tree;
 241        struct atom_str *name;
 242        struct tree_entry_ms {
 243                uint16_t mode;
 244                struct object_id oid;
 245        } versions[2];
 246};
 247
 248struct tree_content {
 249        unsigned int entry_capacity; /* must match avail_tree_content */
 250        unsigned int entry_count;
 251        unsigned int delta_depth;
 252        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 253};
 254
 255struct avail_tree_content {
 256        unsigned int entry_capacity; /* must match tree_content */
 257        struct avail_tree_content *next_avail;
 258};
 259
 260struct branch {
 261        struct branch *table_next_branch;
 262        struct branch *active_next_branch;
 263        const char *name;
 264        struct tree_entry branch_tree;
 265        uintmax_t last_commit;
 266        uintmax_t num_notes;
 267        unsigned active : 1;
 268        unsigned delete : 1;
 269        unsigned pack_id : PACK_ID_BITS;
 270        struct object_id oid;
 271};
 272
 273struct tag {
 274        struct tag *next_tag;
 275        const char *name;
 276        unsigned int pack_id;
 277        struct object_id oid;
 278};
 279
 280struct hash_list {
 281        struct hash_list *next;
 282        struct object_id oid;
 283};
 284
 285typedef enum {
 286        WHENSPEC_RAW = 1,
 287        WHENSPEC_RFC2822,
 288        WHENSPEC_NOW
 289} whenspec_type;
 290
 291struct recent_command {
 292        struct recent_command *prev;
 293        struct recent_command *next;
 294        char *buf;
 295};
 296
 297/* Configured limits on output */
 298static unsigned long max_depth = 50;
 299static off_t max_packsize;
 300static int unpack_limit = 100;
 301static int force_update;
 302
 303/* Stats and misc. counters */
 304static uintmax_t alloc_count;
 305static uintmax_t marks_set_count;
 306static uintmax_t object_count_by_type[1 << TYPE_BITS];
 307static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 308static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 309static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
 310static unsigned long object_count;
 311static unsigned long branch_count;
 312static unsigned long branch_load_count;
 313static int failure;
 314static FILE *pack_edges;
 315static unsigned int show_stats = 1;
 316static int global_argc;
 317static const char **global_argv;
 318
 319/* Memory pools */
 320static struct mem_pool fi_mem_pool =  {0, 2*1024*1024 - sizeof(struct mp_block), 0 };
 321
 322/* Atom management */
 323static unsigned int atom_table_sz = 4451;
 324static unsigned int atom_cnt;
 325static struct atom_str **atom_table;
 326
 327/* The .pack file being generated */
 328static struct pack_idx_option pack_idx_opts;
 329static unsigned int pack_id;
 330static struct hashfile *pack_file;
 331static struct packed_git *pack_data;
 332static struct packed_git **all_packs;
 333static off_t pack_size;
 334
 335/* Table of objects we've written. */
 336static unsigned int object_entry_alloc = 5000;
 337static struct object_entry_pool *blocks;
 338static size_t total_allocd;
 339static struct object_entry *object_table[1 << 16];
 340static struct mark_set *marks;
 341static const char *export_marks_file;
 342static const char *import_marks_file;
 343static int import_marks_file_from_stream;
 344static int import_marks_file_ignore_missing;
 345static int import_marks_file_done;
 346static int relative_marks_paths;
 347
 348/* Our last blob */
 349static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 350
 351/* Tree management */
 352static unsigned int tree_entry_alloc = 1000;
 353static void *avail_tree_entry;
 354static unsigned int avail_tree_table_sz = 100;
 355static struct avail_tree_content **avail_tree_table;
 356static struct strbuf old_tree = STRBUF_INIT;
 357static struct strbuf new_tree = STRBUF_INIT;
 358
 359/* Branch data */
 360static unsigned long max_active_branches = 5;
 361static unsigned long cur_active_branches;
 362static unsigned long branch_table_sz = 1039;
 363static struct branch **branch_table;
 364static struct branch *active_branches;
 365
 366/* Tag data */
 367static struct tag *first_tag;
 368static struct tag *last_tag;
 369
 370/* Input stream parsing */
 371static whenspec_type whenspec = WHENSPEC_RAW;
 372static struct strbuf command_buf = STRBUF_INIT;
 373static int unread_command_buf;
 374static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 375static struct recent_command *cmd_tail = &cmd_hist;
 376static struct recent_command *rc_free;
 377static unsigned int cmd_save = 100;
 378static uintmax_t next_mark;
 379static struct strbuf new_data = STRBUF_INIT;
 380static int seen_data_command;
 381static int require_explicit_termination;
 382
 383/* Signal handling */
 384static volatile sig_atomic_t checkpoint_requested;
 385
 386/* Where to write output of cat-blob commands */
 387static int cat_blob_fd = STDOUT_FILENO;
 388
 389static void parse_argv(void);
 390static void parse_get_mark(const char *p);
 391static void parse_cat_blob(const char *p);
 392static void parse_ls(const char *p, struct branch *b);
 393
 394static void write_branch_report(FILE *rpt, struct branch *b)
 395{
 396        fprintf(rpt, "%s:\n", b->name);
 397
 398        fprintf(rpt, "  status      :");
 399        if (b->active)
 400                fputs(" active", rpt);
 401        if (b->branch_tree.tree)
 402                fputs(" loaded", rpt);
 403        if (is_null_oid(&b->branch_tree.versions[1].oid))
 404                fputs(" dirty", rpt);
 405        fputc('\n', rpt);
 406
 407        fprintf(rpt, "  tip commit  : %s\n", oid_to_hex(&b->oid));
 408        fprintf(rpt, "  old tree    : %s\n",
 409                oid_to_hex(&b->branch_tree.versions[0].oid));
 410        fprintf(rpt, "  cur tree    : %s\n",
 411                oid_to_hex(&b->branch_tree.versions[1].oid));
 412        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 413
 414        fputs("  last pack   : ", rpt);
 415        if (b->pack_id < MAX_PACK_ID)
 416                fprintf(rpt, "%u", b->pack_id);
 417        fputc('\n', rpt);
 418
 419        fputc('\n', rpt);
 420}
 421
 422static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 423
 424static void write_crash_report(const char *err)
 425{
 426        char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 427        FILE *rpt = fopen(loc, "w");
 428        struct branch *b;
 429        unsigned long lu;
 430        struct recent_command *rc;
 431
 432        if (!rpt) {
 433                error_errno("can't write crash report %s", loc);
 434                free(loc);
 435                return;
 436        }
 437
 438        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 439
 440        fprintf(rpt, "fast-import crash report:\n");
 441        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 442        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 443        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
 444        fputc('\n', rpt);
 445
 446        fputs("fatal: ", rpt);
 447        fputs(err, rpt);
 448        fputc('\n', rpt);
 449
 450        fputc('\n', rpt);
 451        fputs("Most Recent Commands Before Crash\n", rpt);
 452        fputs("---------------------------------\n", rpt);
 453        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 454                if (rc->next == &cmd_hist)
 455                        fputs("* ", rpt);
 456                else
 457                        fputs("  ", rpt);
 458                fputs(rc->buf, rpt);
 459                fputc('\n', rpt);
 460        }
 461
 462        fputc('\n', rpt);
 463        fputs("Active Branch LRU\n", rpt);
 464        fputs("-----------------\n", rpt);
 465        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 466                cur_active_branches,
 467                max_active_branches);
 468        fputc('\n', rpt);
 469        fputs("  pos  clock name\n", rpt);
 470        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 471        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 472                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 473                        ++lu, b->last_commit, b->name);
 474
 475        fputc('\n', rpt);
 476        fputs("Inactive Branches\n", rpt);
 477        fputs("-----------------\n", rpt);
 478        for (lu = 0; lu < branch_table_sz; lu++) {
 479                for (b = branch_table[lu]; b; b = b->table_next_branch)
 480                        write_branch_report(rpt, b);
 481        }
 482
 483        if (first_tag) {
 484                struct tag *tg;
 485                fputc('\n', rpt);
 486                fputs("Annotated Tags\n", rpt);
 487                fputs("--------------\n", rpt);
 488                for (tg = first_tag; tg; tg = tg->next_tag) {
 489                        fputs(oid_to_hex(&tg->oid), rpt);
 490                        fputc(' ', rpt);
 491                        fputs(tg->name, rpt);
 492                        fputc('\n', rpt);
 493                }
 494        }
 495
 496        fputc('\n', rpt);
 497        fputs("Marks\n", rpt);
 498        fputs("-----\n", rpt);
 499        if (export_marks_file)
 500                fprintf(rpt, "  exported to %s\n", export_marks_file);
 501        else
 502                dump_marks_helper(rpt, 0, marks);
 503
 504        fputc('\n', rpt);
 505        fputs("-------------------\n", rpt);
 506        fputs("END OF CRASH REPORT\n", rpt);
 507        fclose(rpt);
 508        free(loc);
 509}
 510
 511static void end_packfile(void);
 512static void unkeep_all_packs(void);
 513static void dump_marks(void);
 514
 515static NORETURN void die_nicely(const char *err, va_list params)
 516{
 517        static int zombie;
 518        char message[2 * PATH_MAX];
 519
 520        vsnprintf(message, sizeof(message), err, params);
 521        fputs("fatal: ", stderr);
 522        fputs(message, stderr);
 523        fputc('\n', stderr);
 524
 525        if (!zombie) {
 526                zombie = 1;
 527                write_crash_report(message);
 528                end_packfile();
 529                unkeep_all_packs();
 530                dump_marks();
 531        }
 532        exit(128);
 533}
 534
 535#ifndef SIGUSR1 /* Windows, for example */
 536
 537static void set_checkpoint_signal(void)
 538{
 539}
 540
 541#else
 542
 543static void checkpoint_signal(int signo)
 544{
 545        checkpoint_requested = 1;
 546}
 547
 548static void set_checkpoint_signal(void)
 549{
 550        struct sigaction sa;
 551
 552        memset(&sa, 0, sizeof(sa));
 553        sa.sa_handler = checkpoint_signal;
 554        sigemptyset(&sa.sa_mask);
 555        sa.sa_flags = SA_RESTART;
 556        sigaction(SIGUSR1, &sa, NULL);
 557}
 558
 559#endif
 560
 561static void alloc_objects(unsigned int cnt)
 562{
 563        struct object_entry_pool *b;
 564
 565        b = xmalloc(sizeof(struct object_entry_pool)
 566                + cnt * sizeof(struct object_entry));
 567        b->next_pool = blocks;
 568        b->next_free = b->entries;
 569        b->end = b->entries + cnt;
 570        blocks = b;
 571        alloc_count += cnt;
 572}
 573
 574static struct object_entry *new_object(struct object_id *oid)
 575{
 576        struct object_entry *e;
 577
 578        if (blocks->next_free == blocks->end)
 579                alloc_objects(object_entry_alloc);
 580
 581        e = blocks->next_free++;
 582        oidcpy(&e->idx.oid, oid);
 583        return e;
 584}
 585
 586static struct object_entry *find_object(struct object_id *oid)
 587{
 588        unsigned int h = oid->hash[0] << 8 | oid->hash[1];
 589        struct object_entry *e;
 590        for (e = object_table[h]; e; e = e->next)
 591                if (!oidcmp(oid, &e->idx.oid))
 592                        return e;
 593        return NULL;
 594}
 595
 596static struct object_entry *insert_object(struct object_id *oid)
 597{
 598        unsigned int h = oid->hash[0] << 8 | oid->hash[1];
 599        struct object_entry *e = object_table[h];
 600
 601        while (e) {
 602                if (!oidcmp(oid, &e->idx.oid))
 603                        return e;
 604                e = e->next;
 605        }
 606
 607        e = new_object(oid);
 608        e->next = object_table[h];
 609        e->idx.offset = 0;
 610        object_table[h] = e;
 611        return e;
 612}
 613
 614static void invalidate_pack_id(unsigned int id)
 615{
 616        unsigned int h;
 617        unsigned long lu;
 618        struct tag *t;
 619
 620        for (h = 0; h < ARRAY_SIZE(object_table); h++) {
 621                struct object_entry *e;
 622
 623                for (e = object_table[h]; e; e = e->next)
 624                        if (e->pack_id == id)
 625                                e->pack_id = MAX_PACK_ID;
 626        }
 627
 628        for (lu = 0; lu < branch_table_sz; lu++) {
 629                struct branch *b;
 630
 631                for (b = branch_table[lu]; b; b = b->table_next_branch)
 632                        if (b->pack_id == id)
 633                                b->pack_id = MAX_PACK_ID;
 634        }
 635
 636        for (t = first_tag; t; t = t->next_tag)
 637                if (t->pack_id == id)
 638                        t->pack_id = MAX_PACK_ID;
 639}
 640
 641static unsigned int hc_str(const char *s, size_t len)
 642{
 643        unsigned int r = 0;
 644        while (len-- > 0)
 645                r = r * 31 + *s++;
 646        return r;
 647}
 648
 649static struct mp_block *mem_pool_alloc_block(struct mem_pool *mem_pool, size_t block_alloc)
 650{
 651        struct mp_block *p;
 652
 653        mem_pool->pool_alloc += sizeof(struct mp_block) + block_alloc;
 654        p = xmalloc(st_add(sizeof(struct mp_block), block_alloc));
 655        p->next_block = mem_pool->mp_block;
 656        p->next_free = (char *)p->space;
 657        p->end = p->next_free + block_alloc;
 658        mem_pool->mp_block = p;
 659
 660        return p;
 661}
 662
 663static void *mem_pool_alloc(struct mem_pool *mem_pool, size_t len)
 664{
 665        struct mp_block *p;
 666        void *r;
 667
 668        /* round up to a 'uintmax_t' alignment */
 669        if (len & (sizeof(uintmax_t) - 1))
 670                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 671
 672        for (p = mem_pool->mp_block; p; p = p->next_block)
 673                if (p->end - p->next_free >= len)
 674                        break;
 675
 676        if (!p) {
 677                if (len >= (mem_pool->block_alloc / 2)) {
 678                        mem_pool->pool_alloc += len;
 679                        return xmalloc(len);
 680                }
 681
 682                p = mem_pool_alloc_block(mem_pool, mem_pool->block_alloc);
 683        }
 684
 685        r = p->next_free;
 686        p->next_free += len;
 687        return r;
 688}
 689
 690static void *mem_pool_calloc(struct mem_pool *mem_pool, size_t count, size_t size)
 691{
 692        size_t len = st_mult(count, size);
 693        void *r = mem_pool_alloc(mem_pool, len);
 694        memset(r, 0, len);
 695        return r;
 696}
 697
 698static char *pool_strdup(const char *s)
 699{
 700        size_t len = strlen(s) + 1;
 701        char *r = mem_pool_alloc(&fi_mem_pool, len);
 702        memcpy(r, s, len);
 703        return r;
 704}
 705
 706static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 707{
 708        struct mark_set *s = marks;
 709        while ((idnum >> s->shift) >= 1024) {
 710                s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
 711                s->shift = marks->shift + 10;
 712                s->data.sets[0] = marks;
 713                marks = s;
 714        }
 715        while (s->shift) {
 716                uintmax_t i = idnum >> s->shift;
 717                idnum -= i << s->shift;
 718                if (!s->data.sets[i]) {
 719                        s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
 720                        s->data.sets[i]->shift = s->shift - 10;
 721                }
 722                s = s->data.sets[i];
 723        }
 724        if (!s->data.marked[idnum])
 725                marks_set_count++;
 726        s->data.marked[idnum] = oe;
 727}
 728
 729static struct object_entry *find_mark(uintmax_t idnum)
 730{
 731        uintmax_t orig_idnum = idnum;
 732        struct mark_set *s = marks;
 733        struct object_entry *oe = NULL;
 734        if ((idnum >> s->shift) < 1024) {
 735                while (s && s->shift) {
