fast-import.con commit Merge branch 'js/configurable-tab' (47bfb3d)
   1/*
   2(See Documentation/git-fast-import.txt for maintained documentation.)
   3Format of STDIN stream:
   4
   5  stream ::= cmd*;
   6
   7  cmd ::= new_blob
   8        | new_commit
   9        | new_tag
  10        | reset_branch
  11        | checkpoint
  12        | progress
  13        ;
  14
  15  new_blob ::= 'blob' lf
  16    mark?
  17    file_content;
  18  file_content ::= data;
  19
  20  new_commit ::= 'commit' sp ref_str lf
  21    mark?
  22    ('author' (sp name)? sp '<' email '>' sp when lf)?
  23    'committer' (sp name)? sp '<' email '>' sp when lf
  24    commit_msg
  25    ('from' sp committish lf)?
  26    ('merge' sp committish lf)*
  27    file_change*
  28    lf?;
  29  commit_msg ::= data;
  30
  31  file_change ::= file_clr
  32    | file_del
  33    | file_rnm
  34    | file_cpy
  35    | file_obm
  36    | file_inm;
  37  file_clr ::= 'deleteall' lf;
  38  file_del ::= 'D' sp path_str lf;
  39  file_rnm ::= 'R' sp path_str sp path_str lf;
  40  file_cpy ::= 'C' sp path_str sp path_str lf;
  41  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  42  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  43    data;
  44  note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
  45  note_inm ::= 'N' sp 'inline' sp committish lf
  46    data;
  47
  48  new_tag ::= 'tag' sp tag_str lf
  49    'from' sp committish lf
  50    ('tagger' (sp name)? sp '<' email '>' sp when lf)?
  51    tag_msg;
  52  tag_msg ::= data;
  53
  54  reset_branch ::= 'reset' sp ref_str lf
  55    ('from' sp committish lf)?
  56    lf?;
  57
  58  checkpoint ::= 'checkpoint' lf
  59    lf?;
  60
  61  progress ::= 'progress' sp not_lf* lf
  62    lf?;
  63
  64     # note: the first idnum in a stream should be 1 and subsequent
  65     # idnums should not have gaps between values as this will cause
  66     # the stream parser to reserve space for the gapped values.  An
  67     # idnum can be updated in the future to a new object by issuing
  68     # a new mark directive with the old idnum.
  69     #
  70  mark ::= 'mark' sp idnum lf;
  71  data ::= (delimited_data | exact_data)
  72    lf?;
  73
  74    # note: delim may be any string but must not contain lf.
  75    # data_line may contain any data but must not be exactly
  76    # delim.
  77  delimited_data ::= 'data' sp '<<' delim lf
  78    (data_line lf)*
  79    delim lf;
  80
  81     # note: declen indicates the length of binary_data in bytes.
  82     # declen does not include the lf preceding the binary data.
  83     #
  84  exact_data ::= 'data' sp declen lf
  85    binary_data;
  86
  87     # note: quoted strings are C-style quoting supporting \c for
  88     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  89     # is the signed byte value in octal.  Note that the only
  90     # characters which must actually be escaped to protect the
  91     # stream formatting is: \, " and LF.  Otherwise these values
  92     # are UTF8.
  93     #
  94  committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
  95  ref_str     ::= ref;
  96  sha1exp_str ::= sha1exp;
  97  tag_str     ::= tag;
  98  path_str    ::= path    | '"' quoted(path)    '"' ;
  99  mode        ::= '100644' | '644'
 100                | '100755' | '755'
 101                | '120000'
 102                ;
 103
 104  declen ::= # unsigned 32 bit value, ascii base10 notation;
 105  bigint ::= # unsigned integer value, ascii base10 notation;
 106  binary_data ::= # file content, not interpreted;
 107
 108  when         ::= raw_when | rfc2822_when;
 109  raw_when     ::= ts sp tz;
 110  rfc2822_when ::= # Valid RFC 2822 date and time;
 111
 112  sp ::= # ASCII space character;
 113  lf ::= # ASCII newline (LF) character;
 114
 115     # note: a colon (':') must precede the numerical value assigned to
 116     # an idnum.  This is to distinguish it from a ref or tag name as
 117     # GIT does not permit ':' in ref or tag strings.
 118     #
 119  idnum   ::= ':' bigint;
 120  path    ::= # GIT style file path, e.g. "a/b/c";
 121  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 122  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 123  sha1exp ::= # Any valid GIT SHA1 expression;
 124  hexsha1 ::= # SHA1 in hexadecimal format;
 125
 126     # note: name and email are UTF8 strings, however name must not
 127     # contain '<' or lf and email must not contain any of the
 128     # following: '<', '>', lf.
 129     #
 130  name  ::= # valid GIT author/committer name;
 131  email ::= # valid GIT author/committer email;
 132  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 133  tz    ::= # GIT style timezone;
 134
 135     # note: comments may appear anywhere in the input, except
 136     # within a data command.  Any form of the data command
 137     # always escapes the related input from comment processing.
 138     #
 139     # In case it is not clear, the '#' that starts the comment
 140     # must be the first character on that line (an lf
 141     # preceded it).
 142     #
 143  comment ::= '#' not_lf* lf;
 144  not_lf  ::= # Any byte that is not ASCII newline (LF);
 145*/
 146
 147#include "builtin.h"
 148#include "cache.h"
 149#include "object.h"
 150#include "blob.h"
 151#include "tree.h"
 152#include "commit.h"
 153#include "delta.h"
 154#include "pack.h"
 155#include "refs.h"
 156#include "csum-file.h"
 157#include "quote.h"
 158#include "exec_cmd.h"
 159#include "dir.h"
 160
 161#define PACK_ID_BITS 16
 162#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 163#define DEPTH_BITS 13
 164#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 165
 166struct object_entry
 167{
 168        struct pack_idx_entry idx;
 169        struct object_entry *next;
 170        uint32_t type : TYPE_BITS,
 171                pack_id : PACK_ID_BITS,
 172                depth : DEPTH_BITS;
 173};
 174
 175struct object_entry_pool
 176{
 177        struct object_entry_pool *next_pool;
 178        struct object_entry *next_free;
 179        struct object_entry *end;
 180        struct object_entry entries[FLEX_ARRAY]; /* more */
 181};
 182
 183struct mark_set
 184{
 185        union {
 186                struct object_entry *marked[1024];
 187                struct mark_set *sets[1024];
 188        } data;
 189        unsigned int shift;
 190};
 191
 192struct last_object
 193{
 194        struct strbuf data;
 195        off_t offset;
 196        unsigned int depth;
 197        unsigned no_swap : 1;
 198};
 199
 200struct mem_pool
 201{
 202        struct mem_pool *next_pool;
 203        char *next_free;
 204        char *end;
 205        uintmax_t space[FLEX_ARRAY]; /* more */
 206};
 207
 208struct atom_str
 209{
 210        struct atom_str *next_atom;
 211        unsigned short str_len;
 212        char str_dat[FLEX_ARRAY]; /* more */
 213};
 214
 215struct tree_content;
 216struct tree_entry
 217{
 218        struct tree_content *tree;
 219        struct atom_str *name;
 220        struct tree_entry_ms
 221        {
 222                uint16_t mode;
 223                unsigned char sha1[20];
 224        } versions[2];
 225};
 226
 227struct tree_content
 228{
 229        unsigned int entry_capacity; /* must match avail_tree_content */
 230        unsigned int entry_count;
 231        unsigned int delta_depth;
 232        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 233};
 234
 235struct avail_tree_content
 236{
 237        unsigned int entry_capacity; /* must match tree_content */
 238        struct avail_tree_content *next_avail;
 239};
 240
 241struct branch
 242{
 243        struct branch *table_next_branch;
 244        struct branch *active_next_branch;
 245        const char *name;
 246        struct tree_entry branch_tree;
 247        uintmax_t last_commit;
 248        uintmax_t num_notes;
 249        unsigned active : 1;
 250        unsigned pack_id : PACK_ID_BITS;
 251        unsigned char sha1[20];
 252};
 253
 254struct tag
 255{
 256        struct tag *next_tag;
 257        const char *name;
 258        unsigned int pack_id;
 259        unsigned char sha1[20];
 260};
 261
 262struct hash_list
 263{
 264        struct hash_list *next;
 265        unsigned char sha1[20];
 266};
 267
 268typedef enum {
 269        WHENSPEC_RAW = 1,
 270        WHENSPEC_RFC2822,
 271        WHENSPEC_NOW
 272} whenspec_type;
 273
 274struct recent_command
 275{
 276        struct recent_command *prev;
 277        struct recent_command *next;
 278        char *buf;
 279};
 280
 281/* Configured limits on output */
 282static unsigned long max_depth = 10;
 283static off_t max_packsize;
 284static uintmax_t big_file_threshold = 512 * 1024 * 1024;
 285static int force_update;
 286static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 287static int pack_compression_seen;
 288
 289/* Stats and misc. counters */
 290static uintmax_t alloc_count;
 291static uintmax_t marks_set_count;
 292static uintmax_t object_count_by_type[1 << TYPE_BITS];
 293static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 294static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 295static unsigned long object_count;
 296static unsigned long branch_count;
 297static unsigned long branch_load_count;
 298static int failure;
 299static FILE *pack_edges;
 300static unsigned int show_stats = 1;
 301static int global_argc;
 302static const char **global_argv;
 303
 304/* Memory pools */
 305static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 306static size_t total_allocd;
 307static struct mem_pool *mem_pool;
 308
 309/* Atom management */
 310static unsigned int atom_table_sz = 4451;
 311static unsigned int atom_cnt;
 312static struct atom_str **atom_table;
 313
 314/* The .pack file being generated */
 315static unsigned int pack_id;
 316static struct sha1file *pack_file;
 317static struct packed_git *pack_data;
 318static struct packed_git **all_packs;
 319static off_t pack_size;
 320
 321/* Table of objects we've written. */
 322static unsigned int object_entry_alloc = 5000;
 323static struct object_entry_pool *blocks;
 324static struct object_entry *object_table[1 << 16];
 325static struct mark_set *marks;
 326static const char *export_marks_file;
 327static const char *import_marks_file;
 328static int import_marks_file_from_stream;
 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 struct strbuf old_tree = STRBUF_INIT;
 340static struct strbuf new_tree = STRBUF_INIT;
 341
 342/* Branch data */
 343static unsigned long max_active_branches = 5;
 344static unsigned long cur_active_branches;
 345static unsigned long branch_table_sz = 1039;
 346static struct branch **branch_table;
 347static struct branch *active_branches;
 348
 349/* Tag data */
 350static struct tag *first_tag;
 351static struct tag *last_tag;
 352
 353/* Input stream parsing */
 354static whenspec_type whenspec = WHENSPEC_RAW;
 355static struct strbuf command_buf = STRBUF_INIT;
 356static int unread_command_buf;
 357static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 358static struct recent_command *cmd_tail = &cmd_hist;
 359static struct recent_command *rc_free;
 360static unsigned int cmd_save = 100;
 361static uintmax_t next_mark;
 362static struct strbuf new_data = STRBUF_INIT;
 363static int seen_data_command;
 364
 365static void parse_argv(void);
 366
 367static void write_branch_report(FILE *rpt, struct branch *b)
 368{
 369        fprintf(rpt, "%s:\n", b->name);
 370
 371        fprintf(rpt, "  status      :");
 372        if (b->active)
 373                fputs(" active", rpt);
 374        if (b->branch_tree.tree)
 375                fputs(" loaded", rpt);
 376        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 377                fputs(" dirty", rpt);
 378        fputc('\n', rpt);
 379
 380        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 381        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 382        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 383        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 384
 385        fputs("  last pack   : ", rpt);
 386        if (b->pack_id < MAX_PACK_ID)
 387                fprintf(rpt, "%u", b->pack_id);
 388        fputc('\n', rpt);
 389
 390        fputc('\n', rpt);
 391}
 392
 393static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 394
 395static void write_crash_report(const char *err)
 396{
 397        char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 398        FILE *rpt = fopen(loc, "w");
 399        struct branch *b;
 400        unsigned long lu;
 401        struct recent_command *rc;
 402
 403        if (!rpt) {
 404                error("can't write crash report %s: %s", loc, strerror(errno));
 405                return;
 406        }
 407
 408        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 409
 410        fprintf(rpt, "fast-import crash report:\n");
 411        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 412        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 413        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 414        fputc('\n', rpt);
 415
 416        fputs("fatal: ", rpt);
 417        fputs(err, rpt);
 418        fputc('\n', rpt);
 419
 420        fputc('\n', rpt);
 421        fputs("Most Recent Commands Before Crash\n", rpt);
 422        fputs("---------------------------------\n", rpt);
 423        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 424                if (rc->next == &cmd_hist)
 425                        fputs("* ", rpt);
 426                else
 427                        fputs("  ", rpt);
 428                fputs(rc->buf, rpt);
 429                fputc('\n', rpt);
 430        }
 431
 432        fputc('\n', rpt);
 433        fputs("Active Branch LRU\n", rpt);
 434        fputs("-----------------\n", rpt);
 435        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 436                cur_active_branches,
 437                max_active_branches);
 438        fputc('\n', rpt);
 439        fputs("  pos  clock name\n", rpt);
 440        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 441        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 442                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 443                        ++lu, b->last_commit, b->name);
 444
