fast-import.con commit midx: use midx in approximate_object_count (b8990fb)
   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 "repository.h"
 158#include "config.h"
 159#include "lockfile.h"
 160#include "object.h"
 161#include "blob.h"
 162#include "tree.h"
 163#include "commit.h"
 164#include "delta.h"
 165#include "pack.h"
 166#include "refs.h"
 167#include "csum-file.h"
 168#include "quote.h"
 169#include "dir.h"
 170#include "run-command.h"
 171#include "packfile.h"
 172#include "object-store.h"
 173#include "mem-pool.h"
 174
 175#define PACK_ID_BITS 16
 176#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 177#define DEPTH_BITS 13
 178#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 179
 180/*
 181 * We abuse the setuid bit on directories to mean "do not delta".
 182 */
 183#define NO_DELTA S_ISUID
 184
 185struct object_entry {
 186        struct pack_idx_entry idx;
 187        struct object_entry *next;
 188        uint32_t type : TYPE_BITS,
 189                pack_id : PACK_ID_BITS,
 190                depth : DEPTH_BITS;
 191};
 192
 193struct object_entry_pool {
 194        struct object_entry_pool *next_pool;
 195        struct object_entry *next_free;
 196        struct object_entry *end;
 197        struct object_entry entries[FLEX_ARRAY]; /* more */
 198};
 199
 200struct mark_set {
 201        union {
 202                struct object_entry *marked[1024];
 203                struct mark_set *sets[1024];
 204        } data;
 205        unsigned int shift;
 206};
 207
 208struct last_object {
 209        struct strbuf data;
 210        off_t offset;
 211        unsigned int depth;
 212        unsigned no_swap : 1;
 213};
 214
 215struct atom_str {
 216        struct atom_str *next_atom;
 217        unsigned short str_len;
 218        char str_dat[FLEX_ARRAY]; /* more */
 219};
 220
 221struct tree_content;
 222struct tree_entry {
 223        struct tree_content *tree;
 224        struct atom_str *name;
 225        struct tree_entry_ms {
 226                uint16_t mode;
 227                struct object_id oid;
 228        } versions[2];
 229};
 230
 231struct tree_content {
 232        unsigned int entry_capacity; /* must match avail_tree_content */
 233        unsigned int entry_count;
 234        unsigned int delta_depth;
 235        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 236};
 237
 238struct avail_tree_content {
 239        unsigned int entry_capacity; /* must match tree_content */
 240        struct avail_tree_content *next_avail;
 241};
 242
 243struct branch {
 244        struct branch *table_next_branch;
 245        struct branch *active_next_branch;
 246        const char *name;
 247        struct tree_entry branch_tree;
 248        uintmax_t last_commit;
 249        uintmax_t num_notes;
 250        unsigned active : 1;
 251        unsigned delete : 1;
 252        unsigned pack_id : PACK_ID_BITS;
 253        struct object_id oid;
 254};
 255
 256struct tag {
 257        struct tag *next_tag;
 258        const char *name;
 259        unsigned int pack_id;
 260        struct object_id oid;
 261};
 262
 263struct hash_list {
 264        struct hash_list *next;
 265        struct object_id oid;
 266};
 267
 268typedef enum {
 269        WHENSPEC_RAW = 1,
 270        WHENSPEC_RFC2822,
 271        WHENSPEC_NOW
 272} whenspec_type;
 273
 274struct recent_command {
 275        struct recent_command *prev;
 276        struct recent_command *next;
 277        char *buf;
 278};
 279
 280/* Configured limits on output */
 281static unsigned long max_depth = 50;
 282static off_t max_packsize;
 283static int unpack_limit = 100;
 284static int force_update;
 285
 286/* Stats and misc. counters */
 287static uintmax_t alloc_count;
 288static uintmax_t marks_set_count;
 289static uintmax_t object_count_by_type[1 << TYPE_BITS];
 290static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 291static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 292static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
 293static unsigned long object_count;
 294static unsigned long branch_count;
 295static unsigned long branch_load_count;
 296static int failure;
 297static FILE *pack_edges;
 298static unsigned int show_stats = 1;
 299static int global_argc;
 300static const char **global_argv;
 301
 302/* Memory pools */
 303static struct mem_pool fi_mem_pool =  {NULL, 2*1024*1024 -
 304                                       sizeof(struct mp_block), 0 };
 305
 306/* Atom management */
 307static unsigned int atom_table_sz = 4451;
 308static unsigned int atom_cnt;
 309static struct atom_str **atom_table;
 310
 311/* The .pack file being generated */
 312static struct pack_idx_option pack_idx_opts;
 313static unsigned int pack_id;
 314static struct hashfile *pack_file;
 315static struct packed_git *pack_data;
 316static struct packed_git **all_packs;
 317static off_t pack_size;
 318
 319/* Table of objects we've written. */
 320static unsigned int object_entry_alloc = 5000;
 321static struct object_entry_pool *blocks;
 322static struct object_entry *object_table[1 << 16];
 323static struct mark_set *marks;
 324static const char *export_marks_file;
 325static const char *import_marks_file;
 326static int import_marks_file_from_stream;
 327static int import_marks_file_ignore_missing;
 328static int import_marks_file_done;
 329static int relative_marks_paths;
 330
 331/* Our last blob */
 332static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 333
 334/* Tree management */
 335static unsigned int tree_entry_alloc = 1000;
 336static void *avail_tree_entry;
 337static unsigned int avail_tree_table_sz = 100;
 338static struct avail_tree_content **avail_tree_table;
 339static size_t tree_entry_allocd;
 340static struct strbuf old_tree = STRBUF_INIT;
 341static struct strbuf new_tree = STRBUF_INIT;
 342
 343/* Branch data */
 344static unsigned long max_active_branches = 5;
 345static unsigned long cur_active_branches;
 346static unsigned long branch_table_sz = 1039;
 347static struct branch **branch_table;
 348static struct branch *active_branches;
 349
 350/* Tag data */
 351static struct tag *first_tag;
 352static struct tag *last_tag;
 353
 354/* Input stream parsing */
 355static whenspec_type whenspec = WHENSPEC_RAW;
 356static struct strbuf command_buf = STRBUF_INIT;
 357static int unread_command_buf;
 358static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 359static struct recent_command *cmd_tail = &cmd_hist;
 360static struct recent_command *rc_free;
 361static unsigned int cmd_save = 100;
 362static uintmax_t next_mark;
 363static struct strbuf new_data = STRBUF_INIT;
 364static int seen_data_command;
 365static int require_explicit_termination;
 366
 367/* Signal handling */
 368static volatile sig_atomic_t checkpoint_requested;
 369
 370/* Where to write output of cat-blob commands */
 371static int cat_blob_fd = STDOUT_FILENO;
 372
 373static void parse_argv(void);
 374static void parse_get_mark(const char *p);
 375static void parse_cat_blob(const char *p);
 376static void parse_ls(const char *p, struct branch *b);
 377
 378static void write_branch_report(FILE *rpt, struct branch *b)
 379{
 380        fprintf(rpt, "%s:\n", b->name);
 381
 382        fprintf(rpt, "  status      :");
 383        if (b->active)
 384                fputs(" active", rpt);
 385        if (b->branch_tree.tree)
 386                fputs(" loaded", rpt);
 387        if (is_null_oid(&b->branch_tree.versions[1].oid))
 388                fputs(" dirty", rpt);
 389        fputc('\n', rpt);
 390
 391        fprintf(rpt, "  tip commit  : %s\n", oid_to_hex(&b->oid));
 392        fprintf(rpt, "  old tree    : %s\n",
 393                oid_to_hex(&b->branch_tree.versions[0].oid));
 394        fprintf(rpt, "  cur tree    : %s\n",
 395                oid_to_hex(&b->branch_tree.versions[1].oid));
 396        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 397
 398        fputs("  last pack   : ", rpt);
 399        if (b->pack_id < MAX_PACK_ID)
 400                fprintf(rpt, "%u", b->pack_id);
 401        fputc('\n', rpt);
 402
 403        fputc('\n', rpt);
 404}
 405
 406static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 407
 408static void write_crash_report(const char *err)
 409{
 410        char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 411        FILE *rpt = fopen(loc, "w");
 412        struct branch *b;
 413        unsigned long lu;
 414        struct recent_command *rc;
 415
 416        if (!rpt) {
 417                error_errno("can't write crash report %s", loc);
 418                free(loc);
 419                return;
 420        }
 421
 422        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 423
 424        fprintf(rpt, "fast-import crash report:\n");
 425        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 426        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 427        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
 428        fputc('\n', rpt);
 429
 430        fputs("fatal: ", rpt);
 431        fputs(err, rpt);
 432        fputc('\n', rpt);
 433
 434        fputc('\n', rpt);
 435        fputs("Most Recent Commands Before Crash\n", rpt);
 436        fputs("---------------------------------\n", rpt);
 437        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 438                if (rc->next == &cmd_hist)
 439                        fputs("* ", rpt);
 440                else
 441                        fputs("  ", rpt);
 442                fputs(rc->buf, rpt);
 443                fputc('\n', rpt);
 444        }
 445
 446        fputc('\n', rpt);
 447        fputs("Active Branch LRU\n", rpt);
 448        fputs("-----------------\n", rpt);
 449        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 450                cur_active_branches,
 451                max_active_branches);
 452        fputc('\n', rpt);
 453        fputs("  pos  clock name\n", rpt);
 454        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 455        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 456                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 457                        ++lu, b->last_commit, b->name);
 458
 459        fputc('\n', rpt);
 460        fputs("Inactive Branches\n", rpt);
 461        fputs("-----------------\n", rpt);
 462        for (lu = 0; lu < branch_table_sz; lu++) {
 463                for (b = branch_table[lu]; b; b = b->table_next_branch)
 464                        write_branch_report(rpt, b);
 465        }
 466
 467        if (first_tag) {
 468                struct tag *tg;
 469                fputc('\n', rpt);
 470                fputs("Annotated Tags\n", rpt);
 471                fputs("--------------\n", rpt);
 472                for (tg = first_tag; tg; tg = tg->next_tag) {
 473                        fputs(oid_to_hex(&tg->oid), rpt);
 474                        fputc(' ', rpt);
 475                        fputs(tg->name, rpt);
 476                        fputc('\n', rpt);
 477                }
 478        }
 479
 480        fputc('\n', rpt);
 481        fputs("Marks\n", rpt);
 482        fputs("-----\n", rpt);
 483        if (export_marks_file)
 484                fprintf(rpt, "  exported to %s\n", export_marks_file);
 485        else
 486                dump_marks_helper(rpt, 0, marks);
 487
 488        fputc('\n', rpt);
 489        fputs("-------------------\n", rpt);
 490        fputs("END OF CRASH REPORT\n", rpt);
 491        fclose(rpt);
 492        free(loc);
 493}
 494
 495static void end_packfile(void);
 496static void unkeep_all_packs(void);
 497static void dump_marks(void);
 498
 499static NORETURN void die_nicely(const char *err, va_list params)
 500{
 501        static int zombie;
 502        char message[2 * PATH_MAX];
 503
 504        vsnprintf(message, sizeof(message), err, params);
 505        fputs("fatal: ", stderr);
 506        fputs(message, stderr);
 507        fputc('\n', stderr);
 508
 509        if (!zombie) {
 510                zombie = 1;
 511                write_crash_report(message);
 512                end_packfile();
 513                unkeep_all_packs();
 514                dump_marks();
 515        }
 516        exit(128);
 517}
 518
 519#ifndef SIGUSR1 /* Windows, for example */
 520
 521static void set_checkpoint_signal(void)
 522{
 523}
 524
 525#else
 526
 527static void checkpoint_signal(int signo)
 528{
 529        checkpoint_requested = 1;
 530}
 531
 532static void set_checkpoint_signal(void)
 533{
 534        struct sigaction sa;
 535
 536        memset(&sa, 0, sizeof(sa));
 537        sa.sa_handler = checkpoint_signal;
 538        sigemptyset(&sa.sa_mask);
 539        sa.sa_flags = SA_RESTART;
 540        sigaction(SIGUSR1, &sa, NULL);
 541}
 542
 543#endif
 544
 545static void alloc_objects(unsigned int cnt)
 546{
 547        struct object_entry_pool *b;
 548
 549        b = xmalloc(sizeof(struct object_entry_pool)
 550                + cnt * sizeof(struct object_entry));
 551        b->next_pool = blocks;
 552        b->next_free = b->entries;
 553        b->end = b->entries + cnt;
 554        blocks = b;
 555        alloc_count += cnt;
 556}
 557
 558static struct object_entry *new_object(struct object_id *oid)
 559{
 560        struct object_entry *e;
 561
 562        if (blocks->next_free == blocks->end)
 563                alloc_objects(object_entry_alloc);
 564
 565        e = blocks->next_free++;
 566        oidcpy(&e->idx.oid, oid);
 567        return e;
 568}
 569
 570static struct object_entry *find_object(struct object_id *oid)
 571{
 572        unsigned int h = oid->hash[0] << 8 | oid->hash[1];
 573        struct object_entry *e;
 574        for (e = object_table[h]; e; e = e->next)
 575                if (!oidcmp(oid, &e->idx.oid))
 576                        return e;
 577        return NULL;
 578}
 579
 580static struct object_entry *insert_object(struct object_id *oid)
 581{
 582        unsigned int h = oid->hash[0] << 8 | oid->hash[1];
 583        struct object_entry *e = object_table[h];
 584
 585        while (e) {
 586                if (!oidcmp(oid, &e->idx.oid))
 587                        return e;
 588                e = e->next;
 589        }
 590
 591        e = new_object(oid);
 592        e->next = object_table[h];
 593        e->idx.offset = 0;
 594        object_table[h] = e;
 595        return e;
 596}
 597
 598static void invalidate_pack_id(unsigned int id)
 599{
 600        unsigned int h;
 601        unsigned long lu;
 602        struct tag *t;
 603
 604        for (h = 0; h < ARRAY_SIZE(object_table); h++) {
 605                struct object_entry *e;
 606
 607                for (e = object_table[h]; e; e = e->next)
 608                        if (e->pack_id == id)
 609                                e->pack_id = MAX_PACK_ID;
 610        }
 611
 612        for (lu = 0; lu < branch_table_sz; lu++) {
 613                struct branch *b;
 614
 615                for (b = branch_table[lu]; b; b = b->table_next_branch)
 616                        if (b->pack_id == id)
 617                                b->pack_id = MAX_PACK_ID;
 618        }
 619
 620        for (t = first_tag; t; t = t->next_tag)
 621                if (t->pack_id == id)
 622                        t->pack_id = MAX_PACK_ID;
 623}
 624
 625static unsigned int hc_str(const char *s, size_t len)
 626{
 627        unsigned int r = 0;
 628        while (len-- > 0)
 629                r = r * 31 + *s++;
 630        return r;
 631}
 632
 633static char *pool_strdup(const char *s)
 634{
 635        size_t len = strlen(s) + 1;
 636        char *r = mem_pool_alloc(&fi_mem_pool, len);
 637        memcpy(r, s, len);
 638        return r;
 639}
 640
 641static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 642{
 643        struct mark_set *s = marks;
 644        while ((idnum >> s->shift) >= 1024) {
 645                s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
 646                s->shift = marks->shift + 10;
 647                s->data.sets[0] = marks;
 648                marks = s;
 649        }
 650        while (s->shift) {
 651                uintmax_t i = idnum >> s->shift;
 652                idnum -= i << s->shift;
 653                if (!s->data.sets[i]) {
 654                        s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
 655                        s->data.sets[i]->shift = s->shift - 10;
 656                }
 657                s = s->data.sets[i];
 658        }
 659        if (!s->data.marked[idnum])
 660                marks_set_count++;
 661        s->data.marked[idnum] = oe;
 662}
 663
 664static struct object_entry *find_mark(uintmax_t idnum)
 665{
 666        uintmax_t orig_idnum = idnum;
 667        struct mark_set *s = marks;