 736                        uintmax_t i = idnum >> s->shift;
 737                        idnum -= i << s->shift;
 738                        s = s->data.sets[i];
 739                }
 740                if (s)
 741                        oe = s->data.marked[idnum];
 742        }
 743        if (!oe)
 744                die("mark :%" PRIuMAX " not declared", orig_idnum);
 745        return oe;
 746}
 747
 748static struct atom_str *to_atom(const char *s, unsigned short len)
 749{
 750        unsigned int hc = hc_str(s, len) % atom_table_sz;
 751        struct atom_str *c;
 752
 753        for (c = atom_table[hc]; c; c = c->next_atom)
 754                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 755                        return c;
 756
 757        c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
 758        c->str_len = len;
 759        memcpy(c->str_dat, s, len);
 760        c->str_dat[len] = 0;
 761        c->next_atom = atom_table[hc];
 762        atom_table[hc] = c;
 763        atom_cnt++;
 764        return c;
 765}
 766
 767static struct branch *lookup_branch(const char *name)
 768{
 769        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 770        struct branch *b;
 771
 772        for (b = branch_table[hc]; b; b = b->table_next_branch)
 773                if (!strcmp(name, b->name))
 774                        return b;
 775        return NULL;
 776}
 777
 778static struct branch *new_branch(const char *name)
 779{
 780        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 781        struct branch *b = lookup_branch(name);
 782
 783        if (b)
 784                die("Invalid attempt to create duplicate branch: %s", name);
 785        if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
 786                die("Branch name doesn't conform to GIT standards: %s", name);
 787
 788        b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
 789        b->name = pool_strdup(name);
 790        b->table_next_branch = branch_table[hc];
 791        b->branch_tree.versions[0].mode = S_IFDIR;
 792        b->branch_tree.versions[1].mode = S_IFDIR;
 793        b->num_notes = 0;
 794        b->active = 0;
 795        b->pack_id = MAX_PACK_ID;
 796        branch_table[hc] = b;
 797        branch_count++;
 798        return b;
 799}
 800
 801static unsigned int hc_entries(unsigned int cnt)
 802{
 803        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 804        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 805}
 806
 807static struct tree_content *new_tree_content(unsigned int cnt)
 808{
 809        struct avail_tree_content *f, *l = NULL;
 810        struct tree_content *t;
 811        unsigned int hc = hc_entries(cnt);
 812
 813        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 814                if (f->entry_capacity >= cnt)
 815                        break;
 816
 817        if (f) {
 818                if (l)
 819                        l->next_avail = f->next_avail;
 820                else
 821                        avail_tree_table[hc] = f->next_avail;
 822        } else {
 823                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 824                f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
 825                f->entry_capacity = cnt;
 826        }
 827
 828        t = (struct tree_content*)f;
 829        t->entry_count = 0;
 830        t->delta_depth = 0;
 831        return t;
 832}
 833
 834static void release_tree_entry(struct tree_entry *e);
 835static void release_tree_content(struct tree_content *t)
 836{
 837        struct avail_tree_content *f = (struct avail_tree_content*)t;
 838        unsigned int hc = hc_entries(f->entry_capacity);
 839        f->next_avail = avail_tree_table[hc];
 840        avail_tree_table[hc] = f;
 841}
 842
 843static void release_tree_content_recursive(struct tree_content *t)
 844{
 845        unsigned int i;
 846        for (i = 0; i < t->entry_count; i++)
 847                release_tree_entry(t->entries[i]);
 848        release_tree_content(t);
 849}
 850
 851static struct tree_content *grow_tree_content(
 852        struct tree_content *t,
 853        int amt)
 854{
 855        struct tree_content *r = new_tree_content(t->entry_count + amt);
 856        r->entry_count = t->entry_count;
 857        r->delta_depth = t->delta_depth;
 858        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 859        release_tree_content(t);
 860        return r;
 861}
 862
 863static struct tree_entry *new_tree_entry(void)
 864{
 865        struct tree_entry *e;
 866
 867        if (!avail_tree_entry) {
 868                unsigned int n = tree_entry_alloc;
 869                total_allocd += n * sizeof(struct tree_entry);
 870                ALLOC_ARRAY(e, n);
 871                avail_tree_entry = e;
 872                while (n-- > 1) {
 873                        *((void**)e) = e + 1;
 874                        e++;
 875                }
 876                *((void**)e) = NULL;
 877        }
 878
 879        e = avail_tree_entry;
 880        avail_tree_entry = *((void**)e);
 881        return e;
 882}
 883
 884static void release_tree_entry(struct tree_entry *e)
 885{
 886        if (e->tree)
 887                release_tree_content_recursive(e->tree);
 888        *((void**)e) = avail_tree_entry;
 889        avail_tree_entry = e;
 890}
 891
 892static struct tree_content *dup_tree_content(struct tree_content *s)
 893{
 894        struct tree_content *d;
 895        struct tree_entry *a, *b;
 896        unsigned int i;
 897
 898        if (!s)
 899                return NULL;
 900        d = new_tree_content(s->entry_count);
 901        for (i = 0; i < s->entry_count; i++) {
 902                a = s->entries[i];
 903                b = new_tree_entry();
 904                memcpy(b, a, sizeof(*a));
 905                if (a->tree && is_null_oid(&b->versions[1].oid))
 906                        b->tree = dup_tree_content(a->tree);
 907                else
 908                        b->tree = NULL;
 909                d->entries[i] = b;
 910        }
 911        d->entry_count = s->entry_count;
 912        d->delta_depth = s->delta_depth;
 913
 914        return d;
 915}
 916
 917static void start_packfile(void)
 918{
 919        struct strbuf tmp_file = STRBUF_INIT;
 920        struct packed_git *p;
 921        struct pack_header hdr;
 922        int pack_fd;
 923
 924        pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
 925        FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
 926        strbuf_release(&tmp_file);
 927
 928        p->pack_fd = pack_fd;
 929        p->do_not_close = 1;
 930        pack_file = hashfd(pack_fd, p->pack_name);
 931
 932        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 933        hdr.hdr_version = htonl(2);
 934        hdr.hdr_entries = 0;
 935        hashwrite(pack_file, &hdr, sizeof(hdr));
 936
 937        pack_data = p;
 938        pack_size = sizeof(hdr);
 939        object_count = 0;
 940
 941        REALLOC_ARRAY(all_packs, pack_id + 1);
 942        all_packs[pack_id] = p;
 943}
 944
 945static const char *create_index(void)
 946{
 947        const char *tmpfile;
 948        struct pack_idx_entry **idx, **c, **last;
 949        struct object_entry *e;
 950        struct object_entry_pool *o;
 951
 952        /* Build the table of object IDs. */
 953        ALLOC_ARRAY(idx, object_count);
 954        c = idx;
 955        for (o = blocks; o; o = o->next_pool)
 956                for (e = o->next_free; e-- != o->entries;)
 957                        if (pack_id == e->pack_id)
 958                                *c++ = &e->idx;
 959        last = idx + object_count;
 960        if (c != last)
 961                die("internal consistency error creating the index");
 962
 963        tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
 964        free(idx);
 965        return tmpfile;
 966}
 967
 968static char *keep_pack(const char *curr_index_name)
 969{
 970        static const char *keep_msg = "fast-import";
 971        struct strbuf name = STRBUF_INIT;
 972        int keep_fd;
 973
 974        odb_pack_name(&name, pack_data->sha1, "keep");
 975        keep_fd = odb_pack_keep(name.buf);
 976        if (keep_fd < 0)
 977                die_errno("cannot create keep file");
 978        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 979        if (close(keep_fd))
 980                die_errno("failed to write keep file");
 981
 982        odb_pack_name(&name, pack_data->sha1, "pack");
 983        if (finalize_object_file(pack_data->pack_name, name.buf))
 984                die("cannot store pack file");
 985
 986        odb_pack_name(&name, pack_data->sha1, "idx");
 987        if (finalize_object_file(curr_index_name, name.buf))
 988                die("cannot store index file");
 989        free((void *)curr_index_name);
 990        return strbuf_detach(&name, NULL);
 991}
 992
 993static void unkeep_all_packs(void)
 994{
 995        struct strbuf name = STRBUF_INIT;
 996        int k;
 997
 998        for (k = 0; k < pack_id; k++) {
 999                struct packed_git *p = all_packs[k];
1000                odb_pack_name(&name, p->sha1, "keep");
1001                unlink_or_warn(name.buf);
1002        }
1003        strbuf_release(&name);
1004}
1005
1006static int loosen_small_pack(const struct packed_git *p)
1007{
1008        struct child_process unpack = CHILD_PROCESS_INIT;
1009
1010        if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
1011                die_errno("Failed seeking to start of '%s'", p->pack_name);
1012
1013        unpack.in = p->pack_fd;
1014        unpack.git_cmd = 1;
1015        unpack.stdout_to_stderr = 1;
1016        argv_array_push(&unpack.args, "unpack-objects");
1017        if (!show_stats)
1018                argv_array_push(&unpack.args, "-q");
1019
1020        return run_command(&unpack);
1021}
1022
1023static void end_packfile(void)
1024{
1025        static int running;
1026
1027        if (running || !pack_data)
1028                return;
1029
1030        running = 1;
1031        clear_delta_base_cache();
1032        if (object_count) {
1033                struct packed_git *new_p;
1034                struct object_id cur_pack_oid;
1035                char *idx_name;
1036                int i;
1037                struct branch *b;
1038                struct tag *t;
1039
1040                close_pack_windows(pack_data);
1041                hashclose(pack_file, cur_pack_oid.hash, 0);
1042                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
1043                                    pack_data->pack_name, object_count,
1044                                    cur_pack_oid.hash, pack_size);
1045
1046                if (object_count <= unpack_limit) {
1047                        if (!loosen_small_pack(pack_data)) {
1048                                invalidate_pack_id(pack_id);
1049                                goto discard_pack;
1050                        }
1051                }
1052
1053                close(pack_data->pack_fd);
1054                idx_name = keep_pack(create_index());
1055
1056                /* Register the packfile with core git's machinery. */
1057                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
1058                if (!new_p)
1059                        die("core git rejected index %s", idx_name);
1060                all_packs[pack_id] = new_p;
1061                install_packed_git(new_p);
1062                free(idx_name);
1063
1064                /* Print the boundary */
1065                if (pack_edges) {
1066                        fprintf(pack_edges, "%s:", new_p->pack_name);
1067                        for (i = 0; i < branch_table_sz; i++) {
1068                                for (b = branch_table[i]; b; b = b->table_next_branch) {
1069                                        if (b->pack_id == pack_id)
1070                                                fprintf(pack_edges, " %s",
1071                                                        oid_to_hex(&b->oid));
1072                                }
1073                        }
1074                        for (t = first_tag; t; t = t->next_tag) {
1075                                if (t->pack_id == pack_id)
1076                                        fprintf(pack_edges, " %s",
1077                                                oid_to_hex(&t->oid));
1078                        }
1079                        fputc('\n', pack_edges);
1080                        fflush(pack_edges);
1081                }
1082
1083                pack_id++;
1084        }
1085        else {
1086discard_pack:
1087                close(pack_data->pack_fd);
1088                unlink_or_warn(pack_data->pack_name);
1089        }
1090        FREE_AND_NULL(pack_data);
1091        running = 0;
1092
1093        /* We can't carry a delta across packfiles. */
1094        strbuf_release(&last_blob.data);
1095        last_blob.offset = 0;
1096        last_blob.depth = 0;
1097}
1098
1099static void cycle_packfile(void)
1100{
1101        end_packfile();
1102        start_packfile();
1103}
1104
1105static int store_object(
1106        enum object_type type,
1107        struct strbuf *dat,
1108        struct last_object *last,
1109        struct object_id *oidout,
1110        uintmax_t mark)
1111{
1112        void *out, *delta;
1113        struct object_entry *e;
1114        unsigned char hdr[96];
1115        struct object_id oid;
1116        unsigned long hdrlen, deltalen;
1117        git_hash_ctx c;
1118        git_zstream s;
1119
1120        hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1121                           type_name(type), (unsigned long)dat->len) + 1;
1122        the_hash_algo->init_fn(&c);
1123        the_hash_algo->update_fn(&c, hdr, hdrlen);
1124        the_hash_algo->update_fn(&c, dat->buf, dat->len);
1125        the_hash_algo->final_fn(oid.hash, &c);
1126        if (oidout)
1127                oidcpy(oidout, &oid);
1128
1129        e = insert_object(&oid);
1130        if (mark)
1131                insert_mark(mark, e);
1132        if (e->idx.offset) {
1133                duplicate_count_by_type[type]++;
1134                return 1;
1135        } else if (find_sha1_pack(oid.hash, packed_git)) {
1136                e->type = type;
1137                e->pack_id = MAX_PACK_ID;
1138                e->idx.offset = 1; /* just not zero! */
1139                duplicate_count_by_type[type]++;
1140                return 1;
1141        }
1142
1143        if (last && last->data.buf && last->depth < max_depth
1144                && dat->len > the_hash_algo->rawsz) {
1145
1146                delta_count_attempts_by_type[type]++;
1147                delta = diff_delta(last->data.buf, last->data.len,
1148                        dat->buf, dat->len,
1149                        &deltalen, dat->len - the_hash_algo->rawsz);
1150        } else
1151                delta = NULL;
1152
1153        git_deflate_init(&s, pack_compression_level);
1154        if (delta) {
1155                s.next_in = delta;
1156                s.avail_in = deltalen;
1157        } else {
1158                s.next_in = (void *)dat->buf;
1159                s.avail_in = dat->len;
1160        }
1161        s.avail_out = git_deflate_bound(&s, s.avail_in);
1162        s.next_out = out = xmalloc(s.avail_out);
1163        while (git_deflate(&s, Z_FINISH) == Z_OK)
1164                ; /* nothing */
1165        git_deflate_end(&s);
1166
1167        /* Determine if we should auto-checkpoint. */
1168        if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1169                || (pack_size + 60 + s.total_out) < pack_size) {
1170
1171                /* This new object needs to *not* have the current pack_id. */
1172                e->pack_id = pack_id + 1;
1173                cycle_packfile();
1174
1175                /* We cannot carry a delta into the new pack. */
1176                if (delta) {
1177                        FREE_AND_NULL(delta);
1178
1179                        git_deflate_init(&s, pack_compression_level);
1180                        s.next_in = (void *)dat->buf;
1181                        s.avail_in = dat->len;
1182                        s.avail_out = git_deflate_bound(&s, s.avail_in);
1183                        s.next_out = out = xrealloc(out, s.avail_out);
1184                        while (git_deflate(&s, Z_FINISH) == Z_OK)
1185                                ; /* nothing */
1186                        git_deflate_end(&s);
1187                }
1188        }
1189
1190        e->type = type;