 445        fputc('\n', rpt);
 446        fputs("Inactive Branches\n", rpt);
 447        fputs("-----------------\n", rpt);
 448        for (lu = 0; lu < branch_table_sz; lu++) {
 449                for (b = branch_table[lu]; b; b = b->table_next_branch)
 450                        write_branch_report(rpt, b);
 451        }
 452
 453        if (first_tag) {
 454                struct tag *tg;
 455                fputc('\n', rpt);
 456                fputs("Annotated Tags\n", rpt);
 457                fputs("--------------\n", rpt);
 458                for (tg = first_tag; tg; tg = tg->next_tag) {
 459                        fputs(sha1_to_hex(tg->sha1), rpt);
 460                        fputc(' ', rpt);
 461                        fputs(tg->name, rpt);
 462                        fputc('\n', rpt);
 463                }
 464        }
 465
 466        fputc('\n', rpt);
 467        fputs("Marks\n", rpt);
 468        fputs("-----\n", rpt);
 469        if (export_marks_file)
 470                fprintf(rpt, "  exported to %s\n", export_marks_file);
 471        else
 472                dump_marks_helper(rpt, 0, marks);
 473
 474        fputc('\n', rpt);
 475        fputs("-------------------\n", rpt);
 476        fputs("END OF CRASH REPORT\n", rpt);
 477        fclose(rpt);
 478}
 479
 480static void end_packfile(void);
 481static void unkeep_all_packs(void);
 482static void dump_marks(void);
 483
 484static NORETURN void die_nicely(const char *err, va_list params)
 485{
 486        static int zombie;
 487        char message[2 * PATH_MAX];
 488
 489        vsnprintf(message, sizeof(message), err, params);
 490        fputs("fatal: ", stderr);
 491        fputs(message, stderr);
 492        fputc('\n', stderr);
 493
 494        if (!zombie) {
 495                zombie = 1;
 496                write_crash_report(message);
 497                end_packfile();
 498                unkeep_all_packs();
 499                dump_marks();
 500        }
 501        exit(128);
 502}
 503
 504static void alloc_objects(unsigned int cnt)
 505{
 506        struct object_entry_pool *b;
 507
 508        b = xmalloc(sizeof(struct object_entry_pool)
 509                + cnt * sizeof(struct object_entry));
 510        b->next_pool = blocks;
 511        b->next_free = b->entries;
 512        b->end = b->entries + cnt;
 513        blocks = b;
 514        alloc_count += cnt;
 515}
 516
 517static struct object_entry *new_object(unsigned char *sha1)
 518{
 519        struct object_entry *e;
 520
 521        if (blocks->next_free == blocks->end)
 522                alloc_objects(object_entry_alloc);
 523
 524        e = blocks->next_free++;
 525        hashcpy(e->idx.sha1, sha1);
 526        return e;
 527}
 528
 529static struct object_entry *find_object(unsigned char *sha1)
 530{
 531        unsigned int h = sha1[0] << 8 | sha1[1];
 532        struct object_entry *e;
 533        for (e = object_table[h]; e; e = e->next)
 534                if (!hashcmp(sha1, e->idx.sha1))
 535                        return e;
 536        return NULL;
 537}
 538
 539static struct object_entry *insert_object(unsigned char *sha1)
 540{
 541        unsigned int h = sha1[0] << 8 | sha1[1];
 542        struct object_entry *e = object_table[h];
 543        struct object_entry *p = NULL;
 544
 545        while (e) {
 546                if (!hashcmp(sha1, e->idx.sha1))
 547                        return e;
 548                p = e;
 549                e = e->next;
 550        }
 551
 552        e = new_object(sha1);
 553        e->next = NULL;
 554        e->idx.offset = 0;
 555        if (p)
 556                p->next = e;
 557        else
 558                object_table[h] = e;
 559        return e;
 560}
 561
 562static unsigned int hc_str(const char *s, size_t len)
 563{
 564        unsigned int r = 0;
 565        while (len-- > 0)
 566                r = r * 31 + *s++;
 567        return r;
 568}
 569
 570static void *pool_alloc(size_t len)
 571{
 572        struct mem_pool *p;
 573        void *r;
 574
 575        /* round up to a 'uintmax_t' alignment */
 576        if (len & (sizeof(uintmax_t) - 1))
 577                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 578
 579        for (p = mem_pool; p; p = p->next_pool)
 580                if ((p->end - p->next_free >= len))
 581                        break;
 582
 583        if (!p) {
 584                if (len >= (mem_pool_alloc/2)) {
 585                        total_allocd += len;
 586                        return xmalloc(len);
 587                }
 588                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 589                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 590                p->next_pool = mem_pool;
 591                p->next_free = (char *) p->space;
 592                p->end = p->next_free + mem_pool_alloc;
 593                mem_pool = p;
 594        }
 595
 596        r = p->next_free;
 597        p->next_free += len;
 598        return r;
 599}
 600
 601static void *pool_calloc(size_t count, size_t size)
 602{
 603        size_t len = count * size;
 604        void *r = pool_alloc(len);
 605        memset(r, 0, len);
 606        return r;
 607}
 608
 609static char *pool_strdup(const char *s)
 610{
 611        char *r = pool_alloc(strlen(s) + 1);
 612        strcpy(r, s);
 613        return r;
 614}
 615
 616static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 617{
 618        struct mark_set *s = marks;
 619        while ((idnum >> s->shift) >= 1024) {
 620                s = pool_calloc(1, sizeof(struct mark_set));
 621                s->shift = marks->shift + 10;
 622                s->data.sets[0] = marks;
 623                marks = s;
 624        }
 625        while (s->shift) {
 626                uintmax_t i = idnum >> s->shift;
 627                idnum -= i << s->shift;
 628                if (!s->data.sets[i]) {
 629                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 630                        s->data.sets[i]->shift = s->shift - 10;
 631                }
 632                s = s->data.sets[i];
 633        }
 634        if (!s->data.marked[idnum])
 635                marks_set_count++;
 636        s->data.marked[idnum] = oe;
 637}
 638
 639static struct object_entry *find_mark(uintmax_t idnum)
 640{
 641        uintmax_t orig_idnum = idnum;
 642        struct mark_set *s = marks;
 643        struct object_entry *oe = NULL;
 644        if ((idnum >> s->shift) < 1024) {
 645                while (s && s->shift) {
 646                        uintmax_t i = idnum >> s->shift;
 647                        idnum -= i << s->shift;
 648                        s = s->data.sets[i];
 649                }
 650                if (s)
 651                        oe = s->data.marked[idnum];
 652        }
 653        if (!oe)
 654                die("mark :%" PRIuMAX " not declared", orig_idnum);
 655        return oe;
 656}
 657
 658static struct atom_str *to_atom(const char *s, unsigned short len)
 659{
 660        unsigned int hc = hc_str(s, len) % atom_table_sz;
 661        struct atom_str *c;
 662
 663        for (c = atom_table[hc]; c; c = c->next_atom)
 664                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 665                        return c;
 666
 667        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 668        c->str_len = len;
 669        strncpy(c->str_dat, s, len);
 670        c->str_dat[len] = 0;
 671        c->next_atom = atom_table[hc];
 672        atom_table[hc] = c;
 673        atom_cnt++;
 674        return c;
 675}
 676
 677static struct branch *lookup_branch(const char *name)
 678{
 679        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 680        struct branch *b;
 681
 682        for (b = branch_table[hc]; b; b = b->table_next_branch)
 683                if (!strcmp(name, b->name))
 684                        return b;
 685        return NULL;
 686}
 687
 688static struct branch *new_branch(const char *name)
 689{
 690        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 691        struct branch *b = lookup_branch(name);
 692
 693        if (b)
 694                die("Invalid attempt to create duplicate branch: %s", name);
 695        switch (check_ref_format(name)) {
 696        case 0: break; /* its valid */
 697        case CHECK_REF_FORMAT_ONELEVEL:
 698                break; /* valid, but too few '/', allow anyway */
 699        default:
 700                die("Branch name doesn't conform to GIT standards: %s", name);
 701        }
 702
 703        b = pool_calloc(1, sizeof(struct branch));
 704        b->name = pool_strdup(name);
 705        b->table_next_branch = branch_table[hc];
 706        b->branch_tree.versions[0].mode = S_IFDIR;
 707        b->branch_tree.versions[1].mode = S_IFDIR;
 708        b->num_notes = 0;
 709        b->active = 0;
 710        b->pack_id = MAX_PACK_ID;
 711        branch_table[hc] = b;
 712        branch_count++;
 713        return b;
 714}
 715
 716static unsigned int hc_entries(unsigned int cnt)
 717{
 718        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 719        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 720}
 721
 722static struct tree_content *new_tree_content(unsigned int cnt)
 723{
 724        struct avail_tree_content *f, *l = NULL;
 725        struct tree_content *t;
 726        unsigned int hc = hc_entries(cnt);
 727
 728        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 729                if (f->entry_capacity >= cnt)
 730                        break;
 731
 732        if (f) {
 733                if (l)
 734                        l->next_avail = f->next_avail;
 735                else
 736                        avail_tree_table[hc] = f->next_avail;
 737        } else {
 738                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 739                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 740                f->entry_capacity = cnt;
 741        }
 742
 743        t = (struct tree_content*)f;
 744        t->entry_count = 0;
 745        t->delta_depth = 0;
 746        return t;
 747}
 748
 749static void release_tree_entry(struct tree_entry *e);
 750static void release_tree_content(struct tree_content *t)
 751{
 752        struct avail_tree_content *f = (struct avail_tree_content*)t;
 753        unsigned int hc = hc_entries(f->entry_capacity);
 754        f->next_avail = avail_tree_table[hc];
 755        avail_tree_table[hc] = f;
 756}
 757
 758static void release_tree_content_recursive(struct tree_content *t)
 759{
 760        unsigned int i;
 761        for (i = 0; i < t->entry_count; i++)
 762                release_tree_entry(t->entries[i]);
 763        release_tree_content(t);
 764}
 765
 766static struct tree_content *grow_tree_content(
 767        struct tree_content *t,
 768        int amt)
 769{
 770        struct tree_content *r = new_tree_content(t->entry_count + amt);
 771        r->entry_count = t->entry_count;
 772        r->delta_depth = t->delta_depth;
 773        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 774        release_tree_content(t);
 775        return r;
 776}
 777
 778static struct tree_entry *new_tree_entry(void)
 779{
 780        struct tree_entry *e;
 781
 782        if (!avail_tree_entry) {
 783                unsigned int n = tree_entry_alloc;
 784                total_allocd += n * sizeof(struct tree_entry);
 785                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 786                while (n-- > 1) {
 787                        *((void**)e) = e + 1;
 788                        e++;
 789                }
 790                *((void**)e) = NULL;
 791        }
 792
 793        e = avail_tree_entry;
 794        avail_tree_entry = *((void**)e);
 795        return e;
 796}
 797
 798static void release_tree_entry(struct tree_entry *e)
 799{
 800        if (e->tree)
 801                release_tree_content_recursive(e->tree);
 802        *((void**)e) = avail_tree_entry;
 803        avail_tree_entry = e;
 804}
 805
 806static struct tree_content *dup_tree_content(struct tree_content *s)
 807{
 808        struct tree_content *d;
 809        struct tree_entry *a, *b;
 810        unsigned int i;
 811
 812        if (!s)
 813                return NULL;
 814        d = new_tree_content(s->entry_count);
 815        for (i = 0; i < s->entry_count; i++) {
 816                a = s->entries[i];
 817                b = new_tree_entry();
 818                memcpy(b, a, sizeof(*a));
 819                if (a->tree && is_null_sha1(b->versions[1].sha1))
 820                        b->tree = dup_tree_content(a->tree);
 821                else
 822                        b->tree = NULL;
 823                d->entries[i] = b;
 824        }
 825        d->entry_count = s->entry_count;
 826        d->delta_depth = s->delta_depth;
 827
 828        return d;
 829}
 830
 831static void start_packfile(void)
 832{
 833        static char tmpfile[PATH_MAX];
 834        struct packed_git *p;
 835        struct pack_header hdr;
 836        int pack_fd;
 837
 838        pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 839                              "pack/tmp_pack_XXXXXX");
 840        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 841        strcpy(p->pack_name, tmpfile);
 842        p->pack_fd = pack_fd;
 843        pack_file = sha1fd(pack_fd, p->pack_name);
 844
 845        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 846        hdr.hdr_version = htonl(2);
 847        hdr.hdr_entries = 0;
 848        sha1write(pack_file, &hdr, sizeof(hdr));
 849
 850        pack_data = p;
 851        pack_size = sizeof(hdr);
 852        object_count = 0;
 853
 854        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 855        all_packs[pack_id] = p;
 856}
 857
 858static const char *create_index(void)
 859{
 860        const char *tmpfile;
 861        struct pack_idx_entry **idx, **c, **last;
 862        struct object_entry *e;
 863        struct object_entry_pool *o;
 864
 865        /* Build the table of object IDs. */
 866        idx = xmalloc(object_count * sizeof(*idx));
 867        c = idx;
 868        for (o = blocks; o; o = o->next_pool)
 869                for (e = o->next_free; e-- != o->entries;)
 870                        if (pack_id == e->pack_id)
 871                                *c++ = &e->idx;