 668        struct object_entry *oe = NULL;
 669        if ((idnum >> s->shift) < 1024) {
 670                while (s && s->shift) {
 671                        uintmax_t i = idnum >> s->shift;
 672                        idnum -= i << s->shift;
 673                        s = s->data.sets[i];
 674                }
 675                if (s)
 676                        oe = s->data.marked[idnum];
 677        }
 678        if (!oe)
 679                die("mark :%" PRIuMAX " not declared", orig_idnum);
 680        return oe;
 681}
 682
 683static struct atom_str *to_atom(const char *s, unsigned short len)
 684{
 685        unsigned int hc = hc_str(s, len) % atom_table_sz;
 686        struct atom_str *c;
 687
 688        for (c = atom_table[hc]; c; c = c->next_atom)
 689                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 690                        return c;
 691
 692        c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
 693        c->str_len = len;
 694        memcpy(c->str_dat, s, len);
 695        c->str_dat[len] = 0;
 696        c->next_atom = atom_table[hc];
 697        atom_table[hc] = c;
 698        atom_cnt++;
 699        return c;
 700}
 701
 702static struct branch *lookup_branch(const char *name)
 703{
 704        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 705        struct branch *b;
 706
 707        for (b = branch_table[hc]; b; b = b->table_next_branch)
 708                if (!strcmp(name, b->name))
 709                        return b;
 710        return NULL;
 711}
 712
 713static struct branch *new_branch(const char *name)
 714{
 715        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 716        struct branch *b = lookup_branch(name);
 717
 718        if (b)
 719                die("Invalid attempt to create duplicate branch: %s", name);
 720        if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
 721                die("Branch name doesn't conform to GIT standards: %s", name);
 722
 723        b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
 724        b->name = pool_strdup(name);
 725        b->table_next_branch = branch_table[hc];
 726        b->branch_tree.versions[0].mode = S_IFDIR;
 727        b->branch_tree.versions[1].mode = S_IFDIR;
 728        b->num_notes = 0;
 729        b->active = 0;
 730        b->pack_id = MAX_PACK_ID;
 731        branch_table[hc] = b;
 732        branch_count++;
 733        return b;
 734}
 735
 736static unsigned int hc_entries(unsigned int cnt)
 737{
 738        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 739        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 740}
 741
 742static struct tree_content *new_tree_content(unsigned int cnt)
 743{
 744        struct avail_tree_content *f, *l = NULL;
 745        struct tree_content *t;
 746        unsigned int hc = hc_entries(cnt);
 747
 748        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 749                if (f->entry_capacity >= cnt)
 750                        break;
 751
 752        if (f) {
 753                if (l)
 754                        l->next_avail = f->next_avail;
 755                else
 756                        avail_tree_table[hc] = f->next_avail;
 757        } else {
 758                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 759                f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
 760                f->entry_capacity = cnt;
 761        }
 762
 763        t = (struct tree_content*)f;
 764        t->entry_count = 0;
 765        t->delta_depth = 0;
 766        return t;
 767}
 768
 769static void release_tree_entry(struct tree_entry *e);
 770static void release_tree_content(struct tree_content *t)
 771{
 772        struct avail_tree_content *f = (struct avail_tree_content*)t;
 773        unsigned int hc = hc_entries(f->entry_capacity);
 774        f->next_avail = avail_tree_table[hc];
 775        avail_tree_table[hc] = f;
 776}
 777
 778static void release_tree_content_recursive(struct tree_content *t)
 779{
 780        unsigned int i;
 781        for (i = 0; i < t->entry_count; i++)
 782                release_tree_entry(t->entries[i]);
 783        release_tree_content(t);
 784}
 785
 786static struct tree_content *grow_tree_content(
 787        struct tree_content *t,
 788        int amt)
 789{
 790        struct tree_content *r = new_tree_content(t->entry_count + amt);
 791        r->entry_count = t->entry_count;
 792        r->delta_depth = t->delta_depth;
 793        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 794        release_tree_content(t);
 795        return r;
 796}
 797
 798static struct tree_entry *new_tree_entry(void)
 799{
 800        struct tree_entry *e;
 801
 802        if (!avail_tree_entry) {
 803                unsigned int n = tree_entry_alloc;
 804                tree_entry_allocd += n * sizeof(struct tree_entry);
 805                ALLOC_ARRAY(e, n);
 806                avail_tree_entry = e;
 807                while (n-- > 1) {
 808                        *((void**)e) = e + 1;
 809                        e++;
 810                }
 811                *((void**)e) = NULL;
 812        }
 813
 814        e = avail_tree_entry;
 815        avail_tree_entry = *((void**)e);
 816        return e;
 817}
 818
 819static void release_tree_entry(struct tree_entry *e)
 820{
 821        if (e->tree)
 822                release_tree_content_recursive(e->tree);
 823        *((void**)e) = avail_tree_entry;
 824        avail_tree_entry = e;
 825}
 826
 827static struct tree_content *dup_tree_content(struct tree_content *s)
 828{
 829        struct tree_content *d;
 830        struct tree_entry *a, *b;
 831        unsigned int i;
 832
 833        if (!s)
 834                return NULL;
 835        d = new_tree_content(s->entry_count);
 836        for (i = 0; i < s->entry_count; i++) {
 837                a = s->entries[i];
 838                b = new_tree_entry();
 839                memcpy(b, a, sizeof(*a));
 840                if (a->tree && is_null_oid(&b->versions[1].oid))
 841                        b->tree = dup_tree_content(a->tree);
 842                else
 843                        b->tree = NULL;
 844                d->entries[i] = b;
 845        }
 846        d->entry_count = s->entry_count;
 847        d->delta_depth = s->delta_depth;
 848
 849        return d;
 850}
 851
 852static void start_packfile(void)
 853{
 854        struct strbuf tmp_file = STRBUF_INIT;
 855        struct packed_git *p;
 856        struct pack_header hdr;
 857        int pack_fd;
 858
 859        pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
 860        FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
 861        strbuf_release(&tmp_file);
 862
 863        p->pack_fd = pack_fd;
 864        p->do_not_close = 1;
 865        pack_file = hashfd(pack_fd, p->pack_name);
 866
 867        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 868        hdr.hdr_version = htonl(2);
 869        hdr.hdr_entries = 0;
 870        hashwrite(pack_file, &hdr, sizeof(hdr));
 871
 872        pack_data = p;
 873        pack_size = sizeof(hdr);
 874        object_count = 0;
 875
 876        REALLOC_ARRAY(all_packs, pack_id + 1);
 877        all_packs[pack_id] = p;
 878}
 879
 880static const char *create_index(void)
 881{
 882        const char *tmpfile;
 883        struct pack_idx_entry **idx, **c, **last;
 884        struct object_entry *e;
 885        struct object_entry_pool *o;
 886
 887        /* Build the table of object IDs. */
 888        ALLOC_ARRAY(idx, object_count);
 889        c = idx;
 890        for (o = blocks; o; o = o->next_pool)
 891                for (e = o->next_free; e-- != o->entries;)
 892                        if (pack_id == e->pack_id)
 893                                *c++ = &e->idx;
 894        last = idx + object_count;
 895        if (c != last)
 896                die("internal consistency error creating the index");
 897
 898        tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
 899        free(idx);
 900        return tmpfile;
 901}
 902
 903static char *keep_pack(const char *curr_index_name)
 904{
 905        static const char *keep_msg = "fast-import";
 906        struct strbuf name = STRBUF_INIT;
 907        int keep_fd;
 908
 909        odb_pack_name(&name, pack_data->sha1, "keep");
 910        keep_fd = odb_pack_keep(name.buf);
 911        if (keep_fd < 0)
 912                die_errno("cannot create keep file");
 913        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 914        if (close(keep_fd))
 915                die_errno("failed to write keep file");
 916
 917        odb_pack_name(&name, pack_data->sha1, "pack");
 918        if (finalize_object_file(pack_data->pack_name, name.buf))
 919                die("cannot store pack file");
 920
 921        odb_pack_name(&name, pack_data->sha1, "idx");
 922        if (finalize_object_file(curr_index_name, name.buf))
 923                die("cannot store index file");
 924        free((void *)curr_index_name);
 925        return strbuf_detach(&name, NULL);
 926}
 927
 928static void unkeep_all_packs(void)
 929{
 930        struct strbuf name = STRBUF_INIT;
 931        int k;
 932
 933        for (k = 0; k < pack_id; k++) {
 934                struct packed_git *p = all_packs[k];
 935                odb_pack_name(&name, p->sha1, "keep");
 936                unlink_or_warn(name.buf);
 937        }
 938        strbuf_release(&name);
 939}
 940
 941static int loosen_small_pack(const struct packed_git *p)
 942{
 943        struct child_process unpack = CHILD_PROCESS_INIT;
 944
 945        if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
 946                die_errno("Failed seeking to start of '%s'", p->pack_name);
 947
 948        unpack.in = p->pack_fd;
 949        unpack.git_cmd = 1;
 950        unpack.stdout_to_stderr = 1;
 951        argv_array_push(&unpack.args, "unpack-objects");
 952        if (!show_stats)
 953                argv_array_push(&unpack.args, "-q");
 954
 955        return run_command(&unpack);
 956}
 957
 958static void end_packfile(void)
 959{
 960        static int running;
 961
 962        if (running || !pack_data)
 963                return;
 964
 965        running = 1;
 966        clear_delta_base_cache();
 967        if (object_count) {
 968                struct packed_git *new_p;
 969                struct object_id cur_pack_oid;
 970                char *idx_name;
 971                int i;
 972                struct branch *b;
 973                struct tag *t;
 974
 975                close_pack_windows(pack_data);
 976                finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
 977                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 978                                    pack_data->pack_name, object_count,
 979                                    cur_pack_oid.hash, pack_size);
 980
 981                if (object_count <= unpack_limit) {
 982                        if (!loosen_small_pack(pack_data)) {
 983                                invalidate_pack_id(pack_id);
 984                                goto discard_pack;
 985                        }
 986                }
 987
 988                close(pack_data->pack_fd);
 989                idx_name = keep_pack(create_index());
 990
 991                /* Register the packfile with core git's machinery. */
 992                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 993                if (!new_p)
 994                        die("core git rejected index %s", idx_name);
 995                all_packs[pack_id] = new_p;
 996                install_packed_git(the_repository, new_p);
 997                free(idx_name);
 998
 999                /* Print the boundary */
1000                if (pack_edges) {
1001                        fprintf(pack_edges, "%s:", new_p->pack_name);
1002                        for (i = 0; i < branch_table_sz; i++) {
1003                                for (b = branch_table[i]; b; b = b->table_next_branch) {
1004                                        if (b->pack_id == pack_id)
1005                                                fprintf(pack_edges, " %s",
1006                                                        oid_to_hex(&b->oid));
1007                                }
1008                        }
1009                        for (t = first_tag; t; t = t->next_tag) {
1010                                if (t->pack_id == pack_id)
1011                                        fprintf(pack_edges, " %s",
1012                                                oid_to_hex(&t->oid));
1013                        }
1014                        fputc('\n', pack_edges);
1015                        fflush(pack_edges);
1016                }
1017
1018                pack_id++;
1019        }
1020        else {
1021discard_pack:
1022                close(pack_data->pack_fd);
1023                unlink_or_warn(pack_data->pack_name);
1024        }
1025        FREE_AND_NULL(pack_data);
1026        running = 0;
1027
1028        /* We can't carry a delta across packfiles. */
1029        strbuf_release(&last_blob.data);
1030        last_blob.offset = 0;
1031        last_blob.depth = 0;
1032}
1033
1034static void cycle_packfile(void)
1035{
1036        end_packfile();
1037        start_packfile();
1038}
1039
1040static int store_object(
1041        enum object_type type,
1042        struct strbuf *dat,
1043        struct last_object *last,
1044        struct object_id *oidout,
1045        uintmax_t mark)
1046{
1047        void *out, *delta;
1048        struct object_entry *e;
1049        unsigned char hdr[96];
1050        struct object_id oid;
1051        unsigned long hdrlen, deltalen;
1052        git_hash_ctx c;
1053        git_zstream s;
1054
1055        hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
1056                           type_name(type), (unsigned long)dat->len) + 1;
1057        the_hash_algo->init_fn(&c);
1058        the_hash_algo->update_fn(&c, hdr, hdrlen);
1059        the_hash_algo->update_fn(&c, dat->buf, dat->len);
1060        the_hash_algo->final_fn(oid.hash, &c);
1061        if (oidout)
1062                oidcpy(oidout, &oid);
1063
1064        e = insert_object(&oid);
1065        if (mark)
1066                insert_mark(mark, e);
1067        if (e->idx.offset) {
1068                duplicate_count_by_type[type]++;
1069                return 1;
1070        } else if (find_sha1_pack(oid.hash,
1071                                  get_packed_git(the_repository))) {
1072                e->type = type;
1073                e->pack_id = MAX_PACK_ID;
1074                e->idx.offset = 1; /* just not zero! */
1075                duplicate_count_by_type[type]++;
1076                return 1;
1077        }
1078
1079        if (last && last->data.buf && last->depth < max_depth
1080                && dat->len > the_hash_algo->rawsz) {
1081
1082                delta_count_attempts_by_type[type]++;
1083                delta = diff_delta(last->data.buf, last->data.len,
1084                        dat->buf, dat->len,
1085                        &deltalen, dat->len - the_hash_algo->rawsz);
1086        } else
1087                delta = NULL;
1088
1089        git_deflate_init(&s, pack_compression_level);
1090        if (delta) {
1091                s.next_in = delta;
1092                s.avail_in = deltalen;
1093        } else {
1094                s.next_in = (void *)dat->buf;
1095                s.avail_in = dat->len;
1096        }
1097        s.avail_out = git_deflate_bound(&s, s.avail_in);
1098        s.next_out = out = xmalloc(s.avail_out);
1099        while (git_deflate(&s, Z_FINISH) == Z_OK)
1100                ; /* nothing */
1101        git_deflate_end(&s);
1102
1103        /* Determine if we should auto-checkpoint. */
1104        if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1105                || (pack_size + 60 + s.total_out) < pack_size) {
1106
1107                /* This new object needs to *not* have the current pack_id. */
1108                e->pack_id = pack_id + 1;
1109                cycle_packfile();
1110
1111                /* We cannot carry a delta into the new pack. */
1112                if (delta) {
1113                        FREE_AND_NULL(delta);
1114
1115                        git_deflate_init(&s, pack_compression_level);
1116                        s.next_in = (void *)dat->buf;
1117                        s.avail_in = dat->len;
1118                        s.avail_out = git_deflate_bound(&s, s.avail_in);
1119                        s.next_out = out = xrealloc(out, s.avail_out);
1120                        while (git_deflate(&s, Z_FINISH) == Z_OK)
1121                                ; /* nothing */