1191        e->pack_id = pack_id;
1192        e->idx.offset = pack_size;
1193        object_count++;
1194        object_count_by_type[type]++;
1195
1196        crc32_begin(pack_file);
1197
1198        if (delta) {
1199                off_t ofs = e->idx.offset - last->offset;
1200                unsigned pos = sizeof(hdr) - 1;
1201
1202                delta_count_by_type[type]++;
1203                e->depth = last->depth + 1;
1204
1205                hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1206                                                      OBJ_OFS_DELTA, deltalen);
1207                hashwrite(pack_file, hdr, hdrlen);
1208                pack_size += hdrlen;
1209
1210                hdr[pos] = ofs & 127;
1211                while (ofs >>= 7)
1212                        hdr[--pos] = 128 | (--ofs & 127);
1213                hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1214                pack_size += sizeof(hdr) - pos;
1215        } else {
1216                e->depth = 0;
1217                hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1218                                                      type, dat->len);
1219                hashwrite(pack_file, hdr, hdrlen);
1220                pack_size += hdrlen;
1221        }
1222
1223        hashwrite(pack_file, out, s.total_out);
1224        pack_size += s.total_out;
1225
1226        e->idx.crc32 = crc32_end(pack_file);
1227
1228        free(out);
1229        free(delta);
1230        if (last) {
1231                if (last->no_swap) {
1232                        last->data = *dat;
1233                } else {
1234                        strbuf_swap(&last->data, dat);
1235                }
1236                last->offset = e->idx.offset;
1237                last->depth = e->depth;
1238        }
1239        return 0;
1240}
1241
1242static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1243{
1244        if (hashfile_truncate(pack_file, checkpoint))
1245                die_errno("cannot truncate pack to skip duplicate");
1246        pack_size = checkpoint->offset;
1247}
1248
1249static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1250{
1251        size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1252        unsigned char *in_buf = xmalloc(in_sz);
1253        unsigned char *out_buf = xmalloc(out_sz);
1254        struct object_entry *e;
1255        struct object_id oid;
1256        unsigned long hdrlen;
1257        off_t offset;
1258        git_hash_ctx c;
1259        git_zstream s;
1260        struct hashfile_checkpoint checkpoint;
1261        int status = Z_OK;
1262
1263        /* Determine if we should auto-checkpoint. */
1264        if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1265                || (pack_size + 60 + len) < pack_size)
1266                cycle_packfile();
1267
1268        hashfile_checkpoint(pack_file, &checkpoint);
1269        offset = checkpoint.offset;
1270
1271        hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1272
1273        the_hash_algo->init_fn(&c);
1274        the_hash_algo->update_fn(&c, out_buf, hdrlen);
1275
1276        crc32_begin(pack_file);
1277
1278        git_deflate_init(&s, pack_compression_level);
1279
1280        hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1281
1282        s.next_out = out_buf + hdrlen;
1283        s.avail_out = out_sz - hdrlen;
1284
1285        while (status != Z_STREAM_END) {
1286                if (0 < len && !s.avail_in) {
1287                        size_t cnt = in_sz < len ? in_sz : (size_t)len;
1288                        size_t n = fread(in_buf, 1, cnt, stdin);
1289                        if (!n && feof(stdin))
1290                                die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1291
1292                        the_hash_algo->update_fn(&c, in_buf, n);
1293                        s.next_in = in_buf;
1294                        s.avail_in = n;
1295                        len -= n;
1296                }
1297
1298                status = git_deflate(&s, len ? 0 : Z_FINISH);
1299
1300                if (!s.avail_out || status == Z_STREAM_END) {
1301                        size_t n = s.next_out - out_buf;
1302                        hashwrite(pack_file, out_buf, n);
1303                        pack_size += n;
1304                        s.next_out = out_buf;
1305                        s.avail_out = out_sz;
1306                }
1307
1308                switch (status) {
1309                case Z_OK:
1310                case Z_BUF_ERROR:
1311                case Z_STREAM_END:
1312                        continue;
1313                default:
1314                        die("unexpected deflate failure: %d", status);
1315                }
1316        }
1317        git_deflate_end(&s);
1318        the_hash_algo->final_fn(oid.hash, &c);
1319
1320        if (oidout)
1321                oidcpy(oidout, &oid);
1322
1323        e = insert_object(&oid);
1324
1325        if (mark)
1326                insert_mark(mark, e);
1327
1328        if (e->idx.offset) {
1329                duplicate_count_by_type[OBJ_BLOB]++;
1330                truncate_pack(&checkpoint);
1331
1332        } else if (find_sha1_pack(oid.hash, packed_git)) {
1333                e->type = OBJ_BLOB;
1334                e->pack_id = MAX_PACK_ID;
1335                e->idx.offset = 1; /* just not zero! */
1336                duplicate_count_by_type[OBJ_BLOB]++;
1337                truncate_pack(&checkpoint);
1338
1339        } else {
1340                e->depth = 0;
1341                e->type = OBJ_BLOB;
1342                e->pack_id = pack_id;
1343                e->idx.offset = offset;
1344                e->idx.crc32 = crc32_end(pack_file);
1345                object_count++;
1346                object_count_by_type[OBJ_BLOB]++;
1347        }
1348
1349        free(in_buf);
1350        free(out_buf);
1351}
1352
1353/* All calls must be guarded by find_object() or find_mark() to
1354 * ensure the 'struct object_entry' passed was written by this
1355 * process instance.  We unpack the entry by the offset, avoiding
1356 * the need for the corresponding .idx file.  This unpacking rule
1357 * works because we only use OBJ_REF_DELTA within the packfiles
1358 * created by fast-import.
1359 *
1360 * oe must not be NULL.  Such an oe usually comes from giving
1361 * an unknown SHA-1 to find_object() or an undefined mark to
1362 * find_mark().  Callers must test for this condition and use
1363 * the standard read_sha1_file() when it happens.
1364 *
1365 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1366 * find_mark(), where the mark was reloaded from an existing marks
1367 * file and is referencing an object that this fast-import process
1368 * instance did not write out to a packfile.  Callers must test for
1369 * this condition and use read_sha1_file() instead.
1370 */
1371static void *gfi_unpack_entry(
1372        struct object_entry *oe,
1373        unsigned long *sizep)
1374{
1375        enum object_type type;
1376        struct packed_git *p = all_packs[oe->pack_id];
1377        if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1378                /* The object is stored in the packfile we are writing to
1379                 * and we have modified it since the last time we scanned
1380                 * back to read a previously written object.  If an old
1381                 * window covered [p->pack_size, p->pack_size + rawsz) its
1382                 * data is stale and is not valid.  Closing all windows
1383                 * and updating the packfile length ensures we can read
1384                 * the newly written data.
1385                 */
1386                close_pack_windows(p);
1387                hashflush(pack_file);
1388
1389                /* We have to offer rawsz bytes additional on the end of
1390                 * the packfile as the core unpacker code assumes the
1391                 * footer is present at the file end and must promise
1392                 * at least rawsz bytes within any window it maps.  But
1393                 * we don't actually create the footer here.
1394                 */
1395                p->pack_size = pack_size + the_hash_algo->rawsz;
1396        }
1397        return unpack_entry(p, oe->idx.offset, &type, sizep);
1398}
1399
1400static const char *get_mode(const char *str, uint16_t *modep)
1401{
1402        unsigned char c;
1403        uint16_t mode = 0;
1404
1405        while ((c = *str++) != ' ') {
1406                if (c < '0' || c > '7')
1407                        return NULL;
1408                mode = (mode << 3) + (c - '0');
1409        }
1410        *modep = mode;
1411        return str;
1412}
1413
1414static void load_tree(struct tree_entry *root)
1415{
1416        struct object_id *oid = &root->versions[1].oid;
1417        struct object_entry *myoe;
1418        struct tree_content *t;
1419        unsigned long size;
1420        char *buf;
1421        const char *c;
1422
1423        root->tree = t = new_tree_content(8);
1424        if (is_null_oid(oid))
1425                return;
1426
1427        myoe = find_object(oid);
1428        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1429                if (myoe->type != OBJ_TREE)
1430                        die("Not a tree: %s", oid_to_hex(oid));
1431                t->delta_depth = myoe->depth;
1432                buf = gfi_unpack_entry(myoe, &size);
1433                if (!buf)
1434                        die("Can't load tree %s", oid_to_hex(oid));
1435        } else {
1436                enum object_type type;
1437                buf = read_sha1_file(oid->hash, &type, &size);
1438                if (!buf || type != OBJ_TREE)
1439                        die("Can't load tree %s", oid_to_hex(oid));
1440        }
1441
1442        c = buf;
1443        while (c != (buf + size)) {
1444                struct tree_entry *e = new_tree_entry();
1445
1446                if (t->entry_count == t->entry_capacity)
1447                        root->tree = t = grow_tree_content(t, t->entry_count);
1448                t->entries[t->entry_count++] = e;
1449
1450                e->tree = NULL;
1451                c = get_mode(c, &e->versions[1].mode);
1452                if (!c)
1453                        die("Corrupt mode in %s", oid_to_hex(oid));
1454                e->versions[0].mode = e->versions[1].mode;
1455                e->name = to_atom(c, strlen(c));
1456                c += e->name->str_len + 1;
1457                hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1458                hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1459                c += GIT_SHA1_RAWSZ;
1460        }
1461        free(buf);
1462}
1463
1464static int tecmp0 (const void *_a, const void *_b)
1465{
1466        struct tree_entry *a = *((struct tree_entry**)_a);
1467        struct tree_entry *b = *((struct tree_entry**)_b);
1468        return base_name_compare(
1469                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1470                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1471}
1472
1473static int tecmp1 (const void *_a, const void *_b)
1474{
1475        struct tree_entry *a = *((struct tree_entry**)_a);
1476        struct tree_entry *b = *((struct tree_entry**)_b);
1477        return base_name_compare(
1478                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1479                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1480}
1481
1482static void mktree(struct tree_content *t, int v, struct strbuf *b)
1483{
1484        size_t maxlen = 0;
1485        unsigned int i;
1486
1487        if (!v)
1488                QSORT(t->entries, t->entry_count, tecmp0);
1489        else
1490                QSORT(t->entries, t->entry_count, tecmp1);
1491
1492        for (i = 0; i < t->entry_count; i++) {
1493                if (t->entries[i]->versions[v].mode)
1494                        maxlen += t->entries[i]->name->str_len + 34;
1495        }
1496
1497        strbuf_reset(b);
1498        strbuf_grow(b, maxlen);
1499        for (i = 0; i < t->entry_count; i++) {
1500                struct tree_entry *e = t->entries[i];
1501                if (!e->versions[v].mode)
1502                        continue;
1503                strbuf_addf(b, "%o %s%c",
1504                        (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1505                        e->name->str_dat, '\0');
1506                strbuf_add(b, e->versions[v].oid.hash, GIT_SHA1_RAWSZ);
1507        }
1508}
1509
1510static void store_tree(struct tree_entry *root)
1511{
1512        struct tree_content *t;
1513        unsigned int i, j, del;
1514        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1515        struct object_entry *le = NULL;
1516
1517        if (!is_null_oid(&root->versions[1].oid))
1518                return;
1519
1520        if (!root->tree)
1521                load_tree(root);
1522        t = root->tree;
1523
1524        for (i = 0; i < t->entry_count; i++) {
1525                if (t->entries[i]->tree)
1526                        store_tree(t->entries[i]);
1527        }
1528
1529        if (!(root->versions[0].mode & NO_DELTA))
1530                le = find_object(&root->versions[0].oid);
1531        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1532                mktree(t, 0, &old_tree);
1533                lo.data = old_tree;
1534                lo.offset = le->idx.offset;
1535                lo.depth = t->delta_depth;
1536        }
1537
1538        mktree(t, 1, &new_tree);
1539        store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1540
1541        t->delta_depth = lo.depth;
1542        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1543                struct tree_entry *e = t->entries[i];
1544                if (e->versions[1].mode) {
1545                        e->versions[0].mode = e->versions[1].mode;
1546                        oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1547                        t->entries[j++] = e;
1548                } else {
1549                        release_tree_entry(e);
1550                        del++;
1551                }
1552        }
1553        t->entry_count -= del;
1554}
1555
1556static void tree_content_replace(
1557        struct tree_entry *root,
1558        const struct object_id *oid,
1559        const uint16_t mode,
1560        struct tree_content *newtree)
1561{
1562        if (!S_ISDIR(mode))
1563                die("Root cannot be a non-directory");
1564        oidclr(&root->versions[0].oid);
1565        oidcpy(&root->versions[1].oid, oid);
1566        if (root->tree)
1567                release_tree_content_recursive(root->tree);
1568        root->tree = newtree;
1569}
1570
1571static int tree_content_set(
1572        struct tree_entry *root,
1573        const char *p,
1574        const struct object_id *oid,
1575        const uint16_t mode,
1576        struct tree_content *subtree)
1577{
1578        struct tree_content *t;
1579        const char *slash1;
1580        unsigned int i, n;
1581        struct tree_entry *e;
1582
1583        slash1 = strchrnul(p, '/');
1584        n = slash1 - p;
1585        if (!n)
1586                die("Empty path component found in input");
1587        if (!*slash1 && !S_ISDIR(mode) && subtree)
1588                die("Non-directories cannot have subtrees");
1589
1590        if (!root->tree)
1591                load_tree(root);
1592        t = root->tree;
1593        for (i = 0; i < t->entry_count; i++) {
1594                e = t->entries[i];
1595                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1596                        if (!*slash1) {
1597                                if (!S_ISDIR(mode)
1598                                                && e->versions[1].mode == mode
1599                                                && !oidcmp(&e->versions[1].oid, oid))
1600                                        return 0;
1601                                e->versions[1].mode = mode;
1602                                oidcpy(&e->versions[1].oid, oid);
1603                                if (e->tree)
1604                                        release_tree_content_recursive(e->tree);
1605                                e->tree = subtree;
1606
1607                                /*
1608                                 * We need to leave e->versions[0].sha1 alone
1609                                 * to avoid modifying the preimage tree used
1610                                 * when writing out the parent directory.