 872        last = idx + object_count;
 873        if (c != last)
 874                die("internal consistency error creating the index");
 875
 876        tmpfile = write_idx_file(NULL, idx, object_count, pack_data->sha1);
 877        free(idx);
 878        return tmpfile;
 879}
 880
 881static char *keep_pack(const char *curr_index_name)
 882{
 883        static char name[PATH_MAX];
 884        static const char *keep_msg = "fast-import";
 885        int keep_fd;
 886
 887        keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
 888        if (keep_fd < 0)
 889                die_errno("cannot create keep file");
 890        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 891        if (close(keep_fd))
 892                die_errno("failed to write keep file");
 893
 894        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 895                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 896        if (move_temp_to_file(pack_data->pack_name, name))
 897                die("cannot store pack file");
 898
 899        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 900                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 901        if (move_temp_to_file(curr_index_name, name))
 902                die("cannot store index file");
 903        free((void *)curr_index_name);
 904        return name;
 905}
 906
 907static void unkeep_all_packs(void)
 908{
 909        static char name[PATH_MAX];
 910        int k;
 911
 912        for (k = 0; k < pack_id; k++) {
 913                struct packed_git *p = all_packs[k];
 914                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 915                         get_object_directory(), sha1_to_hex(p->sha1));
 916                unlink_or_warn(name);
 917        }
 918}
 919
 920static void end_packfile(void)
 921{
 922        struct packed_git *old_p = pack_data, *new_p;
 923
 924        clear_delta_base_cache();
 925        if (object_count) {
 926                unsigned char cur_pack_sha1[20];
 927                char *idx_name;
 928                int i;
 929                struct branch *b;
 930                struct tag *t;
 931
 932                close_pack_windows(pack_data);
 933                sha1close(pack_file, cur_pack_sha1, 0);
 934                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 935                                    pack_data->pack_name, object_count,
 936                                    cur_pack_sha1, pack_size);
 937                close(pack_data->pack_fd);
 938                idx_name = keep_pack(create_index());
 939
 940                /* Register the packfile with core git's machinery. */
 941                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 942                if (!new_p)
 943                        die("core git rejected index %s", idx_name);
 944                all_packs[pack_id] = new_p;
 945                install_packed_git(new_p);
 946
 947                /* Print the boundary */
 948                if (pack_edges) {
 949                        fprintf(pack_edges, "%s:", new_p->pack_name);
 950                        for (i = 0; i < branch_table_sz; i++) {
 951                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 952                                        if (b->pack_id == pack_id)
 953                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 954                                }
 955                        }
 956                        for (t = first_tag; t; t = t->next_tag) {
 957                                if (t->pack_id == pack_id)
 958                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 959                        }
 960                        fputc('\n', pack_edges);
 961                        fflush(pack_edges);
 962                }
 963
 964                pack_id++;
 965        }
 966        else {
 967                close(old_p->pack_fd);
 968                unlink_or_warn(old_p->pack_name);
 969        }
 970        free(old_p);
 971
 972        /* We can't carry a delta across packfiles. */
 973        strbuf_release(&last_blob.data);
 974        last_blob.offset = 0;
 975        last_blob.depth = 0;
 976}
 977
 978static void cycle_packfile(void)
 979{
 980        end_packfile();
 981        start_packfile();
 982}
 983
 984static int store_object(
 985        enum object_type type,
 986        struct strbuf *dat,
 987        struct last_object *last,
 988        unsigned char *sha1out,
 989        uintmax_t mark)
 990{
 991        void *out, *delta;
 992        struct object_entry *e;
 993        unsigned char hdr[96];
 994        unsigned char sha1[20];
 995        unsigned long hdrlen, deltalen;
 996        git_SHA_CTX c;
 997        z_stream s;
 998
 999        hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1000                (unsigned long)dat->len) + 1;
1001        git_SHA1_Init(&c);
1002        git_SHA1_Update(&c, hdr, hdrlen);
1003        git_SHA1_Update(&c, dat->buf, dat->len);
1004        git_SHA1_Final(sha1, &c);
1005        if (sha1out)
1006                hashcpy(sha1out, sha1);
1007
1008        e = insert_object(sha1);
1009        if (mark)
1010                insert_mark(mark, e);
1011        if (e->idx.offset) {
1012                duplicate_count_by_type[type]++;
1013                return 1;
1014        } else if (find_sha1_pack(sha1, packed_git)) {
1015                e->type = type;
1016                e->pack_id = MAX_PACK_ID;
1017                e->idx.offset = 1; /* just not zero! */
1018                duplicate_count_by_type[type]++;
1019                return 1;
1020        }
1021
1022        if (last && last->data.buf && last->depth < max_depth && dat->len > 20) {
1023                delta = diff_delta(last->data.buf, last->data.len,
1024                        dat->buf, dat->len,
1025                        &deltalen, dat->len - 20);
1026        } else
1027                delta = NULL;
1028
1029        memset(&s, 0, sizeof(s));
1030        deflateInit(&s, pack_compression_level);
1031        if (delta) {
1032                s.next_in = delta;
1033                s.avail_in = deltalen;
1034        } else {
1035                s.next_in = (void *)dat->buf;
1036                s.avail_in = dat->len;
1037        }
1038        s.avail_out = deflateBound(&s, s.avail_in);
1039        s.next_out = out = xmalloc(s.avail_out);
1040        while (deflate(&s, Z_FINISH) == Z_OK)
1041                /* nothing */;
1042        deflateEnd(&s);
1043
1044        /* Determine if we should auto-checkpoint. */
1045        if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
1046                || (pack_size + 60 + s.total_out) < pack_size) {
1047
1048                /* This new object needs to *not* have the current pack_id. */
1049                e->pack_id = pack_id + 1;
1050                cycle_packfile();
1051
1052                /* We cannot carry a delta into the new pack. */
1053                if (delta) {
1054                        free(delta);
1055                        delta = NULL;
1056
1057                        memset(&s, 0, sizeof(s));
1058                        deflateInit(&s, pack_compression_level);
1059                        s.next_in = (void *)dat->buf;
1060                        s.avail_in = dat->len;
1061                        s.avail_out = deflateBound(&s, s.avail_in);
1062                        s.next_out = out = xrealloc(out, s.avail_out);
1063                        while (deflate(&s, Z_FINISH) == Z_OK)
1064                                /* nothing */;
1065                        deflateEnd(&s);
1066                }
1067        }
1068
1069        e->type = type;
1070        e->pack_id = pack_id;
1071        e->idx.offset = pack_size;
1072        object_count++;
1073        object_count_by_type[type]++;
1074
1075        crc32_begin(pack_file);
1076
1077        if (delta) {
1078                off_t ofs = e->idx.offset - last->offset;
1079                unsigned pos = sizeof(hdr) - 1;
1080
1081                delta_count_by_type[type]++;
1082                e->depth = last->depth + 1;
1083
1084                hdrlen = encode_in_pack_object_header(OBJ_OFS_DELTA, deltalen, hdr);
1085                sha1write(pack_file, hdr, hdrlen);
1086                pack_size += hdrlen;
1087
1088                hdr[pos] = ofs & 127;
1089                while (ofs >>= 7)
1090                        hdr[--pos] = 128 | (--ofs & 127);
1091                sha1write(pack_file, hdr + pos, sizeof(hdr) - pos);
1092                pack_size += sizeof(hdr) - pos;
1093        } else {
1094                e->depth = 0;
1095                hdrlen = encode_in_pack_object_header(type, dat->len, hdr);
1096                sha1write(pack_file, hdr, hdrlen);
1097                pack_size += hdrlen;
1098        }
1099
1100        sha1write(pack_file, out, s.total_out);
1101        pack_size += s.total_out;
1102
1103        e->idx.crc32 = crc32_end(pack_file);
1104
1105        free(out);
1106        free(delta);
1107        if (last) {
1108                if (last->no_swap) {
1109                        last->data = *dat;
1110                } else {
1111                        strbuf_swap(&last->data, dat);
1112                }
1113                last->offset = e->idx.offset;
1114                last->depth = e->depth;
1115        }
1116        return 0;
1117}
1118
1119static void truncate_pack(off_t to, git_SHA_CTX *ctx)
1120{
1121        if (ftruncate(pack_data->pack_fd, to)
1122         || lseek(pack_data->pack_fd, to, SEEK_SET) != to)
1123                die_errno("cannot truncate pack to skip duplicate");
1124        pack_size = to;
1125
1126        /* yes this is a layering violation */
1127        pack_file->total = to;
1128        pack_file->offset = 0;
1129        pack_file->ctx = *ctx;
1130}
1131
1132static void stream_blob(uintmax_t len, unsigned char *sha1out, uintmax_t mark)
1133{
1134        size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1135        unsigned char *in_buf = xmalloc(in_sz);
1136        unsigned char *out_buf = xmalloc(out_sz);
1137        struct object_entry *e;
1138        unsigned char sha1[20];
1139        unsigned long hdrlen;
1140        off_t offset;
1141        git_SHA_CTX c;
1142        git_SHA_CTX pack_file_ctx;
1143        z_stream s;
1144        int status = Z_OK;
1145
1146        /* Determine if we should auto-checkpoint. */
1147        if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1148                || (pack_size + 60 + len) < pack_size)
1149                cycle_packfile();
1150
1151        offset = pack_size;
1152
1153        /* preserve the pack_file SHA1 ctx in case we have to truncate later */
1154        sha1flush(pack_file);
1155        pack_file_ctx = pack_file->ctx;
1156
1157        hdrlen = snprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1158        if (out_sz <= hdrlen)
1159                die("impossibly large object header");
1160
1161        git_SHA1_Init(&c);
1162        git_SHA1_Update(&c, out_buf, hdrlen);
1163
1164        crc32_begin(pack_file);
1165
1166        memset(&s, 0, sizeof(s));
1167        deflateInit(&s, pack_compression_level);
1168
1169        hdrlen = encode_in_pack_object_header(OBJ_BLOB, len, out_buf);
1170        if (out_sz <= hdrlen)
1171                die("impossibly large object header");
1172
1173        s.next_out = out_buf + hdrlen;
1174        s.avail_out = out_sz - hdrlen;
1175
1176        while (status != Z_STREAM_END) {
1177                if (0 < len && !s.avail_in) {
1178                        size_t cnt = in_sz < len ? in_sz : (size_t)len;
1179                        size_t n = fread(in_buf, 1, cnt, stdin);
1180                        if (!n && feof(stdin))
1181                                die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1182
1183                        git_SHA1_Update(&c, in_buf, n);
1184                        s.next_in = in_buf;
1185                        s.avail_in = n;
1186                        len -= n;
1187                }
1188
1189                status = deflate(&s, len ? 0 : Z_FINISH);
1190
1191                if (!s.avail_out || status == Z_STREAM_END) {
1192                        size_t n = s.next_out - out_buf;
1193                        sha1write(pack_file, out_buf, n);
1194                        pack_size += n;
1195                        s.next_out = out_buf;
1196                        s.avail_out = out_sz;
1197                }
1198
1199                switch (status) {
1200                case Z_OK:
1201                case Z_BUF_ERROR:
1202                case Z_STREAM_END:
1203                        continue;
1204                default:
1205                        die("unexpected deflate failure: %d", status);
1206                }
1207        }
1208        deflateEnd(&s);
1209        git_SHA1_Final(sha1, &c);
1210
1211        if (sha1out)
1212                hashcpy(sha1out, sha1);
1213
1214        e = insert_object(sha1);
1215
1216        if (mark)
1217                insert_mark(mark, e);
1218
1219        if (e->idx.offset) {
1220                duplicate_count_by_type[OBJ_BLOB]++;
1221                truncate_pack(offset, &pack_file_ctx);
1222
1223        } else if (find_sha1_pack(sha1, packed_git)) {
1224                e->type = OBJ_BLOB;
1225                e->pack_id = MAX_PACK_ID;
1226                e->idx.offset = 1; /* just not zero! */
1227                duplicate_count_by_type[OBJ_BLOB]++;
1228                truncate_pack(offset, &pack_file_ctx);
1229
1230        } else {
1231                e->depth = 0;
1232                e->type = OBJ_BLOB;
1233                e->pack_id = pack_id;
1234                e->idx.offset = offset;
1235                e->idx.crc32 = crc32_end(pack_file);
1236                object_count++;
1237                object_count_by_type[OBJ_BLOB]++;
1238        }
1239
1240        free(in_buf);
1241        free(out_buf);
1242}
1243
1244/* All calls must be guarded by find_object() or find_mark() to
1245 * ensure the 'struct object_entry' passed was written by this
1246 * process instance.  We unpack the entry by the offset, avoiding
1247 * the need for the corresponding .idx file.  This unpacking rule
1248 * works because we only use OBJ_REF_DELTA within the packfiles
1249 * created by fast-import.