1122                        git_deflate_end(&s);
1123                }
1124        }
1125
1126        e->type = type;
1127        e->pack_id = pack_id;
1128        e->idx.offset = pack_size;
1129        object_count++;
1130        object_count_by_type[type]++;
1131
1132        crc32_begin(pack_file);
1133
1134        if (delta) {
1135                off_t ofs = e->idx.offset - last->offset;
1136                unsigned pos = sizeof(hdr) - 1;
1137
1138                delta_count_by_type[type]++;
1139                e->depth = last->depth + 1;
1140
1141                hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1142                                                      OBJ_OFS_DELTA, deltalen);
1143                hashwrite(pack_file, hdr, hdrlen);
1144                pack_size += hdrlen;
1145
1146                hdr[pos] = ofs & 127;
1147                while (ofs >>= 7)
1148                        hdr[--pos] = 128 | (--ofs & 127);
1149                hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1150                pack_size += sizeof(hdr) - pos;
1151        } else {
1152                e->depth = 0;
1153                hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1154                                                      type, dat->len);
1155                hashwrite(pack_file, hdr, hdrlen);
1156                pack_size += hdrlen;
1157        }
1158
1159        hashwrite(pack_file, out, s.total_out);
1160        pack_size += s.total_out;
1161
1162        e->idx.crc32 = crc32_end(pack_file);
1163
1164        free(out);
1165        free(delta);
1166        if (last) {
1167                if (last->no_swap) {
1168                        last->data = *dat;
1169                } else {
1170                        strbuf_swap(&last->data, dat);
1171                }
1172                last->offset = e->idx.offset;
1173                last->depth = e->depth;
1174        }
1175        return 0;
1176}
1177
1178static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1179{
1180        if (hashfile_truncate(pack_file, checkpoint))
1181                die_errno("cannot truncate pack to skip duplicate");
1182        pack_size = checkpoint->offset;
1183}
1184
1185static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1186{
1187        size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1188        unsigned char *in_buf = xmalloc(in_sz);
1189        unsigned char *out_buf = xmalloc(out_sz);
1190        struct object_entry *e;
1191        struct object_id oid;
1192        unsigned long hdrlen;
1193        off_t offset;
1194        git_hash_ctx c;
1195        git_zstream s;
1196        struct hashfile_checkpoint checkpoint;
1197        int status = Z_OK;
1198
1199        /* Determine if we should auto-checkpoint. */
1200        if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1201                || (pack_size + 60 + len) < pack_size)
1202                cycle_packfile();
1203
1204        hashfile_checkpoint(pack_file, &checkpoint);
1205        offset = checkpoint.offset;
1206
1207        hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1208
1209        the_hash_algo->init_fn(&c);
1210        the_hash_algo->update_fn(&c, out_buf, hdrlen);
1211
1212        crc32_begin(pack_file);
1213
1214        git_deflate_init(&s, pack_compression_level);
1215
1216        hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1217
1218        s.next_out = out_buf + hdrlen;
1219        s.avail_out = out_sz - hdrlen;
1220
1221        while (status != Z_STREAM_END) {
1222                if (0 < len && !s.avail_in) {
1223                        size_t cnt = in_sz < len ? in_sz : (size_t)len;
1224                        size_t n = fread(in_buf, 1, cnt, stdin);
1225                        if (!n && feof(stdin))
1226                                die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1227
1228                        the_hash_algo->update_fn(&c, in_buf, n);
1229                        s.next_in = in_buf;
1230                        s.avail_in = n;
1231                        len -= n;
1232                }
1233
1234                status = git_deflate(&s, len ? 0 : Z_FINISH);
1235
1236                if (!s.avail_out || status == Z_STREAM_END) {
1237                        size_t n = s.next_out - out_buf;
1238                        hashwrite(pack_file, out_buf, n);
1239                        pack_size += n;
1240                        s.next_out = out_buf;
1241                        s.avail_out = out_sz;
1242                }
1243
1244                switch (status) {
1245                case Z_OK:
1246                case Z_BUF_ERROR:
1247                case Z_STREAM_END:
1248                        continue;
1249                default:
1250                        die("unexpected deflate failure: %d", status);
1251                }
1252        }
1253        git_deflate_end(&s);
1254        the_hash_algo->final_fn(oid.hash, &c);
1255
1256        if (oidout)
1257                oidcpy(oidout, &oid);
1258
1259        e = insert_object(&oid);
1260
1261        if (mark)
1262                insert_mark(mark, e);
1263
1264        if (e->idx.offset) {
1265                duplicate_count_by_type[OBJ_BLOB]++;
1266                truncate_pack(&checkpoint);
1267
1268        } else if (find_sha1_pack(oid.hash,
1269                                  get_packed_git(the_repository))) {
1270                e->type = OBJ_BLOB;
1271                e->pack_id = MAX_PACK_ID;
1272                e->idx.offset = 1; /* just not zero! */
1273                duplicate_count_by_type[OBJ_BLOB]++;
1274                truncate_pack(&checkpoint);
1275
1276        } else {
1277                e->depth = 0;
1278                e->type = OBJ_BLOB;
1279                e->pack_id = pack_id;
1280                e->idx.offset = offset;
1281                e->idx.crc32 = crc32_end(pack_file);
1282                object_count++;
1283                object_count_by_type[OBJ_BLOB]++;
1284        }
1285
1286        free(in_buf);
1287        free(out_buf);
1288}
1289
1290/* All calls must be guarded by find_object() or find_mark() to
1291 * ensure the 'struct object_entry' passed was written by this
1292 * process instance.  We unpack the entry by the offset, avoiding
1293 * the need for the corresponding .idx file.  This unpacking rule
1294 * works because we only use OBJ_REF_DELTA within the packfiles
1295 * created by fast-import.
1296 *
1297 * oe must not be NULL.  Such an oe usually comes from giving
1298 * an unknown SHA-1 to find_object() or an undefined mark to
1299 * find_mark().  Callers must test for this condition and use
1300 * the standard read_sha1_file() when it happens.
1301 *
1302 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1303 * find_mark(), where the mark was reloaded from an existing marks
1304 * file and is referencing an object that this fast-import process
1305 * instance did not write out to a packfile.  Callers must test for
1306 * this condition and use read_sha1_file() instead.
1307 */
1308static void *gfi_unpack_entry(
1309        struct object_entry *oe,
1310        unsigned long *sizep)
1311{
1312        enum object_type type;
1313        struct packed_git *p = all_packs[oe->pack_id];
1314        if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1315                /* The object is stored in the packfile we are writing to
1316                 * and we have modified it since the last time we scanned
1317                 * back to read a previously written object.  If an old
1318                 * window covered [p->pack_size, p->pack_size + rawsz) its
1319                 * data is stale and is not valid.  Closing all windows
1320                 * and updating the packfile length ensures we can read
1321                 * the newly written data.
1322                 */
1323                close_pack_windows(p);
1324                hashflush(pack_file);
1325
1326                /* We have to offer rawsz bytes additional on the end of
1327                 * the packfile as the core unpacker code assumes the
1328                 * footer is present at the file end and must promise
1329                 * at least rawsz bytes within any window it maps.  But
1330                 * we don't actually create the footer here.
1331                 */
1332                p->pack_size = pack_size + the_hash_algo->rawsz;
1333        }
1334        return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1335}
1336
1337static const char *get_mode(const char *str, uint16_t *modep)
1338{
1339        unsigned char c;
1340        uint16_t mode = 0;
1341
1342        while ((c = *str++) != ' ') {
1343                if (c < '0' || c > '7')
1344                        return NULL;
1345                mode = (mode << 3) + (c - '0');
1346        }
1347        *modep = mode;
1348        return str;
1349}
1350
1351static void load_tree(struct tree_entry *root)
1352{
1353        struct object_id *oid = &root->versions[1].oid;
1354        struct object_entry *myoe;
1355        struct tree_content *t;
1356        unsigned long size;
1357        char *buf;
1358        const char *c;
1359
1360        root->tree = t = new_tree_content(8);
1361        if (is_null_oid(oid))
1362                return;
1363
1364        myoe = find_object(oid);
1365        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1366                if (myoe->type != OBJ_TREE)
1367                        die("Not a tree: %s", oid_to_hex(oid));
1368                t->delta_depth = myoe->depth;
1369                buf = gfi_unpack_entry(myoe, &size);
1370                if (!buf)
1371                        die("Can't load tree %s", oid_to_hex(oid));
1372        } else {
1373                enum object_type type;
1374                buf = read_object_file(oid, &type, &size);
1375                if (!buf || type != OBJ_TREE)
1376                        die("Can't load tree %s", oid_to_hex(oid));
1377        }
1378
1379        c = buf;
1380        while (c != (buf + size)) {
1381                struct tree_entry *e = new_tree_entry();
1382
1383                if (t->entry_count == t->entry_capacity)
1384                        root->tree = t = grow_tree_content(t, t->entry_count);
1385                t->entries[t->entry_count++] = e;
1386
1387                e->tree = NULL;
1388                c = get_mode(c, &e->versions[1].mode);
1389                if (!c)
1390                        die("Corrupt mode in %s", oid_to_hex(oid));
1391                e->versions[0].mode = e->versions[1].mode;
1392                e->name = to_atom(c, strlen(c));
1393                c += e->name->str_len + 1;
1394                hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1395                hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1396                c += GIT_SHA1_RAWSZ;
1397        }
1398        free(buf);
1399}
1400
1401static int tecmp0 (const void *_a, const void *_b)
1402{
1403        struct tree_entry *a = *((struct tree_entry**)_a);
1404        struct tree_entry *b = *((struct tree_entry**)_b);
1405        return base_name_compare(
1406                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1407                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1408}
1409
1410static int tecmp1 (const void *_a, const void *_b)
1411{
1412        struct tree_entry *a = *((struct tree_entry**)_a);
1413        struct tree_entry *b = *((struct tree_entry**)_b);
1414        return base_name_compare(
1415                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1416                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1417}
1418
1419static void mktree(struct tree_content *t, int v, struct strbuf *b)
1420{
1421        size_t maxlen = 0;
1422        unsigned int i;
1423
1424        if (!v)
1425                QSORT(t->entries, t->entry_count, tecmp0);
1426        else
1427                QSORT(t->entries, t->entry_count, tecmp1);
1428
1429        for (i = 0; i < t->entry_count; i++) {
1430                if (t->entries[i]->versions[v].mode)
1431                        maxlen += t->entries[i]->name->str_len + 34;
1432        }
1433
1434        strbuf_reset(b);
1435        strbuf_grow(b, maxlen);
1436        for (i = 0; i < t->entry_count; i++) {
1437                struct tree_entry *e = t->entries[i];
1438                if (!e->versions[v].mode)
1439                        continue;
1440                strbuf_addf(b, "%o %s%c",
1441                        (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1442                        e->name->str_dat, '\0');
1443                strbuf_add(b, e->versions[v].oid.hash, GIT_SHA1_RAWSZ);
1444        }
1445}
1446
1447static void store_tree(struct tree_entry *root)
1448{
1449        struct tree_content *t;
1450        unsigned int i, j, del;
1451        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1452        struct object_entry *le = NULL;
1453
1454        if (!is_null_oid(&root->versions[1].oid))
1455                return;
1456
1457        if (!root->tree)
1458                load_tree(root);
1459        t = root->tree;
1460
1461        for (i = 0; i < t->entry_count; i++) {
1462                if (t->entries[i]->tree)
1463                        store_tree(t->entries[i]);
1464        }
1465
1466        if (!(root->versions[0].mode & NO_DELTA))
1467                le = find_object(&root->versions[0].oid);
1468        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1469                mktree(t, 0, &old_tree);
1470                lo.data = old_tree;
1471                lo.offset = le->idx.offset;
1472                lo.depth = t->delta_depth;
1473        }
1474
1475        mktree(t, 1, &new_tree);
1476        store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1477
1478        t->delta_depth = lo.depth;
1479        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1480                struct tree_entry *e = t->entries[i];
1481                if (e->versions[1].mode) {
1482                        e->versions[0].mode = e->versions[1].mode;
1483                        oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1484                        t->entries[j++] = e;
1485                } else {
1486                        release_tree_entry(e);
1487                        del++;
1488                }
1489        }
1490        t->entry_count -= del;
1491}
1492
1493static void tree_content_replace(
1494        struct tree_entry *root,
1495        const struct object_id *oid,
1496        const uint16_t mode,
1497        struct tree_content *newtree)
1498{
1499        if (!S_ISDIR(mode))
1500                die("Root cannot be a non-directory");
1501        oidclr(&root->versions[0].oid);
1502        oidcpy(&root->versions[1].oid, oid);
1503        if (root->tree)
1504                release_tree_content_recursive(root->tree);
1505        root->tree = newtree;
1506}
1507
1508static int tree_content_set(
1509        struct tree_entry *root,
1510        const char *p,
1511        const struct object_id *oid,
1512        const uint16_t mode,
1513        struct tree_content *subtree)
1514{
1515        struct tree_content *t;
1516        const char *slash1;
1517        unsigned int i, n;
1518        struct tree_entry *e;
1519
1520        slash1 = strchrnul(p, '/');
1521        n = slash1 - p;
1522        if (!n)
1523                die("Empty path component found in input");
1524        if (!*slash1 && !S_ISDIR(mode) && subtree)
1525                die("Non-directories cannot have subtrees");
1526
1527        if (!root->tree)
1528                load_tree(root);
1529        t = root->tree;
1530        for (i = 0; i < t->entry_count; i++) {
1531                e = t->entries[i];
1532                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1533                        if (!*slash1) {
1534                                if (!S_ISDIR(mode)
1535                                                && e->versions[1].mode == mode
1536                                                && !oidcmp(&e->versions[1].oid, oid))
1537                                        return 0;
1538                                e->versions[1].mode = mode;
1539                                oidcpy(&e->versions[1].oid, oid);
1540                                if (e->tree)
1541                                        release_tree_content_recursive(e->tree);
1542                                e->tree = subtree;
1543
1544                                /*
1545                                 * We need to leave e->versions[0].sha1 alone
1546                                 * to avoid modifying the preimage tree used
1547                                 * when writing out the parent directory.