1611                                 * But after replacing the subdir with a
1612                                 * completely different one, it's not a good
1613                                 * delta base any more, and besides, we've
1614                                 * thrown away the tree entries needed to
1615                                 * make a delta against it.
1616                                 *
1617                                 * So let's just explicitly disable deltas
1618                                 * for the subtree.
1619                                 */
1620                                if (S_ISDIR(e->versions[0].mode))
1621                                        e->versions[0].mode |= NO_DELTA;
1622
1623                                oidclr(&root->versions[1].oid);
1624                                return 1;
1625                        }
1626                        if (!S_ISDIR(e->versions[1].mode)) {
1627                                e->tree = new_tree_content(8);
1628                                e->versions[1].mode = S_IFDIR;
1629                        }
1630                        if (!e->tree)
1631                                load_tree(e);
1632                        if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1633                                oidclr(&root->versions[1].oid);
1634                                return 1;
1635                        }
1636                        return 0;
1637                }
1638        }
1639
1640        if (t->entry_count == t->entry_capacity)
1641                root->tree = t = grow_tree_content(t, t->entry_count);
1642        e = new_tree_entry();
1643        e->name = to_atom(p, n);
1644        e->versions[0].mode = 0;
1645        oidclr(&e->versions[0].oid);
1646        t->entries[t->entry_count++] = e;
1647        if (*slash1) {
1648                e->tree = new_tree_content(8);
1649                e->versions[1].mode = S_IFDIR;
1650                tree_content_set(e, slash1 + 1, oid, mode, subtree);
1651        } else {
1652                e->tree = subtree;
1653                e->versions[1].mode = mode;
1654                oidcpy(&e->versions[1].oid, oid);
1655        }
1656        oidclr(&root->versions[1].oid);
1657        return 1;
1658}
1659
1660static int tree_content_remove(
1661        struct tree_entry *root,
1662        const char *p,
1663        struct tree_entry *backup_leaf,
1664        int allow_root)
1665{
1666        struct tree_content *t;
1667        const char *slash1;
1668        unsigned int i, n;
1669        struct tree_entry *e;
1670
1671        slash1 = strchrnul(p, '/');
1672        n = slash1 - p;
1673
1674        if (!root->tree)
1675                load_tree(root);
1676
1677        if (!*p && allow_root) {
1678                e = root;
1679                goto del_entry;
1680        }
1681
1682        t = root->tree;
1683        for (i = 0; i < t->entry_count; i++) {
1684                e = t->entries[i];
1685                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1686                        if (*slash1 && !S_ISDIR(e->versions[1].mode))
1687                                /*
1688                                 * If p names a file in some subdirectory, and a
1689                                 * file or symlink matching the name of the
1690                                 * parent directory of p exists, then p cannot
1691                                 * exist and need not be deleted.
1692                                 */
1693                                return 1;
1694                        if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1695                                goto del_entry;
1696                        if (!e->tree)
1697                                load_tree(e);
1698                        if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1699                                for (n = 0; n < e->tree->entry_count; n++) {
1700                                        if (e->tree->entries[n]->versions[1].mode) {
1701                                                oidclr(&root->versions[1].oid);
1702                                                return 1;
1703                                        }
1704                                }
1705                                backup_leaf = NULL;
1706                                goto del_entry;
1707                        }
1708                        return 0;
1709                }
1710        }
1711        return 0;
1712
1713del_entry:
1714        if (backup_leaf)
1715                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1716        else if (e->tree)
1717                release_tree_content_recursive(e->tree);
1718        e->tree = NULL;
1719        e->versions[1].mode = 0;
1720        oidclr(&e->versions[1].oid);
1721        oidclr(&root->versions[1].oid);
1722        return 1;
1723}
1724
1725static int tree_content_get(
1726        struct tree_entry *root,
1727        const char *p,
1728        struct tree_entry *leaf,
1729        int allow_root)
1730{
1731        struct tree_content *t;
1732        const char *slash1;
1733        unsigned int i, n;
1734        struct tree_entry *e;
1735
1736        slash1 = strchrnul(p, '/');
1737        n = slash1 - p;
1738        if (!n && !allow_root)
1739                die("Empty path component found in input");
1740
1741        if (!root->tree)
1742                load_tree(root);
1743
1744        if (!n) {
1745                e = root;
1746                goto found_entry;
1747        }
1748
1749        t = root->tree;
1750        for (i = 0; i < t->entry_count; i++) {
1751                e = t->entries[i];
1752                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1753                        if (!*slash1)
1754                                goto found_entry;
1755                        if (!S_ISDIR(e->versions[1].mode))
1756                                return 0;
1757                        if (!e->tree)
1758                                load_tree(e);
1759                        return tree_content_get(e, slash1 + 1, leaf, 0);
1760                }
1761        }
1762        return 0;
1763
1764found_entry:
1765        memcpy(leaf, e, sizeof(*leaf));
1766        if (e->tree && is_null_oid(&e->versions[1].oid))
1767                leaf->tree = dup_tree_content(e->tree);
1768        else
1769                leaf->tree = NULL;
1770        return 1;
1771}
1772
1773static int update_branch(struct branch *b)
1774{
1775        static const char *msg = "fast-import";
1776        struct ref_transaction *transaction;
1777        struct object_id old_oid;
1778        struct strbuf err = STRBUF_INIT;
1779
1780        if (is_null_oid(&b->oid)) {
1781                if (b->delete)
1782                        delete_ref(NULL, b->name, NULL, 0);
1783                return 0;
1784        }
1785        if (read_ref(b->name, &old_oid))
1786                oidclr(&old_oid);
1787        if (!force_update && !is_null_oid(&old_oid)) {
1788                struct commit *old_cmit, *new_cmit;
1789
1790                old_cmit = lookup_commit_reference_gently(&old_oid, 0);
1791                new_cmit = lookup_commit_reference_gently(&b->oid, 0);
1792                if (!old_cmit || !new_cmit)
1793                        return error("Branch %s is missing commits.", b->name);
1794
1795                if (!in_merge_bases(old_cmit, new_cmit)) {
1796                        warning("Not updating %s"
1797                                " (new tip %s does not contain %s)",
1798                                b->name, oid_to_hex(&b->oid),
1799                                oid_to_hex(&old_oid));
1800                        return -1;
1801                }
1802        }
1803        transaction = ref_transaction_begin(&err);
1804        if (!transaction ||
1805            ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1806                                   0, msg, &err) ||
1807            ref_transaction_commit(transaction, &err)) {
1808                ref_transaction_free(transaction);
1809                error("%s", err.buf);
1810                strbuf_release(&err);
1811                return -1;
1812        }
1813        ref_transaction_free(transaction);
1814        strbuf_release(&err);
1815        return 0;
1816}
1817
1818static void dump_branches(void)
1819{
1820        unsigned int i;
1821        struct branch *b;
1822
1823        for (i = 0; i < branch_table_sz; i++) {
1824                for (b = branch_table[i]; b; b = b->table_next_branch)
1825                        failure |= update_branch(b);
1826        }
1827}
1828
1829static void dump_tags(void)
1830{
1831        static const char *msg = "fast-import";
1832        struct tag *t;
1833        struct strbuf ref_name = STRBUF_INIT;
1834        struct strbuf err = STRBUF_INIT;
1835        struct ref_transaction *transaction;
1836
1837        transaction = ref_transaction_begin(&err);
1838        if (!transaction) {
1839                failure |= error("%s", err.buf);
1840                goto cleanup;
1841        }
1842        for (t = first_tag; t; t = t->next_tag) {
1843                strbuf_reset(&ref_name);
1844                strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1845
1846                if (ref_transaction_update(transaction, ref_name.buf,
1847                                           &t->oid, NULL, 0, msg, &err)) {
1848                        failure |= error("%s", err.buf);
1849                        goto cleanup;
1850                }
1851        }
1852        if (ref_transaction_commit(transaction, &err))
1853                failure |= error("%s", err.buf);
1854
1855 cleanup:
1856        ref_transaction_free(transaction);
1857        strbuf_release(&ref_name);
1858        strbuf_release(&err);
1859}
1860
1861static void dump_marks_helper(FILE *f,
1862        uintmax_t base,
1863        struct mark_set *m)
1864{
1865        uintmax_t k;
1866        if (m->shift) {
1867                for (k = 0; k < 1024; k++) {
1868                        if (m->data.sets[k])
1869                                dump_marks_helper(f, base + (k << m->shift),
1870                                        m->data.sets[k]);
1871                }
1872        } else {
1873                for (k = 0; k < 1024; k++) {
1874                        if (m->data.marked[k])
1875                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1876                                        oid_to_hex(&m->data.marked[k]->idx.oid));
1877                }
1878        }
1879}
1880
1881static void dump_marks(void)
1882{
1883        static struct lock_file mark_lock;
1884        FILE *f;
1885
1886        if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1887                return;
1888
1889        if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1890                failure |= error_errno("Unable to write marks file %s",
1891                                       export_marks_file);
1892                return;
1893        }
1894
1895        f = fdopen_lock_file(&mark_lock, "w");
1896        if (!f) {
1897                int saved_errno = errno;
1898                rollback_lock_file(&mark_lock);
1899                failure |= error("Unable to write marks file %s: %s",
1900                        export_marks_file, strerror(saved_errno));
1901                return;
1902        }
1903
1904        dump_marks_helper(f, 0, marks);
1905        if (commit_lock_file(&mark_lock)) {
1906                failure |= error_errno("Unable to write file %s",
1907                                       export_marks_file);
1908                return;
1909        }
1910}
1911
1912static void read_marks(void)
1913{
1914        char line[512];
1915        FILE *f = fopen(import_marks_file, "r");
1916        if (f)
1917                ;
1918        else if (import_marks_file_ignore_missing && errno == ENOENT)
1919                goto done; /* Marks file does not exist */
1920        else
1921                die_errno("cannot read '%s'", import_marks_file);
1922        while (fgets(line, sizeof(line), f)) {
1923                uintmax_t mark;
1924                char *end;
1925                struct object_id oid;
1926                struct object_entry *e;
1927
1928                end = strchr(line, '\n');
1929                if (line[0] != ':' || !end)
1930                        die("corrupt mark line: %s", line);
1931                *end = 0;
1932                mark = strtoumax(line + 1, &end, 10);
1933                if (!mark || end == line + 1
1934                        || *end != ' ' || get_oid_hex(end + 1, &oid))
1935                        die("corrupt mark line: %s", line);
1936                e = find_object(&oid);
1937                if (!e) {
1938                        enum object_type type = sha1_object_info(oid.hash, NULL);
1939                        if (type < 0)
1940                                die("object not found: %s", oid_to_hex(&oid));
1941                        e = insert_object(&oid);
1942                        e->type = type;
1943                        e->pack_id = MAX_PACK_ID;
1944                        e->idx.offset = 1; /* just not zero! */
1945                }
1946                insert_mark(mark, e);
1947        }
1948        fclose(f);
1949done:
1950        import_marks_file_done = 1;
1951}
1952
1953
1954static int read_next_command(void)
1955{
1956        static int stdin_eof = 0;
1957
1958        if (stdin_eof) {
1959                unread_command_buf = 0;
1960                return EOF;
1961        }
1962
1963        for (;;) {
1964                const char *p;
1965
1966                if (unread_command_buf) {
1967                        unread_command_buf = 0;
1968                } else {
1969                        struct recent_command *rc;
1970
1971                        strbuf_detach(&command_buf, NULL);
1972                        stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1973                        if (stdin_eof)
1974                                return EOF;
1975
1976                        if (!seen_data_command
1977                                && !starts_with(command_buf.buf, "feature ")
1978                                && !starts_with(command_buf.buf, "option ")) {
1979                                parse_argv();
1980                        }
1981
1982                        rc = rc_free;
1983                        if (rc)
1984                                rc_free = rc->next;
1985                        else {
1986                                rc = cmd_hist.next;
1987                                cmd_hist.next = rc->next;