1250 *
1251 * oe must not be NULL.  Such an oe usually comes from giving
1252 * an unknown SHA-1 to find_object() or an undefined mark to
1253 * find_mark().  Callers must test for this condition and use
1254 * the standard read_sha1_file() when it happens.
1255 *
1256 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1257 * find_mark(), where the mark was reloaded from an existing marks
1258 * file and is referencing an object that this fast-import process
1259 * instance did not write out to a packfile.  Callers must test for
1260 * this condition and use read_sha1_file() instead.
1261 */
1262static void *gfi_unpack_entry(
1263        struct object_entry *oe,
1264        unsigned long *sizep)
1265{
1266        enum object_type type;
1267        struct packed_git *p = all_packs[oe->pack_id];
1268        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1269                /* The object is stored in the packfile we are writing to
1270                 * and we have modified it since the last time we scanned
1271                 * back to read a previously written object.  If an old
1272                 * window covered [p->pack_size, p->pack_size + 20) its
1273                 * data is stale and is not valid.  Closing all windows
1274                 * and updating the packfile length ensures we can read
1275                 * the newly written data.
1276                 */
1277                close_pack_windows(p);
1278                sha1flush(pack_file);
1279
1280                /* We have to offer 20 bytes additional on the end of
1281                 * the packfile as the core unpacker code assumes the
1282                 * footer is present at the file end and must promise
1283                 * at least 20 bytes within any window it maps.  But
1284                 * we don't actually create the footer here.
1285                 */
1286                p->pack_size = pack_size + 20;
1287        }
1288        return unpack_entry(p, oe->idx.offset, &type, sizep);
1289}
1290
1291static const char *get_mode(const char *str, uint16_t *modep)
1292{
1293        unsigned char c;
1294        uint16_t mode = 0;
1295
1296        while ((c = *str++) != ' ') {
1297                if (c < '0' || c > '7')
1298                        return NULL;
1299                mode = (mode << 3) + (c - '0');
1300        }
1301        *modep = mode;
1302        return str;
1303}
1304
1305static void load_tree(struct tree_entry *root)
1306{
1307        unsigned char *sha1 = root->versions[1].sha1;
1308        struct object_entry *myoe;
1309        struct tree_content *t;
1310        unsigned long size;
1311        char *buf;
1312        const char *c;
1313
1314        root->tree = t = new_tree_content(8);
1315        if (is_null_sha1(sha1))
1316                return;
1317
1318        myoe = find_object(sha1);
1319        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1320                if (myoe->type != OBJ_TREE)
1321                        die("Not a tree: %s", sha1_to_hex(sha1));
1322                t->delta_depth = myoe->depth;
1323                buf = gfi_unpack_entry(myoe, &size);
1324                if (!buf)
1325                        die("Can't load tree %s", sha1_to_hex(sha1));
1326        } else {
1327                enum object_type type;
1328                buf = read_sha1_file(sha1, &type, &size);
1329                if (!buf || type != OBJ_TREE)
1330                        die("Can't load tree %s", sha1_to_hex(sha1));
1331        }
1332
1333        c = buf;
1334        while (c != (buf + size)) {
1335                struct tree_entry *e = new_tree_entry();
1336
1337                if (t->entry_count == t->entry_capacity)
1338                        root->tree = t = grow_tree_content(t, t->entry_count);
1339                t->entries[t->entry_count++] = e;
1340
1341                e->tree = NULL;
1342                c = get_mode(c, &e->versions[1].mode);
1343                if (!c)
1344                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1345                e->versions[0].mode = e->versions[1].mode;
1346                e->name = to_atom(c, strlen(c));
1347                c += e->name->str_len + 1;
1348                hashcpy(e->versions[0].sha1, (unsigned char *)c);
1349                hashcpy(e->versions[1].sha1, (unsigned char *)c);
1350                c += 20;
1351        }
1352        free(buf);
1353}
1354
1355static int tecmp0 (const void *_a, const void *_b)
1356{
1357        struct tree_entry *a = *((struct tree_entry**)_a);
1358        struct tree_entry *b = *((struct tree_entry**)_b);
1359        return base_name_compare(
1360                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1361                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1362}
1363
1364static int tecmp1 (const void *_a, const void *_b)
1365{
1366        struct tree_entry *a = *((struct tree_entry**)_a);
1367        struct tree_entry *b = *((struct tree_entry**)_b);
1368        return base_name_compare(
1369                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1370                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1371}
1372
1373static void mktree(struct tree_content *t, int v, struct strbuf *b)
1374{
1375        size_t maxlen = 0;
1376        unsigned int i;
1377
1378        if (!v)
1379                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1380        else
1381                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1382
1383        for (i = 0; i < t->entry_count; i++) {
1384                if (t->entries[i]->versions[v].mode)
1385                        maxlen += t->entries[i]->name->str_len + 34;
1386        }
1387
1388        strbuf_reset(b);
1389        strbuf_grow(b, maxlen);
1390        for (i = 0; i < t->entry_count; i++) {
1391                struct tree_entry *e = t->entries[i];
1392                if (!e->versions[v].mode)
1393                        continue;
1394                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1395                                        e->name->str_dat, '\0');
1396                strbuf_add(b, e->versions[v].sha1, 20);
1397        }
1398}
1399
1400static void store_tree(struct tree_entry *root)
1401{
1402        struct tree_content *t = root->tree;
1403        unsigned int i, j, del;
1404        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1405        struct object_entry *le;
1406
1407        if (!is_null_sha1(root->versions[1].sha1))
1408                return;
1409
1410        for (i = 0; i < t->entry_count; i++) {
1411                if (t->entries[i]->tree)
1412                        store_tree(t->entries[i]);
1413        }
1414
1415        le = find_object(root->versions[0].sha1);
1416        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1417                mktree(t, 0, &old_tree);
1418                lo.data = old_tree;
1419                lo.offset = le->idx.offset;
1420                lo.depth = t->delta_depth;
1421        }
1422
1423        mktree(t, 1, &new_tree);
1424        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1425
1426        t->delta_depth = lo.depth;
1427        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1428                struct tree_entry *e = t->entries[i];
1429                if (e->versions[1].mode) {
1430                        e->versions[0].mode = e->versions[1].mode;
1431                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1432                        t->entries[j++] = e;
1433                } else {
1434                        release_tree_entry(e);
1435                        del++;
1436                }
1437        }
1438        t->entry_count -= del;
1439}
1440
1441static void tree_content_replace(
1442        struct tree_entry *root,
1443        const unsigned char *sha1,
1444        const uint16_t mode,
1445        struct tree_content *newtree)
1446{
1447        if (!S_ISDIR(mode))
1448                die("Root cannot be a non-directory");
1449        hashcpy(root->versions[1].sha1, sha1);
1450        if (root->tree)
1451                release_tree_content_recursive(root->tree);
1452        root->tree = newtree;
1453}
1454
1455static int tree_content_set(
1456        struct tree_entry *root,
1457        const char *p,
1458        const unsigned char *sha1,
1459        const uint16_t mode,
1460        struct tree_content *subtree)
1461{
1462        struct tree_content *t;
1463        const char *slash1;
1464        unsigned int i, n;
1465        struct tree_entry *e;
1466
1467        slash1 = strchr(p, '/');
1468        if (slash1)
1469                n = slash1 - p;
1470        else
1471                n = strlen(p);
1472        if (!n)
1473                die("Empty path component found in input");
1474        if (!slash1 && !S_ISDIR(mode) && subtree)
1475                die("Non-directories cannot have subtrees");
1476
1477        if (!root->tree)
1478                load_tree(root);
1479        t = root->tree;
1480        for (i = 0; i < t->entry_count; i++) {
1481                e = t->entries[i];
1482                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1483                        if (!slash1) {
1484                                if (!S_ISDIR(mode)
1485                                                && e->versions[1].mode == mode
1486                                                && !hashcmp(e->versions[1].sha1, sha1))
1487                                        return 0;
1488                                e->versions[1].mode = mode;
1489                                hashcpy(e->versions[1].sha1, sha1);
1490                                if (e->tree)
1491                                        release_tree_content_recursive(e->tree);
1492                                e->tree = subtree;
1493                                hashclr(root->versions[1].sha1);
1494                                return 1;
1495                        }
1496                        if (!S_ISDIR(e->versions[1].mode)) {
1497                                e->tree = new_tree_content(8);
1498                                e->versions[1].mode = S_IFDIR;
1499                        }
1500                        if (!e->tree)
1501                                load_tree(e);
1502                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1503                                hashclr(root->versions[1].sha1);
1504                                return 1;
1505                        }
1506                        return 0;
1507                }
1508        }
1509
1510        if (t->entry_count == t->entry_capacity)
1511                root->tree = t = grow_tree_content(t, t->entry_count);
1512        e = new_tree_entry();
1513        e->name = to_atom(p, n);
1514        e->versions[0].mode = 0;
1515        hashclr(e->versions[0].sha1);
1516        t->entries[t->entry_count++] = e;
1517        if (slash1) {
1518                e->tree = new_tree_content(8);
1519                e->versions[1].mode = S_IFDIR;
1520                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1521        } else {
1522                e->tree = subtree;
1523                e->versions[1].mode = mode;
1524                hashcpy(e->versions[1].sha1, sha1);
1525        }
1526        hashclr(root->versions[1].sha1);
1527        return 1;
1528}
1529
1530static int tree_content_remove(
1531        struct tree_entry *root,
1532        const char *p,
1533        struct tree_entry *backup_leaf)
1534{
1535        struct tree_content *t;
1536        const char *slash1;
1537        unsigned int i, n;
1538        struct tree_entry *e;
1539
1540        slash1 = strchr(p, '/');
1541        if (slash1)
1542                n = slash1 - p;
1543        else
1544                n = strlen(p);
1545
1546        if (!root->tree)
1547                load_tree(root);
1548        t = root->tree;
1549        for (i = 0; i < t->entry_count; i++) {
1550                e = t->entries[i];
1551                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1552                        if (slash1 && !S_ISDIR(e->versions[1].mode))
1553                                /*
1554                                 * If p names a file in some subdirectory, and a
1555                                 * file or symlink matching the name of the
1556                                 * parent directory of p exists, then p cannot
1557                                 * exist and need not be deleted.