1548                                 * But after replacing the subdir with a
1549                                 * completely different one, it's not a good
1550                                 * delta base any more, and besides, we've
1551                                 * thrown away the tree entries needed to
1552                                 * make a delta against it.
1553                                 *
1554                                 * So let's just explicitly disable deltas
1555                                 * for the subtree.
1556                                 */
1557                                if (S_ISDIR(e->versions[0].mode))
1558                                        e->versions[0].mode |= NO_DELTA;
1559
1560                                oidclr(&root->versions[1].oid);
1561                                return 1;
1562                        }
1563                        if (!S_ISDIR(e->versions[1].mode)) {
1564                                e->tree = new_tree_content(8);
1565                                e->versions[1].mode = S_IFDIR;
1566                        }
1567                        if (!e->tree)
1568                                load_tree(e);
1569                        if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1570                                oidclr(&root->versions[1].oid);
1571                                return 1;
1572                        }
1573                        return 0;
1574                }
1575        }
1576
1577        if (t->entry_count == t->entry_capacity)
1578                root->tree = t = grow_tree_content(t, t->entry_count);
1579        e = new_tree_entry();
1580        e->name = to_atom(p, n);
1581        e->versions[0].mode = 0;
1582        oidclr(&e->versions[0].oid);
1583        t->entries[t->entry_count++] = e;
1584        if (*slash1) {
1585                e->tree = new_tree_content(8);
1586                e->versions[1].mode = S_IFDIR;
1587                tree_content_set(e, slash1 + 1, oid, mode, subtree);
1588        } else {
1589                e->tree = subtree;
1590                e->versions[1].mode = mode;
1591                oidcpy(&e->versions[1].oid, oid);
1592        }
1593        oidclr(&root->versions[1].oid);
1594        return 1;
1595}
1596
1597static int tree_content_remove(
1598        struct tree_entry *root,
1599        const char *p,
1600        struct tree_entry *backup_leaf,
1601        int allow_root)
1602{
1603        struct tree_content *t;
1604        const char *slash1;
1605        unsigned int i, n;
1606        struct tree_entry *e;
1607
1608        slash1 = strchrnul(p, '/');
1609        n = slash1 - p;
1610
1611        if (!root->tree)
1612                load_tree(root);
1613
1614        if (!*p && allow_root) {
1615                e = root;
1616                goto del_entry;
1617        }
1618
1619        t = root->tree;
1620        for (i = 0; i < t->entry_count; i++) {
1621                e = t->entries[i];
1622                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1623                        if (*slash1 && !S_ISDIR(e->versions[1].mode))
1624                                /*
1625                                 * If p names a file in some subdirectory, and a
1626                                 * file or symlink matching the name of the
1627                                 * parent directory of p exists, then p cannot
1628                                 * exist and need not be deleted.
1629                                 */
1630                                return 1;
1631                        if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1632                                goto del_entry;
1633                        if (!e->tree)
1634                                load_tree(e);
1635                        if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1636                                for (n = 0; n < e->tree->entry_count; n++) {
1637                                        if (e->tree->entries[n]->versions[1].mode) {
1638                                                oidclr(&root->versions[1].oid);
1639                                                return 1;
1640                                        }
1641                                }
1642                                backup_leaf = NULL;
1643                                goto del_entry;
1644                        }
1645                        return 0;
1646                }
1647        }
1648        return 0;
1649
1650del_entry:
1651        if (backup_leaf)
1652                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1653        else if (e->tree)
1654                release_tree_content_recursive(e->tree);
1655        e->tree = NULL;
1656        e->versions[1].mode = 0;
1657        oidclr(&e->versions[1].oid);
1658        oidclr(&root->versions[1].oid);
1659        return 1;
1660}
1661
1662static int tree_content_get(
1663        struct tree_entry *root,
1664        const char *p,
1665        struct tree_entry *leaf,
1666        int allow_root)
1667{
1668        struct tree_content *t;
1669        const char *slash1;
1670        unsigned int i, n;
1671        struct tree_entry *e;
1672
1673        slash1 = strchrnul(p, '/');
1674        n = slash1 - p;
1675        if (!n && !allow_root)
1676                die("Empty path component found in input");
1677
1678        if (!root->tree)
1679                load_tree(root);
1680
1681        if (!n) {
1682                e = root;
1683                goto found_entry;
1684        }
1685
1686        t = root->tree;
1687        for (i = 0; i < t->entry_count; i++) {
1688                e = t->entries[i];
1689                if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1690                        if (!*slash1)
1691                                goto found_entry;
1692                        if (!S_ISDIR(e->versions[1].mode))
1693                                return 0;
1694                        if (!e->tree)
1695                                load_tree(e);
1696                        return tree_content_get(e, slash1 + 1, leaf, 0);
1697                }
1698        }
1699        return 0;
1700
1701found_entry:
1702        memcpy(leaf, e, sizeof(*leaf));
1703        if (e->tree && is_null_oid(&e->versions[1].oid))
1704                leaf->tree = dup_tree_content(e->tree);
1705        else
1706                leaf->tree = NULL;
1707        return 1;
1708}
1709
1710static int update_branch(struct branch *b)
1711{
1712        static const char *msg = "fast-import";
1713        struct ref_transaction *transaction;
1714        struct object_id old_oid;
1715        struct strbuf err = STRBUF_INIT;
1716
1717        if (is_null_oid(&b->oid)) {
1718                if (b->delete)
1719                        delete_ref(NULL, b->name, NULL, 0);
1720                return 0;
1721        }
1722        if (read_ref(b->name, &old_oid))
1723                oidclr(&old_oid);
1724        if (!force_update && !is_null_oid(&old_oid)) {
1725                struct commit *old_cmit, *new_cmit;
1726
1727                old_cmit = lookup_commit_reference_gently(&old_oid, 0);
1728                new_cmit = lookup_commit_reference_gently(&b->oid, 0);
1729                if (!old_cmit || !new_cmit)
1730                        return error("Branch %s is missing commits.", b->name);
1731
1732                if (!in_merge_bases(old_cmit, new_cmit)) {
1733                        warning("Not updating %s"
1734                                " (new tip %s does not contain %s)",
1735                                b->name, oid_to_hex(&b->oid),
1736                                oid_to_hex(&old_oid));
1737                        return -1;
1738                }
1739        }
1740        transaction = ref_transaction_begin(&err);
1741        if (!transaction ||
1742            ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1743                                   0, msg, &err) ||
1744            ref_transaction_commit(transaction, &err)) {
1745                ref_transaction_free(transaction);
1746                error("%s", err.buf);
1747                strbuf_release(&err);
1748                return -1;
1749        }
1750        ref_transaction_free(transaction);
1751        strbuf_release(&err);
1752        return 0;
1753}
1754
1755static void dump_branches(void)
1756{
1757        unsigned int i;
1758        struct branch *b;
1759
1760        for (i = 0; i < branch_table_sz; i++) {
1761                for (b = branch_table[i]; b; b = b->table_next_branch)
1762                        failure |= update_branch(b);
1763        }
1764}
1765
1766static void dump_tags(void)
1767{
1768        static const char *msg = "fast-import";
1769        struct tag *t;
1770        struct strbuf ref_name = STRBUF_INIT;
1771        struct strbuf err = STRBUF_INIT;
1772        struct ref_transaction *transaction;
1773
1774        transaction = ref_transaction_begin(&err);
1775        if (!transaction) {
1776                failure |= error("%s", err.buf);
1777                goto cleanup;
1778        }
1779        for (t = first_tag; t; t = t->next_tag) {
1780                strbuf_reset(&ref_name);
1781                strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1782
1783                if (ref_transaction_update(transaction, ref_name.buf,
1784                                           &t->oid, NULL, 0, msg, &err)) {
1785                        failure |= error("%s", err.buf);
1786                        goto cleanup;
1787                }
1788        }
1789        if (ref_transaction_commit(transaction, &err))
1790                failure |= error("%s", err.buf);
1791
1792 cleanup:
1793        ref_transaction_free(transaction);
1794        strbuf_release(&ref_name);
1795        strbuf_release(&err);
1796}
1797
1798static void dump_marks_helper(FILE *f,
1799        uintmax_t base,
1800        struct mark_set *m)
1801{
1802        uintmax_t k;
1803        if (m->shift) {
1804                for (k = 0; k < 1024; k++) {
1805                        if (m->data.sets[k])
1806                                dump_marks_helper(f, base + (k << m->shift),
1807                                        m->data.sets[k]);
1808                }
1809        } else {
1810                for (k = 0; k < 1024; k++) {
1811                        if (m->data.marked[k])
1812                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1813                                        oid_to_hex(&m->data.marked[k]->idx.oid));
1814                }
1815        }
1816}
1817
1818static void dump_marks(void)
1819{
1820        struct lock_file mark_lock = LOCK_INIT;
1821        FILE *f;
1822
1823        if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1824                return;
1825
1826        if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1827                failure |= error_errno("Unable to write marks file %s",
1828                                       export_marks_file);
1829                return;
1830        }
1831
1832        f = fdopen_lock_file(&mark_lock, "w");
1833        if (!f) {
1834                int saved_errno = errno;
1835                rollback_lock_file(&mark_lock);
1836                failure |= error("Unable to write marks file %s: %s",
1837                        export_marks_file, strerror(saved_errno));
1838                return;
1839        }
1840
1841        dump_marks_helper(f, 0, marks);
1842        if (commit_lock_file(&mark_lock)) {
1843                failure |= error_errno("Unable to write file %s",
1844                                       export_marks_file);
1845                return;
1846        }
1847}
1848
1849static void read_marks(void)
1850{
1851        char line[512];
1852        FILE *f = fopen(import_marks_file, "r");
1853        if (f)
1854                ;
1855        else if (import_marks_file_ignore_missing && errno == ENOENT)
1856                goto done; /* Marks file does not exist */
1857        else
1858                die_errno("cannot read '%s'", import_marks_file);
1859        while (fgets(line, sizeof(line), f)) {
1860                uintmax_t mark;
1861                char *end;
1862                struct object_id oid;
1863                struct object_entry *e;
1864
1865                end = strchr(line, '\n');
1866                if (line[0] != ':' || !end)
1867                        die("corrupt mark line: %s", line);
1868                *end = 0;
1869                mark = strtoumax(line + 1, &end, 10);
1870                if (!mark || end == line + 1
1871                        || *end != ' ' || get_oid_hex(end + 1, &oid))
1872                        die("corrupt mark line: %s", line);
1873                e = find_object(&oid);
1874                if (!e) {
1875                        enum object_type type = oid_object_info(the_repository,
1876                                                                &oid, NULL);
1877                        if (type < 0)
1878                                die("object not found: %s", oid_to_hex(&oid));
1879                        e = insert_object(&oid);
1880                        e->type = type;
1881                        e->pack_id = MAX_PACK_ID;
1882                        e->idx.offset = 1; /* just not zero! */
1883                }
1884                insert_mark(mark, e);
1885        }
1886        fclose(f);
1887done:
1888        import_marks_file_done = 1;
1889}
1890
1891
1892static int read_next_command(void)
1893{
1894        static int stdin_eof = 0;
1895
1896        if (stdin_eof) {
1897                unread_command_buf = 0;
1898                return EOF;
1899        }
1900
1901        for (;;) {
1902                const char *p;
1903
1904                if (unread_command_buf) {
1905                        unread_command_buf = 0;
1906                } else {
1907                        struct recent_command *rc;
1908
1909                        strbuf_detach(&command_buf, NULL);
1910                        stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1911                        if (stdin_eof)
1912                                return EOF;
1913
1914                        if (!seen_data_command
1915                                && !starts_with(command_buf.buf, "feature ")
1916                                && !starts_with(command_buf.buf, "option ")) {
1917                                parse_argv();
1918                        }
1919
1920                        rc = rc_free;
1921                        if (rc)
1922                                rc_free = rc->next;
1923                        else {
1924                                rc = cmd_hist.next;
1925                                cmd_hist.next = rc->next;