1988                                cmd_hist.next->prev = &cmd_hist;
1989                                free(rc->buf);
1990                        }
1991
1992                        rc->buf = command_buf.buf;
1993                        rc->prev = cmd_tail;
1994                        rc->next = cmd_hist.prev;
1995                        rc->prev->next = rc;
1996                        cmd_tail = rc;
1997                }
1998                if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1999                        parse_get_mark(p);
2000                        continue;
2001                }
2002                if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
2003                        parse_cat_blob(p);
2004                        continue;
2005                }
2006                if (command_buf.buf[0] == '#')
2007                        continue;
2008                return 0;
2009        }
2010}
2011
2012static void skip_optional_lf(void)
2013{
2014        int term_char = fgetc(stdin);
2015        if (term_char != '\n' && term_char != EOF)
2016                ungetc(term_char, stdin);
2017}
2018
2019static void parse_mark(void)
2020{
2021        const char *v;
2022        if (skip_prefix(command_buf.buf, "mark :", &v)) {
2023                next_mark = strtoumax(v, NULL, 10);
2024                read_next_command();
2025        }
2026        else
2027                next_mark = 0;
2028}
2029
2030static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
2031{
2032        const char *data;
2033        strbuf_reset(sb);
2034
2035        if (!skip_prefix(command_buf.buf, "data ", &data))
2036                die("Expected 'data n' command, found: %s", command_buf.buf);
2037
2038        if (skip_prefix(data, "<<", &data)) {
2039                char *term = xstrdup(data);
2040                size_t term_len = command_buf.len - (data - command_buf.buf);
2041
2042                strbuf_detach(&command_buf, NULL);
2043                for (;;) {
2044                        if (strbuf_getline_lf(&command_buf, stdin) == EOF)
2045                                die("EOF in data (terminator '%s' not found)", term);
2046                        if (term_len == command_buf.len
2047                                && !strcmp(term, command_buf.buf))
2048                                break;
2049                        strbuf_addbuf(sb, &command_buf);
2050                        strbuf_addch(sb, '\n');
2051                }
2052                free(term);
2053        }
2054        else {
2055                uintmax_t len = strtoumax(data, NULL, 10);
2056                size_t n = 0, length = (size_t)len;
2057
2058                if (limit && limit < len) {
2059                        *len_res = len;
2060                        return 0;
2061                }
2062                if (length < len)
2063                        die("data is too large to use in this context");
2064
2065                while (n < length) {
2066                        size_t s = strbuf_fread(sb, length - n, stdin);
2067                        if (!s && feof(stdin))
2068                                die("EOF in data (%lu bytes remaining)",
2069                                        (unsigned long)(length - n));
2070                        n += s;
2071                }
2072        }
2073
2074        skip_optional_lf();
2075        return 1;
2076}
2077
2078static int validate_raw_date(const char *src, struct strbuf *result)
2079{
2080        const char *orig_src = src;
2081        char *endp;
2082        unsigned long num;
2083
2084        errno = 0;
2085
2086        num = strtoul(src, &endp, 10);
2087        /* NEEDSWORK: perhaps check for reasonable values? */
2088        if (errno || endp == src || *endp != ' ')
2089                return -1;
2090
2091        src = endp + 1;
2092        if (*src != '-' && *src != '+')
2093                return -1;
2094
2095        num = strtoul(src + 1, &endp, 10);
2096        if (errno || endp == src + 1 || *endp || 1400 < num)
2097                return -1;
2098
2099        strbuf_addstr(result, orig_src);
2100        return 0;
2101}
2102
2103static char *parse_ident(const char *buf)
2104{
2105        const char *ltgt;
2106        size_t name_len;
2107        struct strbuf ident = STRBUF_INIT;
2108
2109        /* ensure there is a space delimiter even if there is no name */
2110        if (*buf == '<')
2111                --buf;
2112
2113        ltgt = buf + strcspn(buf, "<>");
2114        if (*ltgt != '<')
2115                die("Missing < in ident string: %s", buf);
2116        if (ltgt != buf && ltgt[-1] != ' ')
2117                die("Missing space before < in ident string: %s", buf);
2118        ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2119        if (*ltgt != '>')
2120                die("Missing > in ident string: %s", buf);
2121        ltgt++;
2122        if (*ltgt != ' ')
2123                die("Missing space after > in ident string: %s", buf);
2124        ltgt++;
2125        name_len = ltgt - buf;
2126        strbuf_add(&ident, buf, name_len);
2127
2128        switch (whenspec) {
2129        case WHENSPEC_RAW:
2130                if (validate_raw_date(ltgt, &ident) < 0)
2131                        die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2132                break;
2133        case WHENSPEC_RFC2822:
2134                if (parse_date(ltgt, &ident) < 0)
2135                        die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2136                break;
2137        case WHENSPEC_NOW:
2138                if (strcmp("now", ltgt))
2139                        die("Date in ident must be 'now': %s", buf);
2140                datestamp(&ident);
2141                break;
2142        }
2143
2144        return strbuf_detach(&ident, NULL);
2145}
2146
2147static void parse_and_store_blob(
2148        struct last_object *last,
2149        struct object_id *oidout,
2150        uintmax_t mark)
2151{
2152        static struct strbuf buf = STRBUF_INIT;
2153        uintmax_t len;
2154
2155        if (parse_data(&buf, big_file_threshold, &len))
2156                store_object(OBJ_BLOB, &buf, last, oidout, mark);
2157        else {
2158                if (last) {
2159                        strbuf_release(&last->data);
2160                        last->offset = 0;
2161                        last->depth = 0;
2162                }
2163                stream_blob(len, oidout, mark);
2164                skip_optional_lf();
2165        }
2166}
2167
2168static void parse_new_blob(void)
2169{
2170        read_next_command();
2171        parse_mark();
2172        parse_and_store_blob(&last_blob, NULL, next_mark);
2173}
2174
2175static void unload_one_branch(void)
2176{
2177        while (cur_active_branches
2178                && cur_active_branches >= max_active_branches) {
2179                uintmax_t min_commit = ULONG_MAX;
2180                struct branch *e, *l = NULL, *p = NULL;
2181
2182                for (e = active_branches; e; e = e->active_next_branch) {
2183                        if (e->last_commit < min_commit) {
2184                                p = l;
2185                                min_commit = e->last_commit;
2186                        }
2187                        l = e;
2188                }
2189
2190                if (p) {
2191                        e = p->active_next_branch;
2192                        p->active_next_branch = e->active_next_branch;
2193                } else {
2194                        e = active_branches;
2195                        active_branches = e->active_next_branch;
2196                }
2197                e->active = 0;
2198                e->active_next_branch = NULL;
2199                if (e->branch_tree.tree) {
2200                        release_tree_content_recursive(e->branch_tree.tree);
2201                        e->branch_tree.tree = NULL;
2202                }
2203                cur_active_branches--;
2204        }
2205}
2206
2207static void load_branch(struct branch *b)
2208{
2209        load_tree(&b->branch_tree);
2210        if (!b->active) {
2211                b->active = 1;
2212                b->active_next_branch = active_branches;
2213                active_branches = b;
2214                cur_active_branches++;
2215                branch_load_count++;
2216        }
2217}
2218
2219static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2220{
2221        unsigned char fanout = 0;
2222        while ((num_notes >>= 8))
2223                fanout++;
2224        return fanout;
2225}
2226
2227static void construct_path_with_fanout(const char *hex_sha1,
2228                unsigned char fanout, char *path)
2229{
2230        unsigned int i = 0, j = 0;
2231        if (fanout >= the_hash_algo->rawsz)
2232                die("Too large fanout (%u)", fanout);
2233        while (fanout) {
2234                path[i++] = hex_sha1[j++];
2235                path[i++] = hex_sha1[j++];
2236                path[i++] = '/';
2237                fanout--;
2238        }
2239        memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2240        path[i + the_hash_algo->hexsz - j] = '\0';
2241}
2242
2243static uintmax_t do_change_note_fanout(
2244                struct tree_entry *orig_root, struct tree_entry *root,
2245                char *hex_oid, unsigned int hex_oid_len,
2246                char *fullpath, unsigned int fullpath_len,
2247                unsigned char fanout)
2248{
2249        struct tree_content *t;
2250        struct tree_entry *e, leaf;
2251        unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2252        uintmax_t num_notes = 0;
2253        struct object_id oid;
2254        char realpath[60];
2255
2256        if (!root->tree)
2257                load_tree(root);
2258        t = root->tree;
2259
2260        for (i = 0; t && i < t->entry_count; i++) {
2261                e = t->entries[i];
2262                tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2263                tmp_fullpath_len = fullpath_len;
2264
2265                /*
2266                 * We're interested in EITHER existing note entries (entries
2267                 * with exactly 40 hex chars in path, not including directory
2268                 * separators), OR directory entries that may contain note
2269                 * entries (with < 40 hex chars in path).
2270                 * Also, each path component in a note entry must be a multiple
2271                 * of 2 chars.
2272                 */
2273                if (!e->versions[1].mode ||
2274                    tmp_hex_oid_len > GIT_SHA1_HEXSZ ||
2275                    e->name->str_len % 2)
2276                        continue;
2277
2278                /* This _may_ be a note entry, or a subdir containing notes */
2279                memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2280                       e->name->str_len);
2281                if (tmp_fullpath_len)
2282                        fullpath[tmp_fullpath_len++] = '/';
2283                memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2284                       e->name->str_len);
2285                tmp_fullpath_len += e->name->str_len;
2286                fullpath[tmp_fullpath_len] = '\0';
2287
2288                if (tmp_hex_oid_len == GIT_SHA1_HEXSZ && !get_oid_hex(hex_oid, &oid)) {
2289                        /* This is a note entry */
2290                        if (fanout == 0xff) {
2291                                /* Counting mode, no rename */
2292                                num_notes++;
2293                                continue;
2294                        }
2295                        construct_path_with_fanout(hex_oid, fanout, realpath);
2296                        if (!strcmp(fullpath, realpath)) {
2297                                /* Note entry is in correct location */
2298                                num_notes++;
2299                                continue;
2300                        }
2301
2302                        /* Rename fullpath to realpath */
2303                        if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2304                                die("Failed to remove path %s", fullpath);
2305                        tree_content_set(orig_root, realpath,
2306                                &leaf.versions[1].oid,
2307                                leaf.versions[1].mode,
2308                                leaf.tree);
2309                } else if (S_ISDIR(e->versions[1].mode)) {
2310                        /* This is a subdir that may contain note entries */
2311                        num_notes += do_change_note_fanout(orig_root, e,
2312                                hex_oid, tmp_hex_oid_len,
2313                                fullpath, tmp_fullpath_len, fanout);
2314                }
2315
2316                /* The above may have reallocated the current tree_content */
2317                t = root->tree;
2318        }
2319        return num_notes;
2320}
2321
2322static uintmax_t change_note_fanout(struct tree_entry *root,
2323                unsigned char fanout)
2324{
2325        /*
2326         * The size of path is due to one slash between every two hex digits,
2327         * plus the terminating NUL.  Note that there is no slash at the end, so
2328         * the number of slashes is one less than half the number of hex
2329         * characters.
2330         */
2331        char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2332        return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2333}
2334
2335/*
2336 * Given a pointer into a string, parse a mark reference:
2337 *
2338 *   idnum ::= ':' bigint;
2339 *
2340 * Return the first character after the value in *endptr.
2341 *
2342 * Complain if the following character is not what is expected,
2343 * either a space or end of the string.
2344 */
2345static uintmax_t parse_mark_ref(const char *p, char **endptr)
2346{
2347        uintmax_t mark;
2348
2349        assert(*p == ':');
2350        p++;
2351        mark = strtoumax(p, endptr, 10);
2352        if (*endptr == p)
2353                die("No value after ':' in mark: %s", command_buf.buf);
2354        return mark;
2355}
2356
2357/*
2358 * Parse the mark reference, and complain if this is not the end of
2359 * the string.
2360 */
2361static uintmax_t parse_mark_ref_eol(const char *p)
2362{
2363        char *end;
2364        uintmax_t mark;
2365
2366        mark = parse_mark_ref(p, &end);
2367        if (*end != '\0')
2368                die("Garbage after mark: %s", command_buf.buf);
2369        return mark;
2370}
2371
2372/*
2373 * Parse the mark reference, demanding a trailing space.  Return a
2374 * pointer to the space.