1558                                 */
1559                                return 1;
1560                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1561                                goto del_entry;
1562                        if (!e->tree)
1563                                load_tree(e);
1564                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1565                                for (n = 0; n < e->tree->entry_count; n++) {
1566                                        if (e->tree->entries[n]->versions[1].mode) {
1567                                                hashclr(root->versions[1].sha1);
1568                                                return 1;
1569                                        }
1570                                }
1571                                backup_leaf = NULL;
1572                                goto del_entry;
1573                        }
1574                        return 0;
1575                }
1576        }
1577        return 0;
1578
1579del_entry:
1580        if (backup_leaf)
1581                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1582        else if (e->tree)
1583                release_tree_content_recursive(e->tree);
1584        e->tree = NULL;
1585        e->versions[1].mode = 0;
1586        hashclr(e->versions[1].sha1);
1587        hashclr(root->versions[1].sha1);
1588        return 1;
1589}
1590
1591static int tree_content_get(
1592        struct tree_entry *root,
1593        const char *p,
1594        struct tree_entry *leaf)
1595{
1596        struct tree_content *t;
1597        const char *slash1;
1598        unsigned int i, n;
1599        struct tree_entry *e;
1600
1601        slash1 = strchr(p, '/');
1602        if (slash1)
1603                n = slash1 - p;
1604        else
1605                n = strlen(p);
1606
1607        if (!root->tree)
1608                load_tree(root);
1609        t = root->tree;
1610        for (i = 0; i < t->entry_count; i++) {
1611                e = t->entries[i];
1612                if (e->name->str_len == n && !strncmp_icase(p, e->name->str_dat, n)) {
1613                        if (!slash1) {
1614                                memcpy(leaf, e, sizeof(*leaf));
1615                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1616                                        leaf->tree = dup_tree_content(e->tree);
1617                                else
1618                                        leaf->tree = NULL;
1619                                return 1;
1620                        }
1621                        if (!S_ISDIR(e->versions[1].mode))
1622                                return 0;
1623                        if (!e->tree)
1624                                load_tree(e);
1625                        return tree_content_get(e, slash1 + 1, leaf);
1626                }
1627        }
1628        return 0;
1629}
1630
1631static int update_branch(struct branch *b)
1632{
1633        static const char *msg = "fast-import";
1634        struct ref_lock *lock;
1635        unsigned char old_sha1[20];
1636
1637        if (is_null_sha1(b->sha1))
1638                return 0;
1639        if (read_ref(b->name, old_sha1))
1640                hashclr(old_sha1);
1641        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1642        if (!lock)
1643                return error("Unable to lock %s", b->name);
1644        if (!force_update && !is_null_sha1(old_sha1)) {
1645                struct commit *old_cmit, *new_cmit;
1646
1647                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1648                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1649                if (!old_cmit || !new_cmit) {
1650                        unlock_ref(lock);
1651                        return error("Branch %s is missing commits.", b->name);
1652                }
1653
1654                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1655                        unlock_ref(lock);
1656                        warning("Not updating %s"
1657                                " (new tip %s does not contain %s)",
1658                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1659                        return -1;
1660                }
1661        }
1662        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1663                return error("Unable to update %s", b->name);
1664        return 0;
1665}
1666
1667static void dump_branches(void)
1668{
1669        unsigned int i;
1670        struct branch *b;
1671
1672        for (i = 0; i < branch_table_sz; i++) {
1673                for (b = branch_table[i]; b; b = b->table_next_branch)
1674                        failure |= update_branch(b);
1675        }
1676}
1677
1678static void dump_tags(void)
1679{
1680        static const char *msg = "fast-import";
1681        struct tag *t;
1682        struct ref_lock *lock;
1683        char ref_name[PATH_MAX];
1684
1685        for (t = first_tag; t; t = t->next_tag) {
1686                sprintf(ref_name, "tags/%s", t->name);
1687                lock = lock_ref_sha1(ref_name, NULL);
1688                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1689                        failure |= error("Unable to update %s", ref_name);
1690        }
1691}
1692
1693static void dump_marks_helper(FILE *f,
1694        uintmax_t base,
1695        struct mark_set *m)
1696{
1697        uintmax_t k;
1698        if (m->shift) {
1699                for (k = 0; k < 1024; k++) {
1700                        if (m->data.sets[k])
1701                                dump_marks_helper(f, base + (k << m->shift),
1702                                        m->data.sets[k]);
1703                }
1704        } else {
1705                for (k = 0; k < 1024; k++) {
1706                        if (m->data.marked[k])
1707                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1708                                        sha1_to_hex(m->data.marked[k]->idx.sha1));
1709                }
1710        }
1711}
1712
1713static void dump_marks(void)
1714{
1715        static struct lock_file mark_lock;
1716        int mark_fd;
1717        FILE *f;
1718
1719        if (!export_marks_file)
1720                return;
1721
1722        mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1723        if (mark_fd < 0) {
1724                failure |= error("Unable to write marks file %s: %s",
1725                        export_marks_file, strerror(errno));
1726                return;
1727        }
1728
1729        f = fdopen(mark_fd, "w");
1730        if (!f) {
1731                int saved_errno = errno;
1732                rollback_lock_file(&mark_lock);
1733                failure |= error("Unable to write marks file %s: %s",
1734                        export_marks_file, strerror(saved_errno));
1735                return;
1736        }
1737
1738        /*
1739         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1740         * Assign -1 to the lock file descriptor so that commit_lock_file()
1741         * won't try to close() it.
1742         */
1743        mark_lock.fd = -1;
1744
1745        dump_marks_helper(f, 0, marks);
1746        if (ferror(f) || fclose(f)) {
1747                int saved_errno = errno;
1748                rollback_lock_file(&mark_lock);
1749                failure |= error("Unable to write marks file %s: %s",
1750                        export_marks_file, strerror(saved_errno));
1751                return;
1752        }
1753
1754        if (commit_lock_file(&mark_lock)) {
1755                int saved_errno = errno;
1756                rollback_lock_file(&mark_lock);
1757                failure |= error("Unable to commit marks file %s: %s",
1758                        export_marks_file, strerror(saved_errno));
1759                return;
1760        }
1761}
1762
1763static void read_marks(void)
1764{
1765        char line[512];
1766        FILE *f = fopen(import_marks_file, "r");
1767        if (!f)
1768                die_errno("cannot read '%s'", import_marks_file);
1769        while (fgets(line, sizeof(line), f)) {
1770                uintmax_t mark;
1771                char *end;
1772                unsigned char sha1[20];
1773                struct object_entry *e;
1774
1775                end = strchr(line, '\n');
1776                if (line[0] != ':' || !end)
1777                        die("corrupt mark line: %s", line);
1778                *end = 0;
1779                mark = strtoumax(line + 1, &end, 10);
1780                if (!mark || end == line + 1
1781                        || *end != ' ' || get_sha1(end + 1, sha1))
1782                        die("corrupt mark line: %s", line);
1783                e = find_object(sha1);
1784                if (!e) {
1785                        enum object_type type = sha1_object_info(sha1, NULL);
1786                        if (type < 0)
1787                                die("object not found: %s", sha1_to_hex(sha1));
1788                        e = insert_object(sha1);
1789                        e->type = type;
1790                        e->pack_id = MAX_PACK_ID;
1791                        e->idx.offset = 1; /* just not zero! */
1792                }
1793                insert_mark(mark, e);
1794        }
1795        fclose(f);
1796}
1797
1798
1799static int read_next_command(void)
1800{
1801        static int stdin_eof = 0;
1802
1803        if (stdin_eof) {
1804                unread_command_buf = 0;
1805                return EOF;
1806        }
1807
1808        do {
1809                if (unread_command_buf) {
1810                        unread_command_buf = 0;
1811                } else {
1812                        struct recent_command *rc;
1813
1814                        strbuf_detach(&command_buf, NULL);
1815                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1816                        if (stdin_eof)
1817                                return EOF;
1818
1819                        if (!seen_data_command
1820                                && prefixcmp(command_buf.buf, "feature ")
1821                                && prefixcmp(command_buf.buf, "option ")) {
1822                                parse_argv();
1823                        }
1824
1825                        rc = rc_free;
1826                        if (rc)
1827                                rc_free = rc->next;
1828                        else {
1829                                rc = cmd_hist.next;
1830                                cmd_hist.next = rc->next;
1831                                cmd_hist.next->prev = &cmd_hist;
1832                                free(rc->buf);
1833                        }
1834
1835                        rc->buf = command_buf.buf;
1836                        rc->prev = cmd_tail;
1837                        rc->next = cmd_hist.prev;
1838                        rc->prev->next = rc;
1839                        cmd_tail = rc;
1840                }
1841        } while (command_buf.buf[0] == '#');
1842
1843        return 0;
1844}
1845
1846static void skip_optional_lf(void)
1847{
1848        int term_char = fgetc(stdin);
1849        if (term_char != '\n' && term_char != EOF)
1850                ungetc(term_char, stdin);
1851}
1852
1853static void parse_mark(void)
1854{
1855        if (!prefixcmp(command_buf.buf, "mark :")) {
1856                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1857                read_next_command();
1858        }
1859        else
1860                next_mark = 0;
1861}
1862
1863static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1864{
1865        strbuf_reset(sb);
1866
1867        if (prefixcmp(command_buf.buf, "data "))
1868                die("Expected 'data n' command, found: %s", command_buf.buf);
1869
1870        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1871                char *term = xstrdup(command_buf.buf + 5 + 2);
1872                size_t term_len = command_buf.len - 5 - 2;
1873
1874                strbuf_detach(&command_buf, NULL);
1875                for (;;) {
1876                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1877                                die("EOF in data (terminator '%s' not found)", term);
1878                        if (term_len == command_buf.len
1879                                && !strcmp(term, command_buf.buf))
1880                                break;
1881                        strbuf_addbuf(sb, &command_buf);
1882                        strbuf_addch(sb, '\n');
1883                }
1884                free(term);
1885        }
1886        else {
1887                uintmax_t len = strtoumax(command_buf.buf + 5, NULL, 10);
1888                size_t n = 0, length = (size_t)len;
1889
1890                if (limit && limit < len) {