1926                                cmd_hist.next->prev = &cmd_hist;
1927                                free(rc->buf);
1928                        }
1929
1930                        rc->buf = command_buf.buf;
1931                        rc->prev = cmd_tail;
1932                        rc->next = cmd_hist.prev;
1933                        rc->prev->next = rc;
1934                        cmd_tail = rc;
1935                }
1936                if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1937                        parse_get_mark(p);
1938                        continue;
1939                }
1940                if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1941                        parse_cat_blob(p);
1942                        continue;
1943                }
1944                if (command_buf.buf[0] == '#')
1945                        continue;
1946                return 0;
1947        }
1948}
1949
1950static void skip_optional_lf(void)
1951{
1952        int term_char = fgetc(stdin);
1953        if (term_char != '\n' && term_char != EOF)
1954                ungetc(term_char, stdin);
1955}
1956
1957static void parse_mark(void)
1958{
1959        const char *v;
1960        if (skip_prefix(command_buf.buf, "mark :", &v)) {
1961                next_mark = strtoumax(v, NULL, 10);
1962                read_next_command();
1963        }
1964        else
1965                next_mark = 0;
1966}
1967
1968static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1969{
1970        const char *data;
1971        strbuf_reset(sb);
1972
1973        if (!skip_prefix(command_buf.buf, "data ", &data))
1974                die("Expected 'data n' command, found: %s", command_buf.buf);
1975
1976        if (skip_prefix(data, "<<", &data)) {
1977                char *term = xstrdup(data);
1978                size_t term_len = command_buf.len - (data - command_buf.buf);
1979
1980                strbuf_detach(&command_buf, NULL);
1981                for (;;) {
1982                        if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1983                                die("EOF in data (terminator '%s' not found)", term);
1984                        if (term_len == command_buf.len
1985                                && !strcmp(term, command_buf.buf))
1986                                break;
1987                        strbuf_addbuf(sb, &command_buf);
1988                        strbuf_addch(sb, '\n');
1989                }
1990                free(term);
1991        }
1992        else {
1993                uintmax_t len = strtoumax(data, NULL, 10);
1994                size_t n = 0, length = (size_t)len;
1995
1996                if (limit && limit < len) {
1997                        *len_res = len;
1998                        return 0;
1999                }
2000                if (length < len)
2001                        die("data is too large to use in this context");
2002
2003                while (n < length) {
2004                        size_t s = strbuf_fread(sb, length - n, stdin);
2005                        if (!s && feof(stdin))
2006                                die("EOF in data (%lu bytes remaining)",
2007                                        (unsigned long)(length - n));
2008                        n += s;
2009                }
2010        }
2011
2012        skip_optional_lf();
2013        return 1;
2014}
2015
2016static int validate_raw_date(const char *src, struct strbuf *result)
2017{
2018        const char *orig_src = src;
2019        char *endp;
2020        unsigned long num;
2021
2022        errno = 0;
2023
2024        num = strtoul(src, &endp, 10);
2025        /* NEEDSWORK: perhaps check for reasonable values? */
2026        if (errno || endp == src || *endp != ' ')
2027                return -1;
2028
2029        src = endp + 1;
2030        if (*src != '-' && *src != '+')
2031                return -1;
2032
2033        num = strtoul(src + 1, &endp, 10);
2034        if (errno || endp == src + 1 || *endp || 1400 < num)
2035                return -1;
2036
2037        strbuf_addstr(result, orig_src);
2038        return 0;
2039}
2040
2041static char *parse_ident(const char *buf)
2042{
2043        const char *ltgt;
2044        size_t name_len;
2045        struct strbuf ident = STRBUF_INIT;
2046
2047        /* ensure there is a space delimiter even if there is no name */
2048        if (*buf == '<')
2049                --buf;
2050
2051        ltgt = buf + strcspn(buf, "<>");
2052        if (*ltgt != '<')
2053                die("Missing < in ident string: %s", buf);
2054        if (ltgt != buf && ltgt[-1] != ' ')
2055                die("Missing space before < in ident string: %s", buf);
2056        ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
2057        if (*ltgt != '>')
2058                die("Missing > in ident string: %s", buf);
2059        ltgt++;
2060        if (*ltgt != ' ')
2061                die("Missing space after > in ident string: %s", buf);
2062        ltgt++;
2063        name_len = ltgt - buf;
2064        strbuf_add(&ident, buf, name_len);
2065
2066        switch (whenspec) {
2067        case WHENSPEC_RAW:
2068                if (validate_raw_date(ltgt, &ident) < 0)
2069                        die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
2070                break;
2071        case WHENSPEC_RFC2822:
2072                if (parse_date(ltgt, &ident) < 0)
2073                        die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
2074                break;
2075        case WHENSPEC_NOW:
2076                if (strcmp("now", ltgt))
2077                        die("Date in ident must be 'now': %s", buf);
2078                datestamp(&ident);
2079                break;
2080        }
2081
2082        return strbuf_detach(&ident, NULL);
2083}
2084
2085static void parse_and_store_blob(
2086        struct last_object *last,
2087        struct object_id *oidout,
2088        uintmax_t mark)
2089{
2090        static struct strbuf buf = STRBUF_INIT;
2091        uintmax_t len;
2092
2093        if (parse_data(&buf, big_file_threshold, &len))
2094                store_object(OBJ_BLOB, &buf, last, oidout, mark);
2095        else {
2096                if (last) {
2097                        strbuf_release(&last->data);
2098                        last->offset = 0;
2099                        last->depth = 0;
2100                }
2101                stream_blob(len, oidout, mark);
2102                skip_optional_lf();
2103        }
2104}
2105
2106static void parse_new_blob(void)
2107{
2108        read_next_command();
2109        parse_mark();
2110        parse_and_store_blob(&last_blob, NULL, next_mark);
2111}
2112
2113static void unload_one_branch(void)
2114{
2115        while (cur_active_branches
2116                && cur_active_branches >= max_active_branches) {
2117                uintmax_t min_commit = ULONG_MAX;
2118                struct branch *e, *l = NULL, *p = NULL;
2119
2120                for (e = active_branches; e; e = e->active_next_branch) {
2121                        if (e->last_commit < min_commit) {
2122                                p = l;
2123                                min_commit = e->last_commit;
2124                        }
2125                        l = e;
2126                }
2127
2128                if (p) {
2129                        e = p->active_next_branch;
2130                        p->active_next_branch = e->active_next_branch;
2131                } else {
2132                        e = active_branches;
2133                        active_branches = e->active_next_branch;
2134                }
2135                e->active = 0;
2136                e->active_next_branch = NULL;
2137                if (e->branch_tree.tree) {
2138                        release_tree_content_recursive(e->branch_tree.tree);
2139                        e->branch_tree.tree = NULL;
2140                }
2141                cur_active_branches--;
2142        }
2143}
2144
2145static void load_branch(struct branch *b)
2146{
2147        load_tree(&b->branch_tree);
2148        if (!b->active) {
2149                b->active = 1;
2150                b->active_next_branch = active_branches;
2151                active_branches = b;
2152                cur_active_branches++;
2153                branch_load_count++;
2154        }
2155}
2156
2157static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2158{
2159        unsigned char fanout = 0;
2160        while ((num_notes >>= 8))
2161                fanout++;
2162        return fanout;
2163}
2164
2165static void construct_path_with_fanout(const char *hex_sha1,
2166                unsigned char fanout, char *path)
2167{
2168        unsigned int i = 0, j = 0;
2169        if (fanout >= the_hash_algo->rawsz)
2170                die("Too large fanout (%u)", fanout);
2171        while (fanout) {
2172                path[i++] = hex_sha1[j++];
2173                path[i++] = hex_sha1[j++];
2174                path[i++] = '/';
2175                fanout--;
2176        }
2177        memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2178        path[i + the_hash_algo->hexsz - j] = '\0';
2179}
2180
2181static uintmax_t do_change_note_fanout(
2182                struct tree_entry *orig_root, struct tree_entry *root,
2183                char *hex_oid, unsigned int hex_oid_len,
2184                char *fullpath, unsigned int fullpath_len,
2185                unsigned char fanout)
2186{
2187        struct tree_content *t;
2188        struct tree_entry *e, leaf;
2189        unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2190        uintmax_t num_notes = 0;
2191        struct object_id oid;
2192        char realpath[60];
2193
2194        if (!root->tree)
2195                load_tree(root);
2196        t = root->tree;
2197
2198        for (i = 0; t && i < t->entry_count; i++) {
2199                e = t->entries[i];
2200                tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2201                tmp_fullpath_len = fullpath_len;
2202
2203                /*
2204                 * We're interested in EITHER existing note entries (entries
2205                 * with exactly 40 hex chars in path, not including directory
2206                 * separators), OR directory entries that may contain note
2207                 * entries (with < 40 hex chars in path).
2208                 * Also, each path component in a note entry must be a multiple
2209                 * of 2 chars.
2210                 */
2211                if (!e->versions[1].mode ||
2212                    tmp_hex_oid_len > GIT_SHA1_HEXSZ ||
2213                    e->name->str_len % 2)
2214                        continue;
2215
2216                /* This _may_ be a note entry, or a subdir containing notes */
2217                memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2218                       e->name->str_len);
2219                if (tmp_fullpath_len)
2220                        fullpath[tmp_fullpath_len++] = '/';
2221                memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2222                       e->name->str_len);
2223                tmp_fullpath_len += e->name->str_len;
2224                fullpath[tmp_fullpath_len] = '\0';
2225
2226                if (tmp_hex_oid_len == GIT_SHA1_HEXSZ && !get_oid_hex(hex_oid, &oid)) {
2227                        /* This is a note entry */
2228                        if (fanout == 0xff) {
2229                                /* Counting mode, no rename */
2230                                num_notes++;
2231                                continue;
2232                        }
2233                        construct_path_with_fanout(hex_oid, fanout, realpath);
2234                        if (!strcmp(fullpath, realpath)) {
2235                                /* Note entry is in correct location */
2236                                num_notes++;
2237                                continue;
2238                        }
2239
2240                        /* Rename fullpath to realpath */
2241                        if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2242                                die("Failed to remove path %s", fullpath);
2243                        tree_content_set(orig_root, realpath,
2244                                &leaf.versions[1].oid,
2245                                leaf.versions[1].mode,
2246                                leaf.tree);
2247                } else if (S_ISDIR(e->versions[1].mode)) {
2248                        /* This is a subdir that may contain note entries */
2249                        num_notes += do_change_note_fanout(orig_root, e,
2250                                hex_oid, tmp_hex_oid_len,
2251                                fullpath, tmp_fullpath_len, fanout);
2252                }
2253
2254                /* The above may have reallocated the current tree_content */
2255                t = root->tree;
2256        }
2257        return num_notes;
2258}
2259
2260static uintmax_t change_note_fanout(struct tree_entry *root,
2261                unsigned char fanout)
2262{
2263        /*
2264         * The size of path is due to one slash between every two hex digits,
2265         * plus the terminating NUL.  Note that there is no slash at the end, so
2266         * the number of slashes is one less than half the number of hex
2267         * characters.
2268         */
2269        char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2270        return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2271}
2272
2273/*
2274 * Given a pointer into a string, parse a mark reference:
2275 *
2276 *   idnum ::= ':' bigint;
2277 *
2278 * Return the first character after the value in *endptr.
2279 *
2280 * Complain if the following character is not what is expected,
2281 * either a space or end of the string.
2282 */
2283static uintmax_t parse_mark_ref(const char *p, char **endptr)
2284{
2285        uintmax_t mark;
2286
2287        assert(*p == ':');
2288        p++;
2289        mark = strtoumax(p, endptr, 10);
2290        if (*endptr == p)
2291                die("No value after ':' in mark: %s", command_buf.buf);
2292        return mark;
2293}
2294
2295/*
2296 * Parse the mark reference, and complain if this is not the end of
2297 * the string.
2298 */
2299static uintmax_t parse_mark_ref_eol(const char *p)
2300{
2301        char *end;
2302        uintmax_t mark;
2303
2304        mark = parse_mark_ref(p, &end);
2305        if (*end != '\0')
2306                die("Garbage after mark: %s", command_buf.buf);
2307        return mark;
2308}
2309
2310/*
2311 * Parse the mark reference, demanding a trailing space.  Return a
2312 * pointer to the space.