2375 */
2376static uintmax_t parse_mark_ref_space(const char **p)
2377{
2378        uintmax_t mark;
2379        char *end;
2380
2381        mark = parse_mark_ref(*p, &end);
2382        if (*end++ != ' ')
2383                die("Missing space after mark: %s", command_buf.buf);
2384        *p = end;
2385        return mark;
2386}
2387
2388static void file_change_m(const char *p, struct branch *b)
2389{
2390        static struct strbuf uq = STRBUF_INIT;
2391        const char *endp;
2392        struct object_entry *oe;
2393        struct object_id oid;
2394        uint16_t mode, inline_data = 0;
2395
2396        p = get_mode(p, &mode);
2397        if (!p)
2398                die("Corrupt mode: %s", command_buf.buf);
2399        switch (mode) {
2400        case 0644:
2401        case 0755:
2402                mode |= S_IFREG;
2403        case S_IFREG | 0644:
2404        case S_IFREG | 0755:
2405        case S_IFLNK:
2406        case S_IFDIR:
2407        case S_IFGITLINK:
2408                /* ok */
2409                break;
2410        default:
2411                die("Corrupt mode: %s", command_buf.buf);
2412        }
2413
2414        if (*p == ':') {
2415                oe = find_mark(parse_mark_ref_space(&p));
2416                oidcpy(&oid, &oe->idx.oid);
2417        } else if (skip_prefix(p, "inline ", &p)) {
2418                inline_data = 1;
2419                oe = NULL; /* not used with inline_data, but makes gcc happy */
2420        } else {
2421                if (parse_oid_hex(p, &oid, &p))
2422                        die("Invalid dataref: %s", command_buf.buf);
2423                oe = find_object(&oid);
2424                if (*p++ != ' ')
2425                        die("Missing space after SHA1: %s", command_buf.buf);
2426        }
2427
2428        strbuf_reset(&uq);
2429        if (!unquote_c_style(&uq, p, &endp)) {
2430                if (*endp)
2431                        die("Garbage after path in: %s", command_buf.buf);
2432                p = uq.buf;
2433        }
2434
2435        /* Git does not track empty, non-toplevel directories. */
2436        if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2437                tree_content_remove(&b->branch_tree, p, NULL, 0);
2438                return;
2439        }
2440
2441        if (S_ISGITLINK(mode)) {
2442                if (inline_data)
2443                        die("Git links cannot be specified 'inline': %s",
2444                                command_buf.buf);
2445                else if (oe) {
2446                        if (oe->type != OBJ_COMMIT)
2447                                die("Not a commit (actually a %s): %s",
2448                                        type_name(oe->type), command_buf.buf);
2449                }
2450                /*
2451                 * Accept the sha1 without checking; it expected to be in
2452                 * another repository.
2453                 */
2454        } else if (inline_data) {
2455                if (S_ISDIR(mode))
2456                        die("Directories cannot be specified 'inline': %s",
2457                                command_buf.buf);
2458                if (p != uq.buf) {
2459                        strbuf_addstr(&uq, p);
2460                        p = uq.buf;
2461                }
2462                read_next_command();
2463                parse_and_store_blob(&last_blob, &oid, 0);
2464        } else {
2465                enum object_type expected = S_ISDIR(mode) ?
2466                                                OBJ_TREE: OBJ_BLOB;
2467                enum object_type type = oe ? oe->type :
2468                                        sha1_object_info(oid.hash, NULL);
2469                if (type < 0)
2470                        die("%s not found: %s",
2471                                        S_ISDIR(mode) ?  "Tree" : "Blob",
2472                                        command_buf.buf);
2473                if (type != expected)
2474                        die("Not a %s (actually a %s): %s",
2475                                type_name(expected), type_name(type),
2476                                command_buf.buf);
2477        }
2478
2479        if (!*p) {
2480                tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2481                return;
2482        }
2483        tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2484}
2485
2486static void file_change_d(const char *p, struct branch *b)
2487{
2488        static struct strbuf uq = STRBUF_INIT;
2489        const char *endp;
2490
2491        strbuf_reset(&uq);
2492        if (!unquote_c_style(&uq, p, &endp)) {
2493                if (*endp)
2494                        die("Garbage after path in: %s", command_buf.buf);
2495                p = uq.buf;
2496        }
2497        tree_content_remove(&b->branch_tree, p, NULL, 1);
2498}
2499
2500static void file_change_cr(const char *s, struct branch *b, int rename)
2501{
2502        const char *d;
2503        static struct strbuf s_uq = STRBUF_INIT;
2504        static struct strbuf d_uq = STRBUF_INIT;
2505        const char *endp;
2506        struct tree_entry leaf;
2507
2508        strbuf_reset(&s_uq);
2509        if (!unquote_c_style(&s_uq, s, &endp)) {
2510                if (*endp != ' ')
2511                        die("Missing space after source: %s", command_buf.buf);
2512        } else {
2513                endp = strchr(s, ' ');
2514                if (!endp)
2515                        die("Missing space after source: %s", command_buf.buf);
2516                strbuf_add(&s_uq, s, endp - s);
2517        }
2518        s = s_uq.buf;
2519
2520        endp++;
2521        if (!*endp)
2522                die("Missing dest: %s", command_buf.buf);
2523
2524        d = endp;
2525        strbuf_reset(&d_uq);
2526        if (!unquote_c_style(&d_uq, d, &endp)) {
2527                if (*endp)
2528                        die("Garbage after dest in: %s", command_buf.buf);
2529                d = d_uq.buf;
2530        }
2531
2532        memset(&leaf, 0, sizeof(leaf));
2533        if (rename)
2534                tree_content_remove(&b->branch_tree, s, &leaf, 1);
2535        else
2536                tree_content_get(&b->branch_tree, s, &leaf, 1);
2537        if (!leaf.versions[1].mode)
2538                die("Path %s not in branch", s);
2539        if (!*d) {      /* C "path/to/subdir" "" */
2540                tree_content_replace(&b->branch_tree,
2541                        &leaf.versions[1].oid,
2542                        leaf.versions[1].mode,
2543                        leaf.tree);
2544                return;
2545        }
2546        tree_content_set(&b->branch_tree, d,
2547                &leaf.versions[1].oid,
2548                leaf.versions[1].mode,
2549                leaf.tree);
2550}
2551
2552static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2553{
2554        static struct strbuf uq = STRBUF_INIT;
2555        struct object_entry *oe;
2556        struct branch *s;
2557        struct object_id oid, commit_oid;
2558        char path[60];
2559        uint16_t inline_data = 0;
2560        unsigned char new_fanout;
2561
2562        /*
2563         * When loading a branch, we don't traverse its tree to count the real
2564         * number of notes (too expensive to do this for all non-note refs).
2565         * This means that recently loaded notes refs might incorrectly have
2566         * b->num_notes == 0, and consequently, old_fanout might be wrong.
2567         *
2568         * Fix this by traversing the tree and counting the number of notes
2569         * when b->num_notes == 0. If the notes tree is truly empty, the
2570         * calculation should not take long.
2571         */
2572        if (b->num_notes == 0 && *old_fanout == 0) {
2573                /* Invoke change_note_fanout() in "counting mode". */
2574                b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2575                *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2576        }
2577
2578        /* Now parse the notemodify command. */
2579        /* <dataref> or 'inline' */
2580        if (*p == ':') {
2581                oe = find_mark(parse_mark_ref_space(&p));
2582                oidcpy(&oid, &oe->idx.oid);
2583        } else if (skip_prefix(p, "inline ", &p)) {
2584                inline_data = 1;
2585                oe = NULL; /* not used with inline_data, but makes gcc happy */
2586        } else {
2587                if (parse_oid_hex(p, &oid, &p))
2588                        die("Invalid dataref: %s", command_buf.buf);
2589                oe = find_object(&oid);
2590                if (*p++ != ' ')
2591                        die("Missing space after SHA1: %s", command_buf.buf);
2592        }
2593
2594        /* <commit-ish> */
2595        s = lookup_branch(p);
2596        if (s) {
2597                if (is_null_oid(&s->oid))
2598                        die("Can't add a note on empty branch.");
2599                oidcpy(&commit_oid, &s->oid);
2600        } else if (*p == ':') {
2601                uintmax_t commit_mark = parse_mark_ref_eol(p);
2602                struct object_entry *commit_oe = find_mark(commit_mark);
2603                if (commit_oe->type != OBJ_COMMIT)
2604                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2605                oidcpy(&commit_oid, &commit_oe->idx.oid);
2606        } else if (!get_oid(p, &commit_oid)) {
2607                unsigned long size;
2608                char *buf = read_object_with_reference(commit_oid.hash,
2609                        commit_type, &size, commit_oid.hash);
2610                if (!buf || size < 46)
2611                        die("Not a valid commit: %s", p);
2612                free(buf);
2613        } else
2614                die("Invalid ref name or SHA1 expression: %s", p);
2615
2616        if (inline_data) {
2617                if (p != uq.buf) {
2618                        strbuf_addstr(&uq, p);
2619                        p = uq.buf;
2620                }
2621                read_next_command();
2622                parse_and_store_blob(&last_blob, &oid, 0);
2623        } else if (oe) {
2624                if (oe->type != OBJ_BLOB)
2625                        die("Not a blob (actually a %s): %s",
2626                                type_name(oe->type), command_buf.buf);
2627        } else if (!is_null_oid(&oid)) {
2628                enum object_type type = sha1_object_info(oid.hash, NULL);
2629                if (type < 0)
2630                        die("Blob not found: %s", command_buf.buf);
2631                if (type != OBJ_BLOB)
2632                        die("Not a blob (actually a %s): %s",
2633                            type_name(type), command_buf.buf);
2634        }
2635
2636        construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2637        if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2638                b->num_notes--;
2639
2640        if (is_null_oid(&oid))
2641                return; /* nothing to insert */
2642
2643        b->num_notes++;
2644        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2645        construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2646        tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2647}
2648
2649static void file_change_deleteall(struct branch *b)
2650{
2651        release_tree_content_recursive(b->branch_tree.tree);
2652        oidclr(&b->branch_tree.versions[0].oid);
2653        oidclr(&b->branch_tree.versions[1].oid);
2654        load_tree(&b->branch_tree);
2655        b->num_notes = 0;
2656}
2657
2658static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2659{
2660        if (!buf || size < GIT_SHA1_HEXSZ + 6)
2661                die("Not a valid commit: %s", oid_to_hex(&b->oid));
2662        if (memcmp("tree ", buf, 5)
2663                || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2664                die("The commit %s is corrupt", oid_to_hex(&b->oid));
2665        oidcpy(&b->branch_tree.versions[0].oid,
2666               &b->branch_tree.versions[1].oid);
2667}
2668
2669static void parse_from_existing(struct branch *b)
2670{
2671        if (is_null_oid(&b->oid)) {
2672                oidclr(&b->branch_tree.versions[0].oid);
2673                oidclr(&b->branch_tree.versions[1].oid);
2674        } else {
2675                unsigned long size;
2676                char *buf;
2677
2678                buf = read_object_with_reference(b->oid.hash,
2679                                                 commit_type, &size,
2680                                                 b->oid.hash);
2681                parse_from_commit(b, buf, size);
2682                free(buf);
2683        }
2684}
2685
2686static int parse_from(struct branch *b)
2687{
2688        const char *from;
2689        struct branch *s;
2690        struct object_id oid;
2691
2692        if (!skip_prefix(command_buf.buf, "from ", &from))
2693                return 0;
2694
2695        oidcpy(&oid, &b->branch_tree.versions[1].oid);
2696
2697        s = lookup_branch(from);
2698        if (b == s)
2699                die("Can't create a branch from itself: %s", b->name);
2700        else if (s) {
2701                struct object_id *t = &s->branch_tree.versions[1].oid;
2702                oidcpy(&b->oid, &s->oid);
2703                oidcpy(&b->branch_tree.versions[0].oid, t);
2704                oidcpy(&b->branch_tree.versions[1].oid, t);
2705        } else if (*from == ':') {
2706                uintmax_t idnum = parse_mark_ref_eol(from);
2707                struct object_entry *oe = find_mark(idnum);
2708                if (oe->type != OBJ_COMMIT)
2709                        die("Mark :%" PRIuMAX " not a commit", idnum);
2710                if (oidcmp(&b->oid, &oe->idx.oid)) {
2711                        oidcpy(&b->oid, &oe->idx.oid);
2712                        if (oe->pack_id != MAX_PACK_ID) {
2713                                unsigned long size;
2714                                char *buf = gfi_unpack_entry(oe, &size);
2715                                parse_from_commit(b, buf, size);
2716                                free(buf);
2717                        } else
2718                                parse_from_existing(b);
2719                }
2720        } else if (!get_oid(from, &b->oid)) {
2721                parse_from_existing(b);
2722                if (is_null_oid(&b->oid))
2723                        b->delete = 1;
2724        }
2725        else
2726                die("Invalid ref name or SHA1 expression: %s", from);
2727
2728        if (b->branch_tree.tree && oidcmp(&oid, &b->branch_tree.versions[1].oid)) {
2729                release_tree_content_recursive(b->branch_tree.tree);
2730                b->branch_tree.tree = NULL;
2731        }
2732
2733        read_next_command();
2734        return 1;
2735}
2736
2737static struct hash_list *parse_merge(unsigned int *count)
2738{
2739        struct hash_list *list = NULL, **tail = &list, *n;
2740        const char *from;
2741        struct branch *s;
2742
2743        *count = 0;
2744        while (skip_prefix(command_buf.buf, "merge ", &from)) {
2745                n = xmalloc(sizeof(*n));
2746                s = lookup_branch(from);
2747                if (s)
2748                        oidcpy(&n->oid, &s->oid);
2749                else if (*from == ':') {
2750                        uintmax_t idnum = parse_mark_ref_eol(from);
2751                        struct object_entry *oe = find_mark(idnum);
2752                        if (oe->type != OBJ_COMMIT)
2753                                die("Mark :%" PRIuMAX " not a commit", idnum);
2754                        oidcpy(&n->oid, &oe->idx.oid);
2755                } else if (!get_oid(from, &n->oid)) {
2756                        unsigned long size;
2757                        char *buf = read_object_with_reference(n->oid.hash,
2758                                commit_type, &size, n->oid.hash);
2759                        if (!buf || size < 46)
2760                                die("Not a valid commit: %s", from);
2761                        free(buf);
2762                } else
2763                        die("Invalid ref name or SHA1 expression: %s", from);
2764
2765                n->next = NULL;
2766                *tail = n;
2767                tail = &n->next;
2768
2769                (*count)++;