1891                        *len_res = len;
1892                        return 0;
1893                }
1894                if (length < len)
1895                        die("data is too large to use in this context");
1896
1897                while (n < length) {
1898                        size_t s = strbuf_fread(sb, length - n, stdin);
1899                        if (!s && feof(stdin))
1900                                die("EOF in data (%lu bytes remaining)",
1901                                        (unsigned long)(length - n));
1902                        n += s;
1903                }
1904        }
1905
1906        skip_optional_lf();
1907        return 1;
1908}
1909
1910static int validate_raw_date(const char *src, char *result, int maxlen)
1911{
1912        const char *orig_src = src;
1913        char *endp;
1914        unsigned long num;
1915
1916        errno = 0;
1917
1918        num = strtoul(src, &endp, 10);
1919        /* NEEDSWORK: perhaps check for reasonable values? */
1920        if (errno || endp == src || *endp != ' ')
1921                return -1;
1922
1923        src = endp + 1;
1924        if (*src != '-' && *src != '+')
1925                return -1;
1926
1927        num = strtoul(src + 1, &endp, 10);
1928        if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1929            1400 < num)
1930                return -1;
1931
1932        strcpy(result, orig_src);
1933        return 0;
1934}
1935
1936static char *parse_ident(const char *buf)
1937{
1938        const char *gt;
1939        size_t name_len;
1940        char *ident;
1941
1942        gt = strrchr(buf, '>');
1943        if (!gt)
1944                die("Missing > in ident string: %s", buf);
1945        gt++;
1946        if (*gt != ' ')
1947                die("Missing space after > in ident string: %s", buf);
1948        gt++;
1949        name_len = gt - buf;
1950        ident = xmalloc(name_len + 24);
1951        strncpy(ident, buf, name_len);
1952
1953        switch (whenspec) {
1954        case WHENSPEC_RAW:
1955                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1956                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1957                break;
1958        case WHENSPEC_RFC2822:
1959                if (parse_date(gt, ident + name_len, 24) < 0)
1960                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1961                break;
1962        case WHENSPEC_NOW:
1963                if (strcmp("now", gt))
1964                        die("Date in ident must be 'now': %s", buf);
1965                datestamp(ident + name_len, 24);
1966                break;
1967        }
1968
1969        return ident;
1970}
1971
1972static void parse_and_store_blob(
1973        struct last_object *last,
1974        unsigned char *sha1out,
1975        uintmax_t mark)
1976{
1977        static struct strbuf buf = STRBUF_INIT;
1978        uintmax_t len;
1979
1980        if (parse_data(&buf, big_file_threshold, &len))
1981                store_object(OBJ_BLOB, &buf, last, sha1out, mark);
1982        else {
1983                if (last) {
1984                        strbuf_release(&last->data);
1985                        last->offset = 0;
1986                        last->depth = 0;
1987                }
1988                stream_blob(len, sha1out, mark);
1989                skip_optional_lf();
1990        }
1991}
1992
1993static void parse_new_blob(void)
1994{
1995        read_next_command();
1996        parse_mark();
1997        parse_and_store_blob(&last_blob, NULL, next_mark);
1998}
1999
2000static void unload_one_branch(void)
2001{
2002        while (cur_active_branches
2003                && cur_active_branches >= max_active_branches) {
2004                uintmax_t min_commit = ULONG_MAX;
2005                struct branch *e, *l = NULL, *p = NULL;
2006
2007                for (e = active_branches; e; e = e->active_next_branch) {
2008                        if (e->last_commit < min_commit) {
2009                                p = l;
2010                                min_commit = e->last_commit;
2011                        }
2012                        l = e;
2013                }
2014
2015                if (p) {
2016                        e = p->active_next_branch;
2017                        p->active_next_branch = e->active_next_branch;
2018                } else {
2019                        e = active_branches;
2020                        active_branches = e->active_next_branch;
2021                }
2022                e->active = 0;
2023                e->active_next_branch = NULL;
2024                if (e->branch_tree.tree) {
2025                        release_tree_content_recursive(e->branch_tree.tree);
2026                        e->branch_tree.tree = NULL;
2027                }
2028                cur_active_branches--;
2029        }
2030}
2031
2032static void load_branch(struct branch *b)
2033{
2034        load_tree(&b->branch_tree);
2035        if (!b->active) {
2036                b->active = 1;
2037                b->active_next_branch = active_branches;
2038                active_branches = b;
2039                cur_active_branches++;
2040                branch_load_count++;
2041        }
2042}
2043
2044static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2045{
2046        unsigned char fanout = 0;
2047        while ((num_notes >>= 8))
2048                fanout++;
2049        return fanout;
2050}
2051
2052static void construct_path_with_fanout(const char *hex_sha1,
2053                unsigned char fanout, char *path)
2054{
2055        unsigned int i = 0, j = 0;
2056        if (fanout >= 20)
2057                die("Too large fanout (%u)", fanout);
2058        while (fanout) {
2059                path[i++] = hex_sha1[j++];
2060                path[i++] = hex_sha1[j++];
2061                path[i++] = '/';
2062                fanout--;
2063        }
2064        memcpy(path + i, hex_sha1 + j, 40 - j);
2065        path[i + 40 - j] = '\0';
2066}
2067
2068static uintmax_t do_change_note_fanout(
2069                struct tree_entry *orig_root, struct tree_entry *root,
2070                char *hex_sha1, unsigned int hex_sha1_len,
2071                char *fullpath, unsigned int fullpath_len,
2072                unsigned char fanout)
2073{
2074        struct tree_content *t = root->tree;
2075        struct tree_entry *e, leaf;
2076        unsigned int i, tmp_hex_sha1_len, tmp_fullpath_len;
2077        uintmax_t num_notes = 0;
2078        unsigned char sha1[20];
2079        char realpath[60];
2080
2081        for (i = 0; t && i < t->entry_count; i++) {
2082                e = t->entries[i];
2083                tmp_hex_sha1_len = hex_sha1_len + e->name->str_len;
2084                tmp_fullpath_len = fullpath_len;
2085
2086                /*
2087                 * We're interested in EITHER existing note entries (entries
2088                 * with exactly 40 hex chars in path, not including directory
2089                 * separators), OR directory entries that may contain note
2090                 * entries (with < 40 hex chars in path).
2091                 * Also, each path component in a note entry must be a multiple
2092                 * of 2 chars.
2093                 */
2094                if (!e->versions[1].mode ||
2095                    tmp_hex_sha1_len > 40 ||
2096                    e->name->str_len % 2)
2097                        continue;
2098
2099                /* This _may_ be a note entry, or a subdir containing notes */
2100                memcpy(hex_sha1 + hex_sha1_len, e->name->str_dat,
2101                       e->name->str_len);
2102                if (tmp_fullpath_len)
2103                        fullpath[tmp_fullpath_len++] = '/';
2104                memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2105                       e->name->str_len);
2106                tmp_fullpath_len += e->name->str_len;
2107                fullpath[tmp_fullpath_len] = '\0';
2108
2109                if (tmp_hex_sha1_len == 40 && !get_sha1_hex(hex_sha1, sha1)) {
2110                        /* This is a note entry */
2111                        construct_path_with_fanout(hex_sha1, fanout, realpath);
2112                        if (!strcmp(fullpath, realpath)) {
2113                                /* Note entry is in correct location */
2114                                num_notes++;
2115                                continue;
2116                        }
2117
2118                        /* Rename fullpath to realpath */
2119                        if (!tree_content_remove(orig_root, fullpath, &leaf))
2120                                die("Failed to remove path %s", fullpath);
2121                        tree_content_set(orig_root, realpath,
2122                                leaf.versions[1].sha1,
2123                                leaf.versions[1].mode,
2124                                leaf.tree);
2125                } else if (S_ISDIR(e->versions[1].mode)) {
2126                        /* This is a subdir that may contain note entries */
2127                        if (!e->tree)
2128                                load_tree(e);
2129                        num_notes += do_change_note_fanout(orig_root, e,
2130                                hex_sha1, tmp_hex_sha1_len,
2131                                fullpath, tmp_fullpath_len, fanout);
2132                }
2133
2134                /* The above may have reallocated the current tree_content */
2135                t = root->tree;
2136        }
2137        return num_notes;
2138}
2139
2140static uintmax_t change_note_fanout(struct tree_entry *root,
2141                unsigned char fanout)
2142{
2143        char hex_sha1[40], path[60];
2144        return do_change_note_fanout(root, root, hex_sha1, 0, path, 0, fanout);
2145}
2146
2147static void file_change_m(struct branch *b)
2148{
2149        const char *p = command_buf.buf + 2;
2150        static struct strbuf uq = STRBUF_INIT;
2151        const char *endp;
2152        struct object_entry *oe = oe;
2153        unsigned char sha1[20];
2154        uint16_t mode, inline_data = 0;
2155
2156        p = get_mode(p, &mode);
2157        if (!p)
2158                die("Corrupt mode: %s", command_buf.buf);
2159        switch (mode) {
2160        case 0644:
2161        case 0755:
2162                mode |= S_IFREG;
2163        case S_IFREG | 0644:
2164        case S_IFREG | 0755:
2165        case S_IFLNK:
2166        case S_IFDIR:
2167        case S_IFGITLINK:
2168                /* ok */
2169                break;
2170        default:
2171                die("Corrupt mode: %s", command_buf.buf);
2172        }
2173
2174        if (*p == ':') {
2175                char *x;
2176                oe = find_mark(strtoumax(p + 1, &x, 10));
2177                hashcpy(sha1, oe->idx.sha1);
2178                p = x;
2179        } else if (!prefixcmp(p, "inline")) {
2180                inline_data = 1;
2181                p += 6;
2182        } else {
2183                if (get_sha1_hex(p, sha1))
2184                        die("Invalid SHA1: %s", command_buf.buf);
2185                oe = find_object(sha1);
2186                p += 40;
2187        }
2188        if (*p++ != ' ')
2189                die("Missing space after SHA1: %s", command_buf.buf);
2190
2191        strbuf_reset(&uq);
2192        if (!unquote_c_style(&uq, p, &endp)) {
2193                if (*endp)
2194                        die("Garbage after path in: %s", command_buf.buf);
2195                p = uq.buf;
2196        }
2197
2198        if (S_ISGITLINK(mode)) {
2199                if (inline_data)
2200                        die("Git links cannot be specified 'inline': %s",
2201                                command_buf.buf);
2202                else if (oe) {
2203                        if (oe->type != OBJ_COMMIT)
2204                                die("Not a commit (actually a %s): %s",
2205                                        typename(oe->type), command_buf.buf);
2206                }
2207                /*
2208                 * Accept the sha1 without checking; it expected to be in
2209                 * another repository.
2210                 */
2211        } else if (inline_data) {
2212                if (S_ISDIR(mode))
2213                        die("Directories cannot be specified 'inline': %s",
2214                                command_buf.buf);
2215                if (p != uq.buf) {
2216                        strbuf_addstr(&uq, p);
2217                        p = uq.buf;
2218                }
2219                read_next_command();
2220                parse_and_store_blob(&last_blob, sha1, 0);
2221        } else {
2222                enum object_type expected = S_ISDIR(mode) ?