2313 */
2314static uintmax_t parse_mark_ref_space(const char **p)
2315{
2316        uintmax_t mark;
2317        char *end;
2318
2319        mark = parse_mark_ref(*p, &end);
2320        if (*end++ != ' ')
2321                die("Missing space after mark: %s", command_buf.buf);
2322        *p = end;
2323        return mark;
2324}
2325
2326static void file_change_m(const char *p, struct branch *b)
2327{
2328        static struct strbuf uq = STRBUF_INIT;
2329        const char *endp;
2330        struct object_entry *oe;
2331        struct object_id oid;
2332        uint16_t mode, inline_data = 0;
2333
2334        p = get_mode(p, &mode);
2335        if (!p)
2336                die("Corrupt mode: %s", command_buf.buf);
2337        switch (mode) {
2338        case 0644:
2339        case 0755:
2340                mode |= S_IFREG;
2341        case S_IFREG | 0644:
2342        case S_IFREG | 0755:
2343        case S_IFLNK:
2344        case S_IFDIR:
2345        case S_IFGITLINK:
2346                /* ok */
2347                break;
2348        default:
2349                die("Corrupt mode: %s", command_buf.buf);
2350        }
2351
2352        if (*p == ':') {
2353                oe = find_mark(parse_mark_ref_space(&p));
2354                oidcpy(&oid, &oe->idx.oid);
2355        } else if (skip_prefix(p, "inline ", &p)) {
2356                inline_data = 1;
2357                oe = NULL; /* not used with inline_data, but makes gcc happy */
2358        } else {
2359                if (parse_oid_hex(p, &oid, &p))
2360                        die("Invalid dataref: %s", command_buf.buf);
2361                oe = find_object(&oid);
2362                if (*p++ != ' ')
2363                        die("Missing space after SHA1: %s", command_buf.buf);
2364        }
2365
2366        strbuf_reset(&uq);
2367        if (!unquote_c_style(&uq, p, &endp)) {
2368                if (*endp)
2369                        die("Garbage after path in: %s", command_buf.buf);
2370                p = uq.buf;
2371        }
2372
2373        /* Git does not track empty, non-toplevel directories. */
2374        if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2375                tree_content_remove(&b->branch_tree, p, NULL, 0);
2376                return;
2377        }
2378
2379        if (S_ISGITLINK(mode)) {
2380                if (inline_data)
2381                        die("Git links cannot be specified 'inline': %s",
2382                                command_buf.buf);
2383                else if (oe) {
2384                        if (oe->type != OBJ_COMMIT)
2385                                die("Not a commit (actually a %s): %s",
2386                                        type_name(oe->type), command_buf.buf);
2387                }
2388                /*
2389                 * Accept the sha1 without checking; it expected to be in
2390                 * another repository.
2391                 */
2392        } else if (inline_data) {
2393                if (S_ISDIR(mode))
2394                        die("Directories cannot be specified 'inline': %s",
2395                                command_buf.buf);
2396                if (p != uq.buf) {
2397                        strbuf_addstr(&uq, p);
2398                        p = uq.buf;
2399                }
2400                read_next_command();
2401                parse_and_store_blob(&last_blob, &oid, 0);
2402        } else {
2403                enum object_type expected = S_ISDIR(mode) ?
2404                                                OBJ_TREE: OBJ_BLOB;
2405                enum object_type type = oe ? oe->type :
2406                                        oid_object_info(the_repository, &oid,
2407                                                        NULL);
2408                if (type < 0)
2409                        die("%s not found: %s",
2410                                        S_ISDIR(mode) ?  "Tree" : "Blob",
2411                                        command_buf.buf);
2412                if (type != expected)
2413                        die("Not a %s (actually a %s): %s",
2414                                type_name(expected), type_name(type),
2415                                command_buf.buf);
2416        }
2417
2418        if (!*p) {
2419                tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2420                return;
2421        }
2422        tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2423}
2424
2425static void file_change_d(const char *p, struct branch *b)
2426{
2427        static struct strbuf uq = STRBUF_INIT;
2428        const char *endp;
2429
2430        strbuf_reset(&uq);
2431        if (!unquote_c_style(&uq, p, &endp)) {
2432                if (*endp)
2433                        die("Garbage after path in: %s", command_buf.buf);
2434                p = uq.buf;
2435        }
2436        tree_content_remove(&b->branch_tree, p, NULL, 1);
2437}
2438
2439static void file_change_cr(const char *s, struct branch *b, int rename)
2440{
2441        const char *d;
2442        static struct strbuf s_uq = STRBUF_INIT;
2443        static struct strbuf d_uq = STRBUF_INIT;
2444        const char *endp;
2445        struct tree_entry leaf;
2446
2447        strbuf_reset(&s_uq);
2448        if (!unquote_c_style(&s_uq, s, &endp)) {
2449                if (*endp != ' ')
2450                        die("Missing space after source: %s", command_buf.buf);
2451        } else {
2452                endp = strchr(s, ' ');
2453                if (!endp)
2454                        die("Missing space after source: %s", command_buf.buf);
2455                strbuf_add(&s_uq, s, endp - s);
2456        }
2457        s = s_uq.buf;
2458
2459        endp++;
2460        if (!*endp)
2461                die("Missing dest: %s", command_buf.buf);
2462
2463        d = endp;
2464        strbuf_reset(&d_uq);
2465        if (!unquote_c_style(&d_uq, d, &endp)) {
2466                if (*endp)
2467                        die("Garbage after dest in: %s", command_buf.buf);
2468                d = d_uq.buf;
2469        }
2470
2471        memset(&leaf, 0, sizeof(leaf));
2472        if (rename)
2473                tree_content_remove(&b->branch_tree, s, &leaf, 1);
2474        else
2475                tree_content_get(&b->branch_tree, s, &leaf, 1);
2476        if (!leaf.versions[1].mode)
2477                die("Path %s not in branch", s);
2478        if (!*d) {      /* C "path/to/subdir" "" */
2479                tree_content_replace(&b->branch_tree,
2480                        &leaf.versions[1].oid,
2481                        leaf.versions[1].mode,
2482                        leaf.tree);
2483                return;
2484        }
2485        tree_content_set(&b->branch_tree, d,
2486                &leaf.versions[1].oid,
2487                leaf.versions[1].mode,
2488                leaf.tree);
2489}
2490
2491static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2492{
2493        static struct strbuf uq = STRBUF_INIT;
2494        struct object_entry *oe;
2495        struct branch *s;
2496        struct object_id oid, commit_oid;
2497        char path[60];
2498        uint16_t inline_data = 0;
2499        unsigned char new_fanout;
2500
2501        /*
2502         * When loading a branch, we don't traverse its tree to count the real
2503         * number of notes (too expensive to do this for all non-note refs).
2504         * This means that recently loaded notes refs might incorrectly have
2505         * b->num_notes == 0, and consequently, old_fanout might be wrong.
2506         *
2507         * Fix this by traversing the tree and counting the number of notes
2508         * when b->num_notes == 0. If the notes tree is truly empty, the
2509         * calculation should not take long.
2510         */
2511        if (b->num_notes == 0 && *old_fanout == 0) {
2512                /* Invoke change_note_fanout() in "counting mode". */
2513                b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2514                *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2515        }
2516
2517        /* Now parse the notemodify command. */
2518        /* <dataref> or 'inline' */
2519        if (*p == ':') {
2520                oe = find_mark(parse_mark_ref_space(&p));
2521                oidcpy(&oid, &oe->idx.oid);
2522        } else if (skip_prefix(p, "inline ", &p)) {
2523                inline_data = 1;
2524                oe = NULL; /* not used with inline_data, but makes gcc happy */
2525        } else {
2526                if (parse_oid_hex(p, &oid, &p))
2527                        die("Invalid dataref: %s", command_buf.buf);
2528                oe = find_object(&oid);
2529                if (*p++ != ' ')
2530                        die("Missing space after SHA1: %s", command_buf.buf);
2531        }
2532
2533        /* <commit-ish> */
2534        s = lookup_branch(p);
2535        if (s) {
2536                if (is_null_oid(&s->oid))
2537                        die("Can't add a note on empty branch.");
2538                oidcpy(&commit_oid, &s->oid);
2539        } else if (*p == ':') {
2540                uintmax_t commit_mark = parse_mark_ref_eol(p);
2541                struct object_entry *commit_oe = find_mark(commit_mark);
2542                if (commit_oe->type != OBJ_COMMIT)
2543                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2544                oidcpy(&commit_oid, &commit_oe->idx.oid);
2545        } else if (!get_oid(p, &commit_oid)) {
2546                unsigned long size;
2547                char *buf = read_object_with_reference(&commit_oid,
2548                                                       commit_type, &size,
2549                                                       &commit_oid);
2550                if (!buf || size < 46)
2551                        die("Not a valid commit: %s", p);
2552                free(buf);
2553        } else
2554                die("Invalid ref name or SHA1 expression: %s", p);
2555
2556        if (inline_data) {
2557                if (p != uq.buf) {
2558                        strbuf_addstr(&uq, p);
2559                        p = uq.buf;
2560                }
2561                read_next_command();
2562                parse_and_store_blob(&last_blob, &oid, 0);
2563        } else if (oe) {
2564                if (oe->type != OBJ_BLOB)
2565                        die("Not a blob (actually a %s): %s",
2566                                type_name(oe->type), command_buf.buf);
2567        } else if (!is_null_oid(&oid)) {
2568                enum object_type type = oid_object_info(the_repository, &oid,
2569                                                        NULL);
2570                if (type < 0)
2571                        die("Blob not found: %s", command_buf.buf);
2572                if (type != OBJ_BLOB)
2573                        die("Not a blob (actually a %s): %s",
2574                            type_name(type), command_buf.buf);
2575        }
2576
2577        construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2578        if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2579                b->num_notes--;
2580
2581        if (is_null_oid(&oid))
2582                return; /* nothing to insert */
2583
2584        b->num_notes++;
2585        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2586        construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2587        tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2588}
2589
2590static void file_change_deleteall(struct branch *b)
2591{
2592        release_tree_content_recursive(b->branch_tree.tree);
2593        oidclr(&b->branch_tree.versions[0].oid);
2594        oidclr(&b->branch_tree.versions[1].oid);
2595        load_tree(&b->branch_tree);
2596        b->num_notes = 0;
2597}
2598
2599static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2600{
2601        if (!buf || size < GIT_SHA1_HEXSZ + 6)
2602                die("Not a valid commit: %s", oid_to_hex(&b->oid));
2603        if (memcmp("tree ", buf, 5)
2604                || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2605                die("The commit %s is corrupt", oid_to_hex(&b->oid));
2606        oidcpy(&b->branch_tree.versions[0].oid,
2607               &b->branch_tree.versions[1].oid);
2608}
2609
2610static void parse_from_existing(struct branch *b)
2611{
2612        if (is_null_oid(&b->oid)) {
2613                oidclr(&b->branch_tree.versions[0].oid);
2614                oidclr(&b->branch_tree.versions[1].oid);
2615        } else {
2616                unsigned long size;
2617                char *buf;
2618
2619                buf = read_object_with_reference(&b->oid, commit_type, &size,
2620                                                 &b->oid);
2621                parse_from_commit(b, buf, size);
2622                free(buf);
2623        }
2624}
2625
2626static int parse_from(struct branch *b)
2627{
2628        const char *from;
2629        struct branch *s;
2630        struct object_id oid;
2631
2632        if (!skip_prefix(command_buf.buf, "from ", &from))
2633                return 0;
2634
2635        oidcpy(&oid, &b->branch_tree.versions[1].oid);
2636
2637        s = lookup_branch(from);
2638        if (b == s)
2639                die("Can't create a branch from itself: %s", b->name);
2640        else if (s) {
2641                struct object_id *t = &s->branch_tree.versions[1].oid;
2642                oidcpy(&b->oid, &s->oid);
2643                oidcpy(&b->branch_tree.versions[0].oid, t);
2644                oidcpy(&b->branch_tree.versions[1].oid, t);
2645        } else if (*from == ':') {
2646                uintmax_t idnum = parse_mark_ref_eol(from);
2647                struct object_entry *oe = find_mark(idnum);
2648                if (oe->type != OBJ_COMMIT)
2649                        die("Mark :%" PRIuMAX " not a commit", idnum);
2650                if (oidcmp(&b->oid, &oe->idx.oid)) {
2651                        oidcpy(&b->oid, &oe->idx.oid);
2652                        if (oe->pack_id != MAX_PACK_ID) {
2653                                unsigned long size;
2654                                char *buf = gfi_unpack_entry(oe, &size);
2655                                parse_from_commit(b, buf, size);
2656                                free(buf);
2657                        } else
2658                                parse_from_existing(b);
2659                }
2660        } else if (!get_oid(from, &b->oid)) {
2661                parse_from_existing(b);
2662                if (is_null_oid(&b->oid))
2663                        b->delete = 1;
2664        }
2665        else
2666                die("Invalid ref name or SHA1 expression: %s", from);
2667
2668        if (b->branch_tree.tree && oidcmp(&oid, &b->branch_tree.versions[1].oid)) {
2669                release_tree_content_recursive(b->branch_tree.tree);
2670                b->branch_tree.tree = NULL;
2671        }
2672
2673        read_next_command();
2674        return 1;
2675}
2676
2677static struct hash_list *parse_merge(unsigned int *count)
2678{
2679        struct hash_list *list = NULL, **tail = &list, *n;
2680        const char *from;
2681        struct branch *s;
2682
2683        *count = 0;
2684        while (skip_prefix(command_buf.buf, "merge ", &from)) {
2685                n = xmalloc(sizeof(*n));
2686                s = lookup_branch(from);
2687                if (s)
2688                        oidcpy(&n->oid, &s->oid);
2689                else if (*from == ':') {
2690                        uintmax_t idnum = parse_mark_ref_eol(from);
2691                        struct object_entry *oe = find_mark(idnum);
2692                        if (oe->type != OBJ_COMMIT)
2693                                die("Mark :%" PRIuMAX " not a commit", idnum);
2694                        oidcpy(&n->oid, &oe->idx.oid);
2695                } else if (!get_oid(from, &n->oid)) {
2696                        unsigned long size;
2697                        char *buf = read_object_with_reference(&n->oid,
2698                                                               commit_type,
2699                                                               &size, &n->oid);
2700                        if (!buf || size < 46)
2701                                die("Not a valid commit: %s", from);
2702                        free(buf);
2703                } else
2704                        die("Invalid ref name or SHA1 expression: %s", from);
2705
2706                n->next = NULL;