2770                read_next_command();
2771        }
2772        return list;
2773}
2774
2775static void parse_new_commit(const char *arg)
2776{
2777        static struct strbuf msg = STRBUF_INIT;
2778        struct branch *b;
2779        char *author = NULL;
2780        char *committer = NULL;
2781        struct hash_list *merge_list = NULL;
2782        unsigned int merge_count;
2783        unsigned char prev_fanout, new_fanout;
2784        const char *v;
2785
2786        b = lookup_branch(arg);
2787        if (!b)
2788                b = new_branch(arg);
2789
2790        read_next_command();
2791        parse_mark();
2792        if (skip_prefix(command_buf.buf, "author ", &v)) {
2793                author = parse_ident(v);
2794                read_next_command();
2795        }
2796        if (skip_prefix(command_buf.buf, "committer ", &v)) {
2797                committer = parse_ident(v);
2798                read_next_command();
2799        }
2800        if (!committer)
2801                die("Expected committer but didn't get one");
2802        parse_data(&msg, 0, NULL);
2803        read_next_command();
2804        parse_from(b);
2805        merge_list = parse_merge(&merge_count);
2806
2807        /* ensure the branch is active/loaded */
2808        if (!b->branch_tree.tree || !max_active_branches) {
2809                unload_one_branch();
2810                load_branch(b);
2811        }
2812
2813        prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2814
2815        /* file_change* */
2816        while (command_buf.len > 0) {
2817                if (skip_prefix(command_buf.buf, "M ", &v))
2818                        file_change_m(v, b);
2819                else if (skip_prefix(command_buf.buf, "D ", &v))
2820                        file_change_d(v, b);
2821                else if (skip_prefix(command_buf.buf, "R ", &v))
2822                        file_change_cr(v, b, 1);
2823                else if (skip_prefix(command_buf.buf, "C ", &v))
2824                        file_change_cr(v, b, 0);
2825                else if (skip_prefix(command_buf.buf, "N ", &v))
2826                        note_change_n(v, b, &prev_fanout);
2827                else if (!strcmp("deleteall", command_buf.buf))
2828                        file_change_deleteall(b);
2829                else if (skip_prefix(command_buf.buf, "ls ", &v))
2830                        parse_ls(v, b);
2831                else {
2832                        unread_command_buf = 1;
2833                        break;
2834                }
2835                if (read_next_command() == EOF)
2836                        break;
2837        }
2838
2839        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2840        if (new_fanout != prev_fanout)
2841                b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2842
2843        /* build the tree and the commit */
2844        store_tree(&b->branch_tree);
2845        oidcpy(&b->branch_tree.versions[0].oid,
2846               &b->branch_tree.versions[1].oid);
2847
2848        strbuf_reset(&new_data);
2849        strbuf_addf(&new_data, "tree %s\n",
2850                oid_to_hex(&b->branch_tree.versions[1].oid));
2851        if (!is_null_oid(&b->oid))
2852                strbuf_addf(&new_data, "parent %s\n",
2853                            oid_to_hex(&b->oid));
2854        while (merge_list) {
2855                struct hash_list *next = merge_list->next;
2856                strbuf_addf(&new_data, "parent %s\n",
2857                            oid_to_hex(&merge_list->oid));
2858                free(merge_list);
2859                merge_list = next;
2860        }
2861        strbuf_addf(&new_data,
2862                "author %s\n"
2863                "committer %s\n"
2864                "\n",
2865                author ? author : committer, committer);
2866        strbuf_addbuf(&new_data, &msg);
2867        free(author);
2868        free(committer);
2869
2870        if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2871                b->pack_id = pack_id;
2872        b->last_commit = object_count_by_type[OBJ_COMMIT];
2873}
2874
2875static void parse_new_tag(const char *arg)
2876{
2877        static struct strbuf msg = STRBUF_INIT;
2878        const char *from;
2879        char *tagger;
2880        struct branch *s;
2881        struct tag *t;
2882        uintmax_t from_mark = 0;
2883        struct object_id oid;
2884        enum object_type type;
2885        const char *v;
2886
2887        t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2888        memset(t, 0, sizeof(struct tag));
2889        t->name = pool_strdup(arg);
2890        if (last_tag)
2891                last_tag->next_tag = t;
2892        else
2893                first_tag = t;
2894        last_tag = t;
2895        read_next_command();
2896
2897        /* from ... */
2898        if (!skip_prefix(command_buf.buf, "from ", &from))
2899                die("Expected from command, got %s", command_buf.buf);
2900        s = lookup_branch(from);
2901        if (s) {
2902                if (is_null_oid(&s->oid))
2903                        die("Can't tag an empty branch.");
2904                oidcpy(&oid, &s->oid);
2905                type = OBJ_COMMIT;
2906        } else if (*from == ':') {
2907                struct object_entry *oe;
2908                from_mark = parse_mark_ref_eol(from);
2909                oe = find_mark(from_mark);
2910                type = oe->type;
2911                oidcpy(&oid, &oe->idx.oid);
2912        } else if (!get_oid(from, &oid)) {
2913                struct object_entry *oe = find_object(&oid);
2914                if (!oe) {
2915                        type = sha1_object_info(oid.hash, NULL);
2916                        if (type < 0)
2917                                die("Not a valid object: %s", from);
2918                } else
2919                        type = oe->type;
2920        } else
2921                die("Invalid ref name or SHA1 expression: %s", from);
2922        read_next_command();
2923
2924        /* tagger ... */
2925        if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2926                tagger = parse_ident(v);
2927                read_next_command();
2928        } else
2929                tagger = NULL;
2930
2931        /* tag payload/message */
2932        parse_data(&msg, 0, NULL);
2933
2934        /* build the tag object */
2935        strbuf_reset(&new_data);
2936
2937        strbuf_addf(&new_data,
2938                    "object %s\n"
2939                    "type %s\n"
2940                    "tag %s\n",
2941                    oid_to_hex(&oid), type_name(type), t->name);
2942        if (tagger)
2943                strbuf_addf(&new_data,
2944                            "tagger %s\n", tagger);
2945        strbuf_addch(&new_data, '\n');
2946        strbuf_addbuf(&new_data, &msg);
2947        free(tagger);
2948
2949        if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2950                t->pack_id = MAX_PACK_ID;
2951        else
2952                t->pack_id = pack_id;
2953}
2954
2955static void parse_reset_branch(const char *arg)
2956{
2957        struct branch *b;
2958
2959        b = lookup_branch(arg);
2960        if (b) {
2961                oidclr(&b->oid);
2962                oidclr(&b->branch_tree.versions[0].oid);
2963                oidclr(&b->branch_tree.versions[1].oid);
2964                if (b->branch_tree.tree) {
2965                        release_tree_content_recursive(b->branch_tree.tree);
2966                        b->branch_tree.tree = NULL;
2967                }
2968        }
2969        else
2970                b = new_branch(arg);
2971        read_next_command();
2972        parse_from(b);
2973        if (command_buf.len > 0)
2974                unread_command_buf = 1;
2975}
2976
2977static void cat_blob_write(const char *buf, unsigned long size)
2978{
2979        if (write_in_full(cat_blob_fd, buf, size) < 0)
2980                die_errno("Write to frontend failed");
2981}
2982
2983static void cat_blob(struct object_entry *oe, struct object_id *oid)
2984{
2985        struct strbuf line = STRBUF_INIT;
2986        unsigned long size;
2987        enum object_type type = 0;
2988        char *buf;
2989
2990        if (!oe || oe->pack_id == MAX_PACK_ID) {
2991                buf = read_sha1_file(oid->hash, &type, &size);
2992        } else {
2993                type = oe->type;
2994                buf = gfi_unpack_entry(oe, &size);
2995        }
2996
2997        /*
2998         * Output based on batch_one_object() from cat-file.c.
2999         */
3000        if (type <= 0) {
3001                strbuf_reset(&line);
3002                strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
3003                cat_blob_write(line.buf, line.len);
3004                strbuf_release(&line);
3005                free(buf);
3006                return;
3007        }
3008        if (!buf)
3009                die("Can't read object %s", oid_to_hex(oid));
3010        if (type != OBJ_BLOB)
3011                die("Object %s is a %s but a blob was expected.",
3012                    oid_to_hex(oid), type_name(type));
3013        strbuf_reset(&line);
3014        strbuf_addf(&line, "%s %s %lu\n", oid_to_hex(oid),
3015                                                type_name(type), size);
3016        cat_blob_write(line.buf, line.len);
3017        strbuf_release(&line);
3018        cat_blob_write(buf, size);
3019        cat_blob_write("\n", 1);
3020        if (oe && oe->pack_id == pack_id) {
3021                last_blob.offset = oe->idx.offset;
3022                strbuf_attach(&last_blob.data, buf, size, size);
3023                last_blob.depth = oe->depth;
3024        } else
3025                free(buf);
3026}
3027
3028static void parse_get_mark(const char *p)
3029{
3030        struct object_entry *oe;
3031        char output[GIT_MAX_HEXSZ + 2];
3032
3033        /* get-mark SP <object> LF */
3034        if (*p != ':')
3035                die("Not a mark: %s", p);
3036
3037        oe = find_mark(parse_mark_ref_eol(p));
3038        if (!oe)
3039                die("Unknown mark: %s", command_buf.buf);
3040
3041        xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
3042        cat_blob_write(output, GIT_SHA1_HEXSZ + 1);
3043}
3044
3045static void parse_cat_blob(const char *p)
3046{
3047        struct object_entry *oe;
3048        struct object_id oid;
3049
3050        /* cat-blob SP <object> LF */
3051        if (*p == ':') {
3052                oe = find_mark(parse_mark_ref_eol(p));
3053                if (!oe)
3054                        die("Unknown mark: %s", command_buf.buf);
3055                oidcpy(&oid, &oe->idx.oid);
3056        } else {
3057                if (parse_oid_hex(p, &oid, &p))
3058                        die("Invalid dataref: %s", command_buf.buf);
3059                if (*p)
3060                        die("Garbage after SHA1: %s", command_buf.buf);
3061                oe = find_object(&oid);
3062        }
3063
3064        cat_blob(oe, &oid);
3065}
3066
3067static struct object_entry *dereference(struct object_entry *oe,
3068                                        struct object_id *oid)
3069{
3070        unsigned long size;
3071        char *buf = NULL;
3072        if (!oe) {
3073                enum object_type type = sha1_object_info(oid->hash, NULL);
3074                if (type < 0)
3075                        die("object not found: %s", oid_to_hex(oid));
3076                /* cache it! */
3077                oe = insert_object(oid);
3078                oe->type = type;
3079                oe->pack_id = MAX_PACK_ID;
3080                oe->idx.offset = 1;
3081        }
3082        switch (oe->type) {
3083        case OBJ_TREE:  /* easy case. */
3084                return oe;
3085        case OBJ_COMMIT:
3086        case OBJ_TAG:
3087                break;
3088        default:
3089                die("Not a tree-ish: %s", command_buf.buf);
3090        }
3091
3092        if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
3093                buf = gfi_unpack_entry(oe, &size);
3094        } else {
3095                enum object_type unused;
3096                buf = read_sha1_file(oid->hash, &unused, &size);
3097        }
3098        if (!buf)
3099                die("Can't load object %s", oid_to_hex(oid));
3100
3101        /* Peel one layer. */
3102        switch (oe->type) {
3103        case OBJ_TAG:
3104                if (size < GIT_SHA1_HEXSZ + strlen("object ") ||
3105                    get_oid_hex(buf + strlen("object "), oid))
3106                        die("Invalid SHA1 in tag: %s", command_buf.buf);
3107                break;
3108        case OBJ_COMMIT:
3109                if (size < GIT_SHA1_HEXSZ + strlen("tree ") ||
3110                    get_oid_hex(buf + strlen("tree "), oid))
3111                        die("Invalid SHA1 in commit: %s", command_buf.buf);
3112        }
3113
3114        free(buf);
3115        return find_object(oid);
3116}
3117
3118static struct object_entry *parse_treeish_dataref(const char **p)
3119{
3120        struct object_id oid;
3121        struct object_entry *e;
3122
3123        if (**p == ':') {       /* <mark> */
3124                e = find_mark(parse_mark_ref_space(p));
3125                if (!e)
3126                        die("Unknown mark: %s", command_buf.buf);
3127                oidcpy(&oid, &e->idx.oid);
3128        } else {        /* <sha1> */
3129                if (parse_oid_hex(*p, &oid, p))
3130                        die("Invalid dataref: %s", command_buf.buf);
3131                e = find_object(&oid);
3132                if (*(*p)++ != ' ')
3133                        die("Missing space after tree-ish: %s", command_buf.buf);
3134        }
3135
3136        while (!e || e->type != OBJ_TREE)
3137                e = dereference(e, &oid);
3138        return e;
3139}
3140
3141static void print_ls(int mode, const unsigned char *sha1, const char *path)
3142{
3143        static struct strbuf line = STRBUF_INIT;
3144
3145        /* See show_tree(). */
3146        const char *type =
3147                S_ISGITLINK(mode) ? commit_type :
3148                S_ISDIR(mode) ? tree_type :
3149                blob_type;
3150
3151        if (!mode) {
3152                /* missing SP path LF */
3153                strbuf_reset(&line);
3154                strbuf_addstr(&line, "missing ");
3155                quote_c_style(path, &line, NULL, 0);
3156                strbuf_addch(&line, '\n');
3157        } else {
3158                /* mode SP type SP object_name TAB path LF */
3159                strbuf_reset(&line);
3160                strbuf_addf(&line, "%06o %s %s\t",
3161                                mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3162                quote_c_style(path, &line, NULL, 0);
3163                strbuf_addch(&line, '\n');
3164        }
3165        cat_blob_write(line.buf, line.len);
3166}
3167
3168static void parse_ls(const char *p, struct branch *b)
3169{
3170        struct tree_entry *root = NULL;
3171        struct tree_entry leaf = {NULL};
3172
3173        /* ls SP (<tree-ish> SP)? <path> */
3174        if (*p == '"') {
3175                if (!b)
3176                        die("Not in a commit: %s", command_buf.buf);
3177                root = &b->branch_tree;
3178        } else {
3179                struct object_entry *e = parse_treeish_dataref(&p);
3180                root = new_tree_entry();
3181                oidcpy(&root->versions[1].oid, &e->idx.oid);
3182                if (!is_null_oid(&root->versions[1].oid))
3183                        root->versions[1].mode = S_IFDIR;
3184                load_tree(root);
3185        }
3186        if (*p == '"') {
3187                static struct strbuf uq = STRBUF_INIT;
3188                const char *endp;
3189                strbuf_reset(&uq);
3190                if (unquote_c_style(&uq, p, &endp))
3191                        die("Invalid path: %s", command_buf.buf);
3192                if (*endp)
3193                        die("Garbage after path in: %s", command_buf.buf);
3194                p = uq.buf;
3195        }
3196        tree_content_get(root, p, &leaf, 1);
3197        /*
3198         * A directory in preparation would have a sha1 of zero
3199         * until it is saved.  Save, for simplicity.