2223                                                OBJ_TREE: OBJ_BLOB;
2224                enum object_type type = oe ? oe->type :
2225                                        sha1_object_info(sha1, NULL);
2226                if (type < 0)
2227                        die("%s not found: %s",
2228                                        S_ISDIR(mode) ?  "Tree" : "Blob",
2229                                        command_buf.buf);
2230                if (type != expected)
2231                        die("Not a %s (actually a %s): %s",
2232                                typename(expected), typename(type),
2233                                command_buf.buf);
2234        }
2235
2236        if (!*p) {
2237                tree_content_replace(&b->branch_tree, sha1, mode, NULL);
2238                return;
2239        }
2240        tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
2241}
2242
2243static void file_change_d(struct branch *b)
2244{
2245        const char *p = command_buf.buf + 2;
2246        static struct strbuf uq = STRBUF_INIT;
2247        const char *endp;
2248
2249        strbuf_reset(&uq);
2250        if (!unquote_c_style(&uq, p, &endp)) {
2251                if (*endp)
2252                        die("Garbage after path in: %s", command_buf.buf);
2253                p = uq.buf;
2254        }
2255        tree_content_remove(&b->branch_tree, p, NULL);
2256}
2257
2258static void file_change_cr(struct branch *b, int rename)
2259{
2260        const char *s, *d;
2261        static struct strbuf s_uq = STRBUF_INIT;
2262        static struct strbuf d_uq = STRBUF_INIT;
2263        const char *endp;
2264        struct tree_entry leaf;
2265
2266        s = command_buf.buf + 2;
2267        strbuf_reset(&s_uq);
2268        if (!unquote_c_style(&s_uq, s, &endp)) {
2269                if (*endp != ' ')
2270                        die("Missing space after source: %s", command_buf.buf);
2271        } else {
2272                endp = strchr(s, ' ');
2273                if (!endp)
2274                        die("Missing space after source: %s", command_buf.buf);
2275                strbuf_add(&s_uq, s, endp - s);
2276        }
2277        s = s_uq.buf;
2278
2279        endp++;
2280        if (!*endp)
2281                die("Missing dest: %s", command_buf.buf);
2282
2283        d = endp;
2284        strbuf_reset(&d_uq);
2285        if (!unquote_c_style(&d_uq, d, &endp)) {
2286                if (*endp)
2287                        die("Garbage after dest in: %s", command_buf.buf);
2288                d = d_uq.buf;
2289        }
2290
2291        memset(&leaf, 0, sizeof(leaf));
2292        if (rename)
2293                tree_content_remove(&b->branch_tree, s, &leaf);
2294        else
2295                tree_content_get(&b->branch_tree, s, &leaf);
2296        if (!leaf.versions[1].mode)
2297                die("Path %s not in branch", s);
2298        if (!*d) {      /* C "path/to/subdir" "" */
2299                tree_content_replace(&b->branch_tree,
2300                        leaf.versions[1].sha1,
2301                        leaf.versions[1].mode,
2302                        leaf.tree);
2303                return;
2304        }
2305        tree_content_set(&b->branch_tree, d,
2306                leaf.versions[1].sha1,
2307                leaf.versions[1].mode,
2308                leaf.tree);
2309}
2310
2311static void note_change_n(struct branch *b, unsigned char old_fanout)
2312{
2313        const char *p = command_buf.buf + 2;
2314        static struct strbuf uq = STRBUF_INIT;
2315        struct object_entry *oe = oe;
2316        struct branch *s;
2317        unsigned char sha1[20], commit_sha1[20];
2318        char path[60];
2319        uint16_t inline_data = 0;
2320        unsigned char new_fanout;
2321
2322        /* <dataref> or 'inline' */
2323        if (*p == ':') {
2324                char *x;
2325                oe = find_mark(strtoumax(p + 1, &x, 10));
2326                hashcpy(sha1, oe->idx.sha1);
2327                p = x;
2328        } else if (!prefixcmp(p, "inline")) {
2329                inline_data = 1;
2330                p += 6;
2331        } else {
2332                if (get_sha1_hex(p, sha1))
2333                        die("Invalid SHA1: %s", command_buf.buf);
2334                oe = find_object(sha1);
2335                p += 40;
2336        }
2337        if (*p++ != ' ')
2338                die("Missing space after SHA1: %s", command_buf.buf);
2339
2340        /* <committish> */
2341        s = lookup_branch(p);
2342        if (s) {
2343                hashcpy(commit_sha1, s->sha1);
2344        } else if (*p == ':') {
2345                uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2346                struct object_entry *commit_oe = find_mark(commit_mark);
2347                if (commit_oe->type != OBJ_COMMIT)
2348                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2349                hashcpy(commit_sha1, commit_oe->idx.sha1);
2350        } else if (!get_sha1(p, commit_sha1)) {
2351                unsigned long size;
2352                char *buf = read_object_with_reference(commit_sha1,
2353                        commit_type, &size, commit_sha1);
2354                if (!buf || size < 46)
2355                        die("Not a valid commit: %s", p);
2356                free(buf);
2357        } else
2358                die("Invalid ref name or SHA1 expression: %s", p);
2359
2360        if (inline_data) {
2361                if (p != uq.buf) {
2362                        strbuf_addstr(&uq, p);
2363                        p = uq.buf;
2364                }
2365                read_next_command();
2366                parse_and_store_blob(&last_blob, sha1, 0);
2367        } else if (oe) {
2368                if (oe->type != OBJ_BLOB)
2369                        die("Not a blob (actually a %s): %s",
2370                                typename(oe->type), command_buf.buf);
2371        } else if (!is_null_sha1(sha1)) {
2372                enum object_type type = sha1_object_info(sha1, NULL);
2373                if (type < 0)
2374                        die("Blob not found: %s", command_buf.buf);
2375                if (type != OBJ_BLOB)
2376                        die("Not a blob (actually a %s): %s",
2377                            typename(type), command_buf.buf);
2378        }
2379
2380        construct_path_with_fanout(sha1_to_hex(commit_sha1), old_fanout, path);
2381        if (tree_content_remove(&b->branch_tree, path, NULL))
2382                b->num_notes--;
2383
2384        if (is_null_sha1(sha1))
2385                return; /* nothing to insert */
2386
2387        b->num_notes++;
2388        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2389        construct_path_with_fanout(sha1_to_hex(commit_sha1), new_fanout, path);
2390        tree_content_set(&b->branch_tree, path, sha1, S_IFREG | 0644, NULL);
2391}
2392
2393static void file_change_deleteall(struct branch *b)
2394{
2395        release_tree_content_recursive(b->branch_tree.tree);
2396        hashclr(b->branch_tree.versions[0].sha1);
2397        hashclr(b->branch_tree.versions[1].sha1);
2398        load_tree(&b->branch_tree);
2399        b->num_notes = 0;
2400}
2401
2402static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2403{
2404        if (!buf || size < 46)
2405                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2406        if (memcmp("tree ", buf, 5)
2407                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2408                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2409        hashcpy(b->branch_tree.versions[0].sha1,
2410                b->branch_tree.versions[1].sha1);
2411}
2412
2413static void parse_from_existing(struct branch *b)
2414{
2415        if (is_null_sha1(b->sha1)) {
2416                hashclr(b->branch_tree.versions[0].sha1);
2417                hashclr(b->branch_tree.versions[1].sha1);
2418        } else {
2419                unsigned long size;
2420                char *buf;
2421
2422                buf = read_object_with_reference(b->sha1,
2423                        commit_type, &size, b->sha1);
2424                parse_from_commit(b, buf, size);
2425                free(buf);
2426        }
2427}
2428
2429static int parse_from(struct branch *b)
2430{
2431        const char *from;
2432        struct branch *s;
2433
2434        if (prefixcmp(command_buf.buf, "from "))
2435                return 0;
2436
2437        if (b->branch_tree.tree) {
2438                release_tree_content_recursive(b->branch_tree.tree);
2439                b->branch_tree.tree = NULL;
2440        }
2441
2442        from = strchr(command_buf.buf, ' ') + 1;
2443        s = lookup_branch(from);
2444        if (b == s)
2445                die("Can't create a branch from itself: %s", b->name);
2446        else if (s) {
2447                unsigned char *t = s->branch_tree.versions[1].sha1;
2448                hashcpy(b->sha1, s->sha1);
2449                hashcpy(b->branch_tree.versions[0].sha1, t);
2450                hashcpy(b->branch_tree.versions[1].sha1, t);
2451        } else if (*from == ':') {
2452                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2453                struct object_entry *oe = find_mark(idnum);
2454                if (oe->type != OBJ_COMMIT)
2455                        die("Mark :%" PRIuMAX " not a commit", idnum);
2456                hashcpy(b->sha1, oe->idx.sha1);
2457                if (oe->pack_id != MAX_PACK_ID) {
2458                        unsigned long size;
2459                        char *buf = gfi_unpack_entry(oe, &size);
2460                        parse_from_commit(b, buf, size);
2461                        free(buf);
2462                } else
2463                        parse_from_existing(b);
2464        } else if (!get_sha1(from, b->sha1))
2465                parse_from_existing(b);
2466        else
2467                die("Invalid ref name or SHA1 expression: %s", from);
2468
2469        read_next_command();
2470        return 1;
2471}
2472
2473static struct hash_list *parse_merge(unsigned int *count)
2474{
2475        struct hash_list *list = NULL, *n, *e = e;
2476        const char *from;
2477        struct branch *s;
2478
2479        *count = 0;
2480        while (!prefixcmp(command_buf.buf, "merge ")) {
2481                from = strchr(command_buf.buf, ' ') + 1;
2482                n = xmalloc(sizeof(*n));
2483                s = lookup_branch(from);
2484                if (s)
2485                        hashcpy(n->sha1, s->sha1);
2486                else if (*from == ':') {
2487                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2488                        struct object_entry *oe = find_mark(idnum);
2489                        if (oe->type != OBJ_COMMIT)
2490                                die("Mark :%" PRIuMAX " not a commit", idnum);
2491                        hashcpy(n->sha1, oe->idx.sha1);
2492                } else if (!get_sha1(from, n->sha1)) {
2493                        unsigned long size;
2494                        char *buf = read_object_with_reference(n->sha1,
2495                                commit_type, &size, n->sha1);
2496                        if (!buf || size < 46)
2497                                die("Not a valid commit: %s", from);
2498                        free(buf);
2499                } else
2500                        die("Invalid ref name or SHA1 expression: %s", from);
2501
2502                n->next = NULL;
2503                if (list)
2504                        e->next = n;
2505                else
2506                        list = n;
2507                e = n;
2508                (*count)++;
2509                read_next_command();
2510        }
2511        return list;
2512}
2513
2514static void parse_new_commit(void)
2515{
2516        static struct strbuf msg = STRBUF_INIT;
2517        struct branch *b;
2518        char *sp;
2519        char *author = NULL;
2520        char *committer = NULL;
2521        struct hash_list *merge_list = NULL;
2522        unsigned int merge_count;
2523        unsigned char prev_fanout, new_fanout;
2524
2525        /* Obtain the branch name from the rest of our command */
2526        sp = strchr(command_buf.buf, ' ') + 1;
2527        b = lookup_branch(sp);
2528        if (!b)
2529                b = new_branch(sp);
2530
2531        read_next_command();
2532        parse_mark();
2533        if (!prefixcmp(command_buf.buf, "author ")) {
2534                author = parse_ident(command_buf.buf + 7);
2535                read_next_command();
2536        }
2537        if (!prefixcmp(command_buf.buf, "committer ")) {
2538                committer = parse_ident(command_buf.buf + 10);
2539                read_next_command();
2540        }
2541        if (!committer)
2542                die("Expected committer but didn't get one");
2543        parse_data(&msg, 0, NULL);
2544        read_next_command();
2545        parse_from(b);
2546        merge_list = parse_merge(&merge_count);
2547
2548        /* ensure the branch is active/loaded */
2549        if (!b->branch_tree.tree || !max_active_branches) {
2550                unload_one_branch();
2551                load_branch(b);
2552        }
2553
2554        prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2555
2556        /* file_change* */
2557        while (command_buf.len > 0) {
2558                if (!prefixcmp(command_buf.buf, "M "))
2559                        file_change_m(b);
2560                else if (!prefixcmp(command_buf.buf, "D "))
2561                        file_change_d(b);
2562                else if (!prefixcmp(command_buf.buf, "R "))
2563                        file_change_cr(b, 1);
2564                else if (!prefixcmp(command_buf.buf, "C "))
2565                        file_change_cr(b, 0);
2566                else if (!prefixcmp(command_buf.buf, "N "))
2567                        note_change_n(b, prev_fanout);
2568                else if (!strcmp("deleteall", command_buf.buf))
2569                        file_change_deleteall(b);
2570                else {
2571                        unread_command_buf = 1;
2572                        break;
2573                }
2574                if (read_next_command() == EOF)
2575                        break;
2576        }
2577
2578        new_fanout = convert_num_notes_to_fanout(b->num_notes);
2579        if (new_fanout != prev_fanout)
2580                b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2581
2582        /* build the tree and the commit */
2583        store_tree(&b->branch_tree);
2584        hashcpy(b->branch_tree.versions[0].sha1,
2585                b->branch_tree.versions[1].sha1);
2586
2587        strbuf_reset(&new_data);
2588        strbuf_addf(&new_data, "tree %s\n",
2589                sha1_to_hex(b->branch_tree.versions[1].sha1));
2590        if (!is_null_sha1(b->sha1))
2591                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2592        while (merge_list) {
2593                struct hash_list *next = merge_list->next;
2594                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2595                free(merge_list);
2596                merge_list = next;
2597        }
2598        strbuf_addf(&new_data,
2599                "author %s\n"
2600                "committer %s\n"
2601                "\n",
2602                author ? author : committer, committer);
2603        strbuf_addbuf(&new_data, &msg);
2604        free(author);
2605        free(committer);
2606
2607        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2608                b->pack_id = pack_id;
2609        b->last_commit = object_count_by_type[OBJ_COMMIT];
2610}
2611
2612static void parse_new_tag(void)
2613{
2614        static struct strbuf msg = STRBUF_INIT;
2615        char *sp;
2616        const char *from;
2617        char *tagger;
2618        struct branch *s;
2619        struct tag *t;
2620        uintmax_t from_mark = 0;
2621        unsigned char sha1[20];
2622        enum object_type type;
2623
2624        /* Obtain the new tag name from the rest of our command */