2707                *tail = n;
2708                tail = &n->next;
2709
2710                (*count)++;
2711                read_next_command();
2712        }
2713        return list;
2714}
2715
2716static void parse_new_commit(const char *arg)
2717{
2718        static struct strbuf msg = STRBUF_INIT;
2719        struct branch *b;
2720        char *author = NULL;
2721        char *committer = NULL;
2722        struct hash_list *merge_list = NULL;
2723        unsigned int merge_count;
2724        unsigned char prev_fanout, new_fanout;
2725        const char *v;
2726
2727        b = lookup_branch(arg);
2728        if (!b)
2729                b = new_branch(arg);
2730
2731        read_next_command();
2732        parse_mark();
2733        if (skip_prefix(command_buf.buf, "author ", &v)) {
2734                author = parse_ident(v);
2735                read_next_command();
2736        }
2737        if (skip_prefix(command_buf.buf, "committer ", &v)) {
2738                committer = parse_ident(v);
2739                read_next_command();
2740        }
2741        if (!committer)
2742                die("Expected committer but didn't get one");
2743        parse_data(&msg, 0, NULL);
2744        read_next_command();
2745        parse_from(b);
2746        merge_list = parse_merge(&merge_count);
2747
2748        /* ensure the branch is active/loaded */
2749        if (!b->branch_tree.tree || !max_active_branches) {
2750                unload_one_branch();
2751                load_branch(b);
2752        }
2753
2754        prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2755
2756        /* file_change* */
2757        while (command_buf.len > 0) {
2758                if (skip_prefix(command_buf.buf, "M ", &v))
2759                        file_change_m(v, b);
2760                else if (skip_prefix(command_buf.buf, "D ", &v))
2761                        file_change_d(v, b);
2762                else if (skip_prefix(command_buf.buf, "R ", &v))
2763                        file_change_cr(v, b, 1);
2764                else if (skip_prefix(command_buf.buf, "C ", &v))
2765                        file_change_cr(v, b, 0);
2766                else if (skip_prefix(command_buf.buf, "N ", &v))
2767                        note_change_n(v, b, &prev_fanout);
2768                else if (!strcmp("deleteall", command_buf.buf))
2769                        file_change_deleteall(b);
2770                else if (skip_prefix(command_buf.buf, "ls ", &v))
2771                        parse_ls(v, b);
2772                else {
2773                        unread_command_buf = 1;
2774                        break;
2775                }
2776                if (read_next_command() == EOF)
2777                        break;
2778        }
2779
2780        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2781        if (new_fanout != prev_fanout)
2782                b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2783
2784        /* build the tree and the commit */
2785        store_tree(&b->branch_tree);
2786        oidcpy(&b->branch_tree.versions[0].oid,
2787               &b->branch_tree.versions[1].oid);
2788
2789        strbuf_reset(&new_data);
2790        strbuf_addf(&new_data, "tree %s\n",
2791                oid_to_hex(&b->branch_tree.versions[1].oid));
2792        if (!is_null_oid(&b->oid))
2793                strbuf_addf(&new_data, "parent %s\n",
2794                            oid_to_hex(&b->oid));
2795        while (merge_list) {
2796                struct hash_list *next = merge_list->next;
2797                strbuf_addf(&new_data, "parent %s\n",
2798                            oid_to_hex(&merge_list->oid));
2799                free(merge_list);
2800                merge_list = next;
2801        }
2802        strbuf_addf(&new_data,
2803                "author %s\n"
2804                "committer %s\n"
2805                "\n",
2806                author ? author : committer, committer);
2807        strbuf_addbuf(&new_data, &msg);
2808        free(author);
2809        free(committer);
2810
2811        if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2812                b->pack_id = pack_id;
2813        b->last_commit = object_count_by_type[OBJ_COMMIT];
2814}
2815
2816static void parse_new_tag(const char *arg)
2817{
2818        static struct strbuf msg = STRBUF_INIT;
2819        const char *from;
2820        char *tagger;
2821        struct branch *s;
2822        struct tag *t;
2823        uintmax_t from_mark = 0;
2824        struct object_id oid;
2825        enum object_type type;
2826        const char *v;
2827
2828        t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2829        memset(t, 0, sizeof(struct tag));
2830        t->name = pool_strdup(arg);
2831        if (last_tag)
2832                last_tag->next_tag = t;
2833        else
2834                first_tag = t;
2835        last_tag = t;
2836        read_next_command();
2837
2838        /* from ... */
2839        if (!skip_prefix(command_buf.buf, "from ", &from))
2840                die("Expected from command, got %s", command_buf.buf);
2841        s = lookup_branch(from);
2842        if (s) {
2843                if (is_null_oid(&s->oid))
2844                        die("Can't tag an empty branch.");
2845                oidcpy(&oid, &s->oid);
2846                type = OBJ_COMMIT;
2847        } else if (*from == ':') {
2848                struct object_entry *oe;
2849                from_mark = parse_mark_ref_eol(from);
2850                oe = find_mark(from_mark);
2851                type = oe->type;
2852                oidcpy(&oid, &oe->idx.oid);
2853        } else if (!get_oid(from, &oid)) {
2854                struct object_entry *oe = find_object(&oid);
2855                if (!oe) {
2856                        type = oid_object_info(the_repository, &oid, NULL);
2857                        if (type < 0)
2858                                die("Not a valid object: %s", from);
2859                } else
2860                        type = oe->type;
2861        } else
2862                die("Invalid ref name or SHA1 expression: %s", from);
2863        read_next_command();
2864
2865        /* tagger ... */
2866        if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2867                tagger = parse_ident(v);
2868                read_next_command();
2869        } else
2870                tagger = NULL;
2871
2872        /* tag payload/message */
2873        parse_data(&msg, 0, NULL);
2874
2875        /* build the tag object */
2876        strbuf_reset(&new_data);
2877
2878        strbuf_addf(&new_data,
2879                    "object %s\n"
2880                    "type %s\n"
2881                    "tag %s\n",
2882                    oid_to_hex(&oid), type_name(type), t->name);
2883        if (tagger)
2884                strbuf_addf(&new_data,
2885                            "tagger %s\n", tagger);
2886        strbuf_addch(&new_data, '\n');
2887        strbuf_addbuf(&new_data, &msg);
2888        free(tagger);
2889
2890        if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2891                t->pack_id = MAX_PACK_ID;
2892        else
2893                t->pack_id = pack_id;
2894}
2895
2896static void parse_reset_branch(const char *arg)
2897{
2898        struct branch *b;
2899
2900        b = lookup_branch(arg);
2901        if (b) {
2902                oidclr(&b->oid);
2903                oidclr(&b->branch_tree.versions[0].oid);
2904                oidclr(&b->branch_tree.versions[1].oid);
2905                if (b->branch_tree.tree) {
2906                        release_tree_content_recursive(b->branch_tree.tree);
2907                        b->branch_tree.tree = NULL;
2908                }
2909        }
2910        else
2911                b = new_branch(arg);
2912        read_next_command();
2913        parse_from(b);
2914        if (command_buf.len > 0)
2915                unread_command_buf = 1;
2916}
2917
2918static void cat_blob_write(const char *buf, unsigned long size)
2919{
2920        if (write_in_full(cat_blob_fd, buf, size) < 0)
2921                die_errno("Write to frontend failed");
2922}
2923
2924static void cat_blob(struct object_entry *oe, struct object_id *oid)
2925{
2926        struct strbuf line = STRBUF_INIT;
2927        unsigned long size;
2928        enum object_type type = 0;
2929        char *buf;
2930
2931        if (!oe || oe->pack_id == MAX_PACK_ID) {
2932                buf = read_object_file(oid, &type, &size);
2933        } else {
2934                type = oe->type;
2935                buf = gfi_unpack_entry(oe, &size);
2936        }
2937
2938        /*
2939         * Output based on batch_one_object() from cat-file.c.
2940         */
2941        if (type <= 0) {
2942                strbuf_reset(&line);
2943                strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2944                cat_blob_write(line.buf, line.len);
2945                strbuf_release(&line);
2946                free(buf);
2947                return;
2948        }
2949        if (!buf)
2950                die("Can't read object %s", oid_to_hex(oid));
2951        if (type != OBJ_BLOB)
2952                die("Object %s is a %s but a blob was expected.",
2953                    oid_to_hex(oid), type_name(type));
2954        strbuf_reset(&line);
2955        strbuf_addf(&line, "%s %s %lu\n", oid_to_hex(oid),
2956                                                type_name(type), size);
2957        cat_blob_write(line.buf, line.len);
2958        strbuf_release(&line);
2959        cat_blob_write(buf, size);
2960        cat_blob_write("\n", 1);
2961        if (oe && oe->pack_id == pack_id) {
2962                last_blob.offset = oe->idx.offset;
2963                strbuf_attach(&last_blob.data, buf, size, size);
2964                last_blob.depth = oe->depth;
2965        } else
2966                free(buf);
2967}
2968
2969static void parse_get_mark(const char *p)
2970{
2971        struct object_entry *oe;
2972        char output[GIT_MAX_HEXSZ + 2];
2973
2974        /* get-mark SP <object> LF */
2975        if (*p != ':')
2976                die("Not a mark: %s", p);
2977
2978        oe = find_mark(parse_mark_ref_eol(p));
2979        if (!oe)
2980                die("Unknown mark: %s", command_buf.buf);
2981
2982        xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2983        cat_blob_write(output, GIT_SHA1_HEXSZ + 1);
2984}
2985
2986static void parse_cat_blob(const char *p)
2987{
2988        struct object_entry *oe;
2989        struct object_id oid;
2990
2991        /* cat-blob SP <object> LF */
2992        if (*p == ':') {
2993                oe = find_mark(parse_mark_ref_eol(p));
2994                if (!oe)
2995                        die("Unknown mark: %s", command_buf.buf);
2996                oidcpy(&oid, &oe->idx.oid);
2997        } else {
2998                if (parse_oid_hex(p, &oid, &p))
2999                        die("Invalid dataref: %s", command_buf.buf);
3000                if (*p)
3001                        die("Garbage after SHA1: %s", command_buf.buf);
3002                oe = find_object(&oid);
3003        }
3004
3005        cat_blob(oe, &oid);
3006}
3007
3008static struct object_entry *dereference(struct object_entry *oe,
3009                                        struct object_id *oid)
3010{
3011        unsigned long size;
3012        char *buf = NULL;
3013        if (!oe) {
3014                enum object_type type = oid_object_info(the_repository, oid,
3015                                                        NULL);
3016                if (type < 0)
3017                        die("object not found: %s", oid_to_hex(oid));
3018                /* cache it! */
3019                oe = insert_object(oid);
3020                oe->type = type;
3021                oe->pack_id = MAX_PACK_ID;
3022                oe->idx.offset = 1;
3023        }
3024        switch (oe->type) {
3025        case OBJ_TREE:  /* easy case. */
3026                return oe;
3027        case OBJ_COMMIT:
3028        case OBJ_TAG:
3029                break;
3030        default:
3031                die("Not a tree-ish: %s", command_buf.buf);
3032        }
3033
3034        if (oe->pack_id != MAX_PACK_ID) {       /* in a pack being written */
3035                buf = gfi_unpack_entry(oe, &size);
3036        } else {
3037                enum object_type unused;
3038                buf = read_object_file(oid, &unused, &size);
3039        }
3040        if (!buf)
3041                die("Can't load object %s", oid_to_hex(oid));
3042
3043        /* Peel one layer. */
3044        switch (oe->type) {
3045        case OBJ_TAG:
3046                if (size < GIT_SHA1_HEXSZ + strlen("object ") ||
3047                    get_oid_hex(buf + strlen("object "), oid))
3048                        die("Invalid SHA1 in tag: %s", command_buf.buf);
3049                break;
3050        case OBJ_COMMIT:
3051                if (size < GIT_SHA1_HEXSZ + strlen("tree ") ||
3052                    get_oid_hex(buf + strlen("tree "), oid))
3053                        die("Invalid SHA1 in commit: %s", command_buf.buf);
3054        }
3055
3056        free(buf);
3057        return find_object(oid);
3058}
3059
3060static struct object_entry *parse_treeish_dataref(const char **p)
3061{
3062        struct object_id oid;
3063        struct object_entry *e;
3064
3065        if (**p == ':') {       /* <mark> */
3066                e = find_mark(parse_mark_ref_space(p));
3067                if (!e)
3068                        die("Unknown mark: %s", command_buf.buf);
3069                oidcpy(&oid, &e->idx.oid);
3070        } else {        /* <sha1> */
3071                if (parse_oid_hex(*p, &oid, p))
3072                        die("Invalid dataref: %s", command_buf.buf);
3073                e = find_object(&oid);
3074                if (*(*p)++ != ' ')
3075                        die("Missing space after tree-ish: %s", command_buf.buf);
3076        }
3077
3078        while (!e || e->type != OBJ_TREE)
3079                e = dereference(e, &oid);
3080        return e;
3081}
3082
3083static void print_ls(int mode, const unsigned char *sha1, const char *path)
3084{
3085        static struct strbuf line = STRBUF_INIT;
3086
3087        /* See show_tree(). */
3088        const char *type =
3089                S_ISGITLINK(mode) ? commit_type :
3090                S_ISDIR(mode) ? tree_type :
3091                blob_type;
3092
3093        if (!mode) {
3094                /* missing SP path LF */
3095                strbuf_reset(&line);
3096                strbuf_addstr(&line, "missing ");
3097                quote_c_style(path, &line, NULL, 0);
3098                strbuf_addch(&line, '\n');
3099        } else {
3100                /* mode SP type SP object_name TAB path LF */
3101                strbuf_reset(&line);
3102                strbuf_addf(&line, "%06o %s %s\t",
3103                                mode & ~NO_DELTA, type, sha1_to_hex(sha1));
3104                quote_c_style(path, &line, NULL, 0);
3105                strbuf_addch(&line, '\n');
3106        }
3107        cat_blob_write(line.buf, line.len);
3108}
3109
3110static void parse_ls(const char *p, struct branch *b)
3111{
3112        struct tree_entry *root = NULL;
3113        struct tree_entry leaf = {NULL};
3114
3115        /* ls SP (<tree-ish> SP)? <path> */
3116        if (*p == '"') {
3117                if (!b)
3118                        die("Not in a commit: %s", command_buf.buf);
3119                root = &b->branch_tree;
3120        } else {
3121                struct object_entry *e = parse_treeish_dataref(&p);
3122                root = new_tree_entry();
3123                oidcpy(&root->versions[1].oid, &e->idx.oid);
3124                if (!is_null_oid(&root->versions[1].oid))
3125                        root->versions[1].mode = S_IFDIR;
3126                load_tree(root);
3127        }
3128        if (*p == '"') {
3129                static struct strbuf uq = STRBUF_INIT;
3130                const char *endp;
3131                strbuf_reset(&uq);
3132                if (unquote_c_style(&uq, p, &endp))
3133                        die("Invalid path: %s", command_buf.buf);
3134                if (*endp)
3135                        die("Garbage after path in: %s", command_buf.buf);
3136                p = uq.buf;
3137        }
3138        tree_content_get(root, p, &leaf, 1);
3139        /*
3140         * A directory in preparation would have a sha1 of zero
3141         * until it is saved.  Save, for simplicity.