3200         */
3201        if (S_ISDIR(leaf.versions[1].mode))
3202                store_tree(&leaf);
3203
3204        print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3205        if (leaf.tree)
3206                release_tree_content_recursive(leaf.tree);
3207        if (!b || root != &b->branch_tree)
3208                release_tree_entry(root);
3209}
3210
3211static void checkpoint(void)
3212{
3213        checkpoint_requested = 0;
3214        if (object_count) {
3215                cycle_packfile();
3216        }
3217        dump_branches();
3218        dump_tags();
3219        dump_marks();
3220}
3221
3222static void parse_checkpoint(void)
3223{
3224        checkpoint_requested = 1;
3225        skip_optional_lf();
3226}
3227
3228static void parse_progress(void)
3229{
3230        fwrite(command_buf.buf, 1, command_buf.len, stdout);
3231        fputc('\n', stdout);
3232        fflush(stdout);
3233        skip_optional_lf();
3234}
3235
3236static char* make_fast_import_path(const char *path)
3237{
3238        if (!relative_marks_paths || is_absolute_path(path))
3239                return xstrdup(path);
3240        return git_pathdup("info/fast-import/%s", path);
3241}
3242
3243static void option_import_marks(const char *marks,
3244                                        int from_stream, int ignore_missing)
3245{
3246        if (import_marks_file) {
3247                if (from_stream)
3248                        die("Only one import-marks command allowed per stream");
3249
3250                /* read previous mark file */
3251                if(!import_marks_file_from_stream)
3252                        read_marks();
3253        }
3254
3255        import_marks_file = make_fast_import_path(marks);
3256        safe_create_leading_directories_const(import_marks_file);
3257        import_marks_file_from_stream = from_stream;
3258        import_marks_file_ignore_missing = ignore_missing;
3259}
3260
3261static void option_date_format(const char *fmt)
3262{
3263        if (!strcmp(fmt, "raw"))
3264                whenspec = WHENSPEC_RAW;
3265        else if (!strcmp(fmt, "rfc2822"))
3266                whenspec = WHENSPEC_RFC2822;
3267        else if (!strcmp(fmt, "now"))
3268                whenspec = WHENSPEC_NOW;
3269        else
3270                die("unknown --date-format argument %s", fmt);
3271}
3272
3273static unsigned long ulong_arg(const char *option, const char *arg)
3274{
3275        char *endptr;
3276        unsigned long rv = strtoul(arg, &endptr, 0);
3277        if (strchr(arg, '-') || endptr == arg || *endptr)
3278                die("%s: argument must be a non-negative integer", option);
3279        return rv;
3280}
3281
3282static void option_depth(const char *depth)
3283{
3284        max_depth = ulong_arg("--depth", depth);
3285        if (max_depth > MAX_DEPTH)
3286                die("--depth cannot exceed %u", MAX_DEPTH);
3287}
3288
3289static void option_active_branches(const char *branches)
3290{
3291        max_active_branches = ulong_arg("--active-branches", branches);
3292}
3293
3294static void option_export_marks(const char *marks)
3295{
3296        export_marks_file = make_fast_import_path(marks);
3297        safe_create_leading_directories_const(export_marks_file);
3298}
3299
3300static void option_cat_blob_fd(const char *fd)
3301{
3302        unsigned long n = ulong_arg("--cat-blob-fd", fd);
3303        if (n > (unsigned long) INT_MAX)
3304                die("--cat-blob-fd cannot exceed %d", INT_MAX);
3305        cat_blob_fd = (int) n;
3306}
3307
3308static void option_export_pack_edges(const char *edges)
3309{
3310        if (pack_edges)
3311                fclose(pack_edges);
3312        pack_edges = xfopen(edges, "a");
3313}
3314
3315static int parse_one_option(const char *option)
3316{
3317        if (skip_prefix(option, "max-pack-size=", &option)) {
3318                unsigned long v;
3319                if (!git_parse_ulong(option, &v))
3320                        return 0;
3321                if (v < 8192) {
3322                        warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3323                        v *= 1024 * 1024;
3324                } else if (v < 1024 * 1024) {
3325                        warning("minimum max-pack-size is 1 MiB");
3326                        v = 1024 * 1024;
3327                }
3328                max_packsize = v;
3329        } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3330                unsigned long v;
3331                if (!git_parse_ulong(option, &v))
3332                        return 0;
3333                big_file_threshold = v;
3334        } else if (skip_prefix(option, "depth=", &option)) {
3335                option_depth(option);
3336        } else if (skip_prefix(option, "active-branches=", &option)) {
3337                option_active_branches(option);
3338        } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3339                option_export_pack_edges(option);
3340        } else if (starts_with(option, "quiet")) {
3341                show_stats = 0;
3342        } else if (starts_with(option, "stats")) {
3343                show_stats = 1;
3344        } else {
3345                return 0;
3346        }
3347
3348        return 1;
3349}
3350
3351static int parse_one_feature(const char *feature, int from_stream)
3352{
3353        const char *arg;
3354
3355        if (skip_prefix(feature, "date-format=", &arg)) {
3356                option_date_format(arg);
3357        } else if (skip_prefix(feature, "import-marks=", &arg)) {
3358                option_import_marks(arg, from_stream, 0);
3359        } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3360                option_import_marks(arg, from_stream, 1);
3361        } else if (skip_prefix(feature, "export-marks=", &arg)) {
3362                option_export_marks(arg);
3363        } else if (!strcmp(feature, "get-mark")) {
3364                ; /* Don't die - this feature is supported */
3365        } else if (!strcmp(feature, "cat-blob")) {
3366                ; /* Don't die - this feature is supported */
3367        } else if (!strcmp(feature, "relative-marks")) {
3368                relative_marks_paths = 1;
3369        } else if (!strcmp(feature, "no-relative-marks")) {
3370                relative_marks_paths = 0;
3371        } else if (!strcmp(feature, "done")) {
3372                require_explicit_termination = 1;
3373        } else if (!strcmp(feature, "force")) {
3374                force_update = 1;
3375        } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3376                ; /* do nothing; we have the feature */
3377        } else {
3378                return 0;
3379        }
3380
3381        return 1;
3382}
3383
3384static void parse_feature(const char *feature)
3385{
3386        if (seen_data_command)
3387                die("Got feature command '%s' after data command", feature);
3388
3389        if (parse_one_feature(feature, 1))
3390                return;
3391
3392        die("This version of fast-import does not support feature %s.", feature);
3393}
3394
3395static void parse_option(const char *option)
3396{
3397        if (seen_data_command)
3398                die("Got option command '%s' after data command", option);
3399
3400        if (parse_one_option(option))
3401                return;
3402
3403        die("This version of fast-import does not support option: %s", option);
3404}
3405
3406static void git_pack_config(void)
3407{
3408        int indexversion_value;
3409        int limit;
3410        unsigned long packsizelimit_value;
3411
3412        if (!git_config_get_ulong("pack.depth", &max_depth)) {
3413                if (max_depth > MAX_DEPTH)
3414                        max_depth = MAX_DEPTH;
3415        }
3416        if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3417                pack_idx_opts.version = indexversion_value;
3418                if (pack_idx_opts.version > 2)
3419                        git_die_config("pack.indexversion",
3420                                        "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3421        }
3422        if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3423                max_packsize = packsizelimit_value;
3424
3425        if (!git_config_get_int("fastimport.unpacklimit", &limit))
3426                unpack_limit = limit;
3427        else if (!git_config_get_int("transfer.unpacklimit", &limit))
3428                unpack_limit = limit;
3429
3430        git_config(git_default_config, NULL);
3431}
3432
3433static const char fast_import_usage[] =
3434"git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3435
3436static void parse_argv(void)
3437{
3438        unsigned int i;
3439
3440        for (i = 1; i < global_argc; i++) {
3441                const char *a = global_argv[i];
3442
3443                if (*a != '-' || !strcmp(a, "--"))
3444                        break;
3445
3446                if (!skip_prefix(a, "--", &a))
3447                        die("unknown option %s", a);
3448
3449                if (parse_one_option(a))
3450                        continue;
3451
3452                if (parse_one_feature(a, 0))
3453                        continue;
3454
3455                if (skip_prefix(a, "cat-blob-fd=", &a)) {
3456                        option_cat_blob_fd(a);
3457                        continue;
3458                }
3459
3460                die("unknown option --%s", a);
3461        }
3462        if (i != global_argc)
3463                usage(fast_import_usage);
3464
3465        seen_data_command = 1;
3466        if (import_marks_file)
3467                read_marks();
3468}
3469
3470int cmd_main(int argc, const char **argv)
3471{
3472        unsigned int i;
3473
3474        if (argc == 2 && !strcmp(argv[1], "-h"))
3475                usage(fast_import_usage);
3476
3477        setup_git_directory();
3478        reset_pack_idx_option(&pack_idx_opts);
3479        git_pack_config();
3480
3481        alloc_objects(object_entry_alloc);
3482        strbuf_init(&command_buf, 0);
3483        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3484        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3485        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3486        marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3487
3488        global_argc = argc;
3489        global_argv = argv;
3490
3491        rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3492        for (i = 0; i < (cmd_save - 1); i++)
3493                rc_free[i].next = &rc_free[i + 1];
3494        rc_free[cmd_save - 1].next = NULL;
3495
3496        prepare_packed_git();
3497        start_packfile();
3498        set_die_routine(die_nicely);
3499        set_checkpoint_signal();
3500        while (read_next_command() != EOF) {
3501                const char *v;
3502                if (!strcmp("blob", command_buf.buf))
3503                        parse_new_blob();
3504                else if (skip_prefix(command_buf.buf, "ls ", &v))
3505                        parse_ls(v, NULL);
3506                else if (skip_prefix(command_buf.buf, "commit ", &v))
3507                        parse_new_commit(v);
3508                else if (skip_prefix(command_buf.buf, "tag ", &v))
3509                        parse_new_tag(v);
3510                else if (skip_prefix(command_buf.buf, "reset ", &v))
3511                        parse_reset_branch(v);
3512                else if (!strcmp("checkpoint", command_buf.buf))
3513                        parse_checkpoint();
3514                else if (!strcmp("done", command_buf.buf))
3515                        break;
3516                else if (starts_with(command_buf.buf, "progress "))
3517                        parse_progress();
3518                else if (skip_prefix(command_buf.buf, "feature ", &v))
3519                        parse_feature(v);
3520                else if (skip_prefix(command_buf.buf, "option git ", &v))
3521                        parse_option(v);
3522                else if (starts_with(command_buf.buf, "option "))
3523                        /* ignore non-git options*/;
3524                else
3525                        die("Unsupported command: %s", command_buf.buf);
3526
3527                if (checkpoint_requested)
3528                        checkpoint();
3529        }
3530
3531        /* argv hasn't been parsed yet, do so */
3532        if (!seen_data_command)
3533                parse_argv();
3534
3535        if (require_explicit_termination && feof(stdin))
3536                die("stream ends early");
3537
3538        end_packfile();
3539
3540        dump_branches();
3541        dump_tags();
3542        unkeep_all_packs();
3543        dump_marks();
3544
3545        if (pack_edges)
3546                fclose(pack_edges);
3547
3548        if (show_stats) {
3549                uintmax_t total_count = 0, duplicate_count = 0;
3550                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3551                        total_count += object_count_by_type[i];
3552                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3553                        duplicate_count += duplicate_count_by_type[i];
3554
3555                fprintf(stderr, "%s statistics:\n", argv[0]);
3556                fprintf(stderr, "---------------------------------------------------------------------\n");
3557                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3558                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3559                fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB], delta_count_attempts_by_type[OBJ_BLOB]);
3560                fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE], delta_count_attempts_by_type[OBJ_TREE]);
3561                fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT], delta_count_attempts_by_type[OBJ_COMMIT]);
3562                fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG], delta_count_attempts_by_type[OBJ_TAG]);
3563                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3564                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3565                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3566                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3567                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)((total_allocd + fi_mem_pool.pool_alloc) /1024));
3568                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3569                fprintf(stderr, "---------------------------------------------------------------------\n");
3570                pack_report();
3571                fprintf(stderr, "---------------------------------------------------------------------\n");
3572                fprintf(stderr, "\n");
3573        }
3574
3575        return failure ? 1 : 0;
3576}