2625        sp = strchr(command_buf.buf, ' ') + 1;
2626        t = pool_alloc(sizeof(struct tag));
2627        t->next_tag = NULL;
2628        t->name = pool_strdup(sp);
2629        if (last_tag)
2630                last_tag->next_tag = t;
2631        else
2632                first_tag = t;
2633        last_tag = t;
2634        read_next_command();
2635
2636        /* from ... */
2637        if (prefixcmp(command_buf.buf, "from "))
2638                die("Expected from command, got %s", command_buf.buf);
2639        from = strchr(command_buf.buf, ' ') + 1;
2640        s = lookup_branch(from);
2641        if (s) {
2642                hashcpy(sha1, s->sha1);
2643                type = OBJ_COMMIT;
2644        } else if (*from == ':') {
2645                struct object_entry *oe;
2646                from_mark = strtoumax(from + 1, NULL, 10);
2647                oe = find_mark(from_mark);
2648                type = oe->type;
2649                hashcpy(sha1, oe->idx.sha1);
2650        } else if (!get_sha1(from, sha1)) {
2651                unsigned long size;
2652                char *buf;
2653
2654                buf = read_sha1_file(sha1, &type, &size);
2655                if (!buf || size < 46)
2656                        die("Not a valid commit: %s", from);
2657                free(buf);
2658        } else
2659                die("Invalid ref name or SHA1 expression: %s", from);
2660        read_next_command();
2661
2662        /* tagger ... */
2663        if (!prefixcmp(command_buf.buf, "tagger ")) {
2664                tagger = parse_ident(command_buf.buf + 7);
2665                read_next_command();
2666        } else
2667                tagger = NULL;
2668
2669        /* tag payload/message */
2670        parse_data(&msg, 0, NULL);
2671
2672        /* build the tag object */
2673        strbuf_reset(&new_data);
2674
2675        strbuf_addf(&new_data,
2676                    "object %s\n"
2677                    "type %s\n"
2678                    "tag %s\n",
2679                    sha1_to_hex(sha1), typename(type), t->name);
2680        if (tagger)
2681                strbuf_addf(&new_data,
2682                            "tagger %s\n", tagger);
2683        strbuf_addch(&new_data, '\n');
2684        strbuf_addbuf(&new_data, &msg);
2685        free(tagger);
2686
2687        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2688                t->pack_id = MAX_PACK_ID;
2689        else
2690                t->pack_id = pack_id;
2691}
2692
2693static void parse_reset_branch(void)
2694{
2695        struct branch *b;
2696        char *sp;
2697
2698        /* Obtain the branch name from the rest of our command */
2699        sp = strchr(command_buf.buf, ' ') + 1;
2700        b = lookup_branch(sp);
2701        if (b) {
2702                hashclr(b->sha1);
2703                hashclr(b->branch_tree.versions[0].sha1);
2704                hashclr(b->branch_tree.versions[1].sha1);
2705                if (b->branch_tree.tree) {
2706                        release_tree_content_recursive(b->branch_tree.tree);
2707                        b->branch_tree.tree = NULL;
2708                }
2709        }
2710        else
2711                b = new_branch(sp);
2712        read_next_command();
2713        parse_from(b);
2714        if (command_buf.len > 0)
2715                unread_command_buf = 1;
2716}
2717
2718static void parse_checkpoint(void)
2719{
2720        if (object_count) {
2721                cycle_packfile();
2722                dump_branches();
2723                dump_tags();
2724                dump_marks();
2725        }
2726        skip_optional_lf();
2727}
2728
2729static void parse_progress(void)
2730{
2731        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2732        fputc('\n', stdout);
2733        fflush(stdout);
2734        skip_optional_lf();
2735}
2736
2737static char* make_fast_import_path(const char *path)
2738{
2739        struct strbuf abs_path = STRBUF_INIT;
2740
2741        if (!relative_marks_paths || is_absolute_path(path))
2742                return xstrdup(path);
2743        strbuf_addf(&abs_path, "%s/info/fast-import/%s", get_git_dir(), path);
2744        return strbuf_detach(&abs_path, NULL);
2745}
2746
2747static void option_import_marks(const char *marks, int from_stream)
2748{
2749        if (import_marks_file) {
2750                if (from_stream)
2751                        die("Only one import-marks command allowed per stream");
2752
2753                /* read previous mark file */
2754                if(!import_marks_file_from_stream)
2755                        read_marks();
2756        }
2757
2758        import_marks_file = make_fast_import_path(marks);
2759        safe_create_leading_directories_const(import_marks_file);
2760        import_marks_file_from_stream = from_stream;
2761}
2762
2763static void option_date_format(const char *fmt)
2764{
2765        if (!strcmp(fmt, "raw"))
2766                whenspec = WHENSPEC_RAW;
2767        else if (!strcmp(fmt, "rfc2822"))
2768                whenspec = WHENSPEC_RFC2822;
2769        else if (!strcmp(fmt, "now"))
2770                whenspec = WHENSPEC_NOW;
2771        else
2772                die("unknown --date-format argument %s", fmt);
2773}
2774
2775static void option_depth(const char *depth)
2776{
2777        max_depth = strtoul(depth, NULL, 0);
2778        if (max_depth > MAX_DEPTH)
2779                die("--depth cannot exceed %u", MAX_DEPTH);
2780}
2781
2782static void option_active_branches(const char *branches)
2783{
2784        max_active_branches = strtoul(branches, NULL, 0);
2785}
2786
2787static void option_export_marks(const char *marks)
2788{
2789        export_marks_file = make_fast_import_path(marks);
2790        safe_create_leading_directories_const(export_marks_file);
2791}
2792
2793static void option_export_pack_edges(const char *edges)
2794{
2795        if (pack_edges)
2796                fclose(pack_edges);
2797        pack_edges = fopen(edges, "a");
2798        if (!pack_edges)
2799                die_errno("Cannot open '%s'", edges);
2800}
2801
2802static int parse_one_option(const char *option)
2803{
2804        if (!prefixcmp(option, "max-pack-size=")) {
2805                unsigned long v;
2806                if (!git_parse_ulong(option + 14, &v))
2807                        return 0;
2808                if (v < 8192) {
2809                        warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
2810                        v *= 1024 * 1024;
2811                } else if (v < 1024 * 1024) {
2812                        warning("minimum max-pack-size is 1 MiB");
2813                        v = 1024 * 1024;
2814                }
2815                max_packsize = v;
2816        } else if (!prefixcmp(option, "big-file-threshold=")) {
2817                unsigned long v;
2818                if (!git_parse_ulong(option + 19, &v))
2819                        return 0;
2820                big_file_threshold = v;
2821        } else if (!prefixcmp(option, "depth=")) {
2822                option_depth(option + 6);
2823        } else if (!prefixcmp(option, "active-branches=")) {
2824                option_active_branches(option + 16);
2825        } else if (!prefixcmp(option, "export-pack-edges=")) {
2826                option_export_pack_edges(option + 18);
2827        } else if (!prefixcmp(option, "quiet")) {
2828                show_stats = 0;
2829        } else if (!prefixcmp(option, "stats")) {
2830                show_stats = 1;
2831        } else {
2832                return 0;
2833        }
2834
2835        return 1;
2836}
2837
2838static int parse_one_feature(const char *feature, int from_stream)
2839{
2840        if (!prefixcmp(feature, "date-format=")) {
2841                option_date_format(feature + 12);
2842        } else if (!prefixcmp(feature, "import-marks=")) {
2843                option_import_marks(feature + 13, from_stream);
2844        } else if (!prefixcmp(feature, "export-marks=")) {
2845                option_export_marks(feature + 13);
2846        } else if (!prefixcmp(feature, "relative-marks")) {
2847                relative_marks_paths = 1;
2848        } else if (!prefixcmp(feature, "no-relative-marks")) {
2849                relative_marks_paths = 0;
2850        } else if (!prefixcmp(feature, "force")) {
2851                force_update = 1;
2852        } else {
2853                return 0;
2854        }
2855
2856        return 1;
2857}
2858
2859static void parse_feature(void)
2860{
2861        char *feature = command_buf.buf + 8;
2862
2863        if (seen_data_command)
2864                die("Got feature command '%s' after data command", feature);
2865
2866        if (parse_one_feature(feature, 1))
2867                return;
2868
2869        die("This version of fast-import does not support feature %s.", feature);
2870}
2871
2872static void parse_option(void)
2873{
2874        char *option = command_buf.buf + 11;
2875
2876        if (seen_data_command)
2877                die("Got option command '%s' after data command", option);
2878
2879        if (parse_one_option(option))
2880                return;
2881
2882        die("This version of fast-import does not support option: %s", option);
2883}
2884
2885static int git_pack_config(const char *k, const char *v, void *cb)
2886{
2887        if (!strcmp(k, "pack.depth")) {
2888                max_depth = git_config_int(k, v);
2889                if (max_depth > MAX_DEPTH)
2890                        max_depth = MAX_DEPTH;
2891                return 0;
2892        }
2893        if (!strcmp(k, "pack.compression")) {
2894                int level = git_config_int(k, v);
2895                if (level == -1)
2896                        level = Z_DEFAULT_COMPRESSION;
2897                else if (level < 0 || level > Z_BEST_COMPRESSION)
2898                        die("bad pack compression level %d", level);
2899                pack_compression_level = level;
2900                pack_compression_seen = 1;
2901                return 0;
2902        }
2903        if (!strcmp(k, "pack.indexversion")) {
2904                pack_idx_default_version = git_config_int(k, v);
2905                if (pack_idx_default_version > 2)
2906                        die("bad pack.indexversion=%"PRIu32,
2907                            pack_idx_default_version);
2908                return 0;
2909        }
2910        if (!strcmp(k, "pack.packsizelimit")) {
2911                max_packsize = git_config_ulong(k, v);
2912                return 0;
2913        }
2914        if (!strcmp(k, "core.bigfilethreshold")) {
2915                long n = git_config_int(k, v);
2916                big_file_threshold = 0 < n ? n : 0;
2917        }
2918        return git_default_config(k, v, cb);
2919}
2920
2921static const char fast_import_usage[] =
2922"git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
2923
2924static void parse_argv(void)
2925{
2926        unsigned int i;
2927
2928        for (i = 1; i < global_argc; i++) {
2929                const char *a = global_argv[i];
2930
2931                if (*a != '-' || !strcmp(a, "--"))
2932                        break;
2933
2934                if (parse_one_option(a + 2))
2935                        continue;
2936
2937                if (parse_one_feature(a + 2, 0))
2938                        continue;
2939
2940                die("unknown option %s", a);
2941        }
2942        if (i != global_argc)
2943                usage(fast_import_usage);
2944
2945        seen_data_command = 1;
2946        if (import_marks_file)
2947                read_marks();
2948}
2949
2950int main(int argc, const char **argv)
2951{
2952        unsigned int i;
2953
2954        git_extract_argv0_path(argv[0]);
2955
2956        if (argc == 2 && !strcmp(argv[1], "-h"))
2957                usage(fast_import_usage);
2958
2959        setup_git_directory();
2960        git_config(git_pack_config, NULL);
2961        if (!pack_compression_seen && core_compression_seen)
2962                pack_compression_level = core_compression_level;
2963
2964        alloc_objects(object_entry_alloc);
2965        strbuf_init(&command_buf, 0);
2966        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2967        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2968        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2969        marks = pool_calloc(1, sizeof(struct mark_set));
2970
2971        global_argc = argc;
2972        global_argv = argv;
2973
2974        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2975        for (i = 0; i < (cmd_save - 1); i++)
2976                rc_free[i].next = &rc_free[i + 1];
2977        rc_free[cmd_save - 1].next = NULL;
2978
2979        prepare_packed_git();
2980        start_packfile();
2981        set_die_routine(die_nicely);
2982        while (read_next_command() != EOF) {
2983                if (!strcmp("blob", command_buf.buf))
2984                        parse_new_blob();
2985                else if (!prefixcmp(command_buf.buf, "commit "))
2986                        parse_new_commit();
2987                else if (!prefixcmp(command_buf.buf, "tag "))
2988                        parse_new_tag();
2989                else if (!prefixcmp(command_buf.buf, "reset "))
2990                        parse_reset_branch();
2991                else if (!strcmp("checkpoint", command_buf.buf))
2992                        parse_checkpoint();
2993                else if (!prefixcmp(command_buf.buf, "progress "))
2994                        parse_progress();
2995                else if (!prefixcmp(command_buf.buf, "feature "))
2996                        parse_feature();
2997                else if (!prefixcmp(command_buf.buf, "option git "))
2998                        parse_option();
2999                else if (!prefixcmp(command_buf.buf, "option "))
3000                        /* ignore non-git options*/;
3001                else
3002                        die("Unsupported command: %s", command_buf.buf);
3003        }
3004
3005        /* argv hasn't been parsed yet, do so */
3006        if (!seen_data_command)
3007                parse_argv();
3008
3009        end_packfile();
3010
3011        dump_branches();
3012        dump_tags();
3013        unkeep_all_packs();
3014        dump_marks();
3015
3016        if (pack_edges)
3017                fclose(pack_edges);
3018
3019        if (show_stats) {
3020                uintmax_t total_count = 0, duplicate_count = 0;
3021                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3022                        total_count += object_count_by_type[i];
3023                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3024                        duplicate_count += duplicate_count_by_type[i];
3025
3026                fprintf(stderr, "%s statistics:\n", argv[0]);
3027                fprintf(stderr, "---------------------------------------------------------------------\n");
3028                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3029                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
3030                fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
3031                fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
3032                fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
3033                fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
3034                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
3035                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3036                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
3037                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
3038                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
3039                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3040                fprintf(stderr, "---------------------------------------------------------------------\n");
3041                pack_report();
3042                fprintf(stderr, "---------------------------------------------------------------------\n");
3043                fprintf(stderr, "\n");
3044        }
3045
3046        return failure ? 1 : 0;
3047}