3142         */
3143        if (S_ISDIR(leaf.versions[1].mode))
3144                store_tree(&leaf);
3145
3146        print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3147        if (leaf.tree)
3148                release_tree_content_recursive(leaf.tree);
3149        if (!b || root != &b->branch_tree)
3150                release_tree_entry(root);
3151}
3152
3153static void checkpoint(void)
3154{
3155        checkpoint_requested = 0;
3156        if (object_count) {
3157                cycle_packfile();
3158        }
3159        dump_branches();
3160        dump_tags();
3161        dump_marks();
3162}
3163
3164static void parse_checkpoint(void)
3165{
3166        checkpoint_requested = 1;
3167        skip_optional_lf();
3168}
3169
3170static void parse_progress(void)
3171{
3172        fwrite(command_buf.buf, 1, command_buf.len, stdout);
3173        fputc('\n', stdout);
3174        fflush(stdout);
3175        skip_optional_lf();
3176}
3177
3178static char* make_fast_import_path(const char *path)
3179{
3180        if (!relative_marks_paths || is_absolute_path(path))
3181                return xstrdup(path);
3182        return git_pathdup("info/fast-import/%s", path);
3183}
3184
3185static void option_import_marks(const char *marks,
3186                                        int from_stream, int ignore_missing)
3187{
3188        if (import_marks_file) {
3189                if (from_stream)
3190                        die("Only one import-marks command allowed per stream");
3191
3192                /* read previous mark file */
3193                if(!import_marks_file_from_stream)
3194                        read_marks();
3195        }
3196
3197        import_marks_file = make_fast_import_path(marks);
3198        safe_create_leading_directories_const(import_marks_file);
3199        import_marks_file_from_stream = from_stream;
3200        import_marks_file_ignore_missing = ignore_missing;
3201}
3202
3203static void option_date_format(const char *fmt)
3204{
3205        if (!strcmp(fmt, "raw"))
3206                whenspec = WHENSPEC_RAW;
3207        else if (!strcmp(fmt, "rfc2822"))
3208                whenspec = WHENSPEC_RFC2822;
3209        else if (!strcmp(fmt, "now"))
3210                whenspec = WHENSPEC_NOW;
3211        else
3212                die("unknown --date-format argument %s", fmt);
3213}
3214
3215static unsigned long ulong_arg(const char *option, const char *arg)
3216{
3217        char *endptr;
3218        unsigned long rv = strtoul(arg, &endptr, 0);
3219        if (strchr(arg, '-') || endptr == arg || *endptr)
3220                die("%s: argument must be a non-negative integer", option);
3221        return rv;
3222}
3223
3224static void option_depth(const char *depth)
3225{
3226        max_depth = ulong_arg("--depth", depth);
3227        if (max_depth > MAX_DEPTH)
3228                die("--depth cannot exceed %u", MAX_DEPTH);
3229}
3230
3231static void option_active_branches(const char *branches)
3232{
3233        max_active_branches = ulong_arg("--active-branches", branches);
3234}
3235
3236static void option_export_marks(const char *marks)
3237{
3238        export_marks_file = make_fast_import_path(marks);
3239        safe_create_leading_directories_const(export_marks_file);
3240}
3241
3242static void option_cat_blob_fd(const char *fd)
3243{
3244        unsigned long n = ulong_arg("--cat-blob-fd", fd);
3245        if (n > (unsigned long) INT_MAX)
3246                die("--cat-blob-fd cannot exceed %d", INT_MAX);
3247        cat_blob_fd = (int) n;
3248}
3249
3250static void option_export_pack_edges(const char *edges)
3251{
3252        if (pack_edges)
3253                fclose(pack_edges);
3254        pack_edges = xfopen(edges, "a");
3255}
3256
3257static int parse_one_option(const char *option)
3258{
3259        if (skip_prefix(option, "max-pack-size=", &option)) {
3260                unsigned long v;
3261                if (!git_parse_ulong(option, &v))
3262                        return 0;
3263                if (v < 8192) {
3264                        warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3265                        v *= 1024 * 1024;
3266                } else if (v < 1024 * 1024) {
3267                        warning("minimum max-pack-size is 1 MiB");
3268                        v = 1024 * 1024;
3269                }
3270                max_packsize = v;
3271        } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3272                unsigned long v;
3273                if (!git_parse_ulong(option, &v))
3274                        return 0;
3275                big_file_threshold = v;
3276        } else if (skip_prefix(option, "depth=", &option)) {
3277                option_depth(option);
3278        } else if (skip_prefix(option, "active-branches=", &option)) {
3279                option_active_branches(option);
3280        } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3281                option_export_pack_edges(option);
3282        } else if (starts_with(option, "quiet")) {
3283                show_stats = 0;
3284        } else if (starts_with(option, "stats")) {
3285                show_stats = 1;
3286        } else {
3287                return 0;
3288        }
3289
3290        return 1;
3291}
3292
3293static int parse_one_feature(const char *feature, int from_stream)
3294{
3295        const char *arg;
3296
3297        if (skip_prefix(feature, "date-format=", &arg)) {
3298                option_date_format(arg);
3299        } else if (skip_prefix(feature, "import-marks=", &arg)) {
3300                option_import_marks(arg, from_stream, 0);
3301        } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3302                option_import_marks(arg, from_stream, 1);
3303        } else if (skip_prefix(feature, "export-marks=", &arg)) {
3304                option_export_marks(arg);
3305        } else if (!strcmp(feature, "get-mark")) {
3306                ; /* Don't die - this feature is supported */
3307        } else if (!strcmp(feature, "cat-blob")) {
3308                ; /* Don't die - this feature is supported */
3309        } else if (!strcmp(feature, "relative-marks")) {
3310                relative_marks_paths = 1;
3311        } else if (!strcmp(feature, "no-relative-marks")) {
3312                relative_marks_paths = 0;
3313        } else if (!strcmp(feature, "done")) {
3314                require_explicit_termination = 1;
3315        } else if (!strcmp(feature, "force")) {
3316                force_update = 1;
3317        } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3318                ; /* do nothing; we have the feature */
3319        } else {
3320                return 0;
3321        }
3322
3323        return 1;
3324}
3325
3326static void parse_feature(const char *feature)
3327{
3328        if (seen_data_command)
3329                die("Got feature command '%s' after data command", feature);
3330
3331        if (parse_one_feature(feature, 1))
3332                return;
3333
3334        die("This version of fast-import does not support feature %s.", feature);
3335}
3336
3337static void parse_option(const char *option)
3338{
3339        if (seen_data_command)
3340                die("Got option command '%s' after data command", option);
3341
3342        if (parse_one_option(option))
3343                return;
3344
3345        die("This version of fast-import does not support option: %s", option);
3346}
3347
3348static void git_pack_config(void)
3349{
3350        int indexversion_value;
3351        int limit;
3352        unsigned long packsizelimit_value;
3353
3354        if (!git_config_get_ulong("pack.depth", &max_depth)) {
3355                if (max_depth > MAX_DEPTH)
3356                        max_depth = MAX_DEPTH;
3357        }
3358        if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3359                pack_idx_opts.version = indexversion_value;
3360                if (pack_idx_opts.version > 2)
3361                        git_die_config("pack.indexversion",
3362                                        "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3363        }
3364        if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3365                max_packsize = packsizelimit_value;
3366
3367        if (!git_config_get_int("fastimport.unpacklimit", &limit))
3368                unpack_limit = limit;
3369        else if (!git_config_get_int("transfer.unpacklimit", &limit))
3370                unpack_limit = limit;
3371
3372        git_config(git_default_config, NULL);
3373}
3374
3375static const char fast_import_usage[] =
3376"git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3377
3378static void parse_argv(void)
3379{
3380        unsigned int i;
3381
3382        for (i = 1; i < global_argc; i++) {
3383                const char *a = global_argv[i];
3384
3385                if (*a != '-' || !strcmp(a, "--"))
3386                        break;
3387
3388                if (!skip_prefix(a, "--", &a))
3389                        die("unknown option %s", a);
3390
3391                if (parse_one_option(a))
3392                        continue;
3393
3394                if (parse_one_feature(a, 0))
3395                        continue;
3396
3397                if (skip_prefix(a, "cat-blob-fd=", &a)) {
3398                        option_cat_blob_fd(a);
3399                        continue;
3400                }
3401
3402                die("unknown option --%s", a);
3403        }
3404        if (i != global_argc)
3405                usage(fast_import_usage);
3406
3407        seen_data_command = 1;
3408        if (import_marks_file)
3409                read_marks();
3410}
3411
3412int cmd_main(int argc, const char **argv)
3413{
3414        unsigned int i;
3415
3416        if (argc == 2 && !strcmp(argv[1], "-h"))
3417                usage(fast_import_usage);
3418
3419        setup_git_directory();
3420        reset_pack_idx_option(&pack_idx_opts);
3421        git_pack_config();
3422
3423        alloc_objects(object_entry_alloc);
3424        strbuf_init(&command_buf, 0);
3425        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3426        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3427        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3428        marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3429
3430        global_argc = argc;
3431        global_argv = argv;
3432
3433        rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3434        for (i = 0; i < (cmd_save - 1); i++)
3435                rc_free[i].next = &rc_free[i + 1];
3436        rc_free[cmd_save - 1].next = NULL;
3437
3438        start_packfile();
3439        set_die_routine(die_nicely);
3440        set_checkpoint_signal();
3441        while (read_next_command() != EOF) {
3442                const char *v;
3443                if (!strcmp("blob", command_buf.buf))
3444                        parse_new_blob();
3445                else if (skip_prefix(command_buf.buf, "ls ", &v))
3446                        parse_ls(v, NULL);
3447                else if (skip_prefix(command_buf.buf, "commit ", &v))
3448                        parse_new_commit(v);
3449                else if (skip_prefix(command_buf.buf, "tag ", &v))
3450                        parse_new_tag(v);
3451                else if (skip_prefix(command_buf.buf, "reset ", &v))
3452                        parse_reset_branch(v);
3453                else if (!strcmp("checkpoint", command_buf.buf))
3454                        parse_checkpoint();
3455                else if (!strcmp("done", command_buf.buf))
3456                        break;
3457                else if (starts_with(command_buf.buf, "progress "))
3458                        parse_progress();
3459                else if (skip_prefix(command_buf.buf, "feature ", &v))
3460                        parse_feature(v);
3461                else if (skip_prefix(command_buf.buf, "option git ", &v))
3462                        parse_option(v);
3463                else if (starts_with(command_buf.buf, "option "))
3464                        /* ignore non-git options*/;
3465                else
3466                        die("Unsupported command: %s", command_buf.buf);
3467
3468                if (checkpoint_requested)
3469                        checkpoint();
3470        }
3471
3472        /* argv hasn't been parsed yet, do so */
3473        if (!seen_data_command)
3474                parse_argv();
3475
3476        if (require_explicit_termination && feof(stdin))
3477                die("stream ends early");
3478
3479        end_packfile();
3480
3481        dump_branches();
3482        dump_tags();
3483        unkeep_all_packs();
3484        dump_marks();
3485
3486        if (pack_edges)
3487                fclose(pack_edges);
3488
3489        if (show_stats) {
3490                uintmax_t total_count = 0, duplicate_count = 0;
3491                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3492                        total_count += object_count_by_type[i];
3493                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3494                        duplicate_count += duplicate_count_by_type[i];
3495
3496                fprintf(stderr, "%s statistics:\n", argv[0]);
3497                fprintf(stderr, "---------------------------------------------------------------------\n");
3498                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3499                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3500                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]);
3501                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]);
3502                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]);
3503                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]);
3504                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3505                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3506                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3507                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3508                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3509                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3510                fprintf(stderr, "---------------------------------------------------------------------\n");
3511                pack_report();
3512                fprintf(stderr, "---------------------------------------------------------------------\n");
3513                fprintf(stderr, "\n");
3514        }
3515
3516        return failure ? 1 : 0;
3517}