0458b03a2a45d7c16335d83b93dba1c33a6766b7
   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
 160#define PACK_ID_BITS 16
 161#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 162#define DEPTH_BITS 13
 163#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 164
 165struct object_entry
 166{
 167        struct object_entry *next;
 168        uint32_t offset;
 169        uint32_t type : TYPE_BITS,
 170                pack_id : PACK_ID_BITS,
 171                depth : DEPTH_BITS;
 172        unsigned char sha1[20];
 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        uint32_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        unsigned active : 1;
 249        unsigned pack_id : PACK_ID_BITS;
 250        unsigned char sha1[20];
 251};
 252
 253struct tag
 254{
 255        struct tag *next_tag;
 256        const char *name;
 257        unsigned int pack_id;
 258        unsigned char sha1[20];
 259};
 260
 261struct hash_list
 262{
 263        struct hash_list *next;
 264        unsigned char sha1[20];
 265};
 266
 267typedef enum {
 268        WHENSPEC_RAW = 1,
 269        WHENSPEC_RFC2822,
 270        WHENSPEC_NOW,
 271} whenspec_type;
 272
 273struct recent_command
 274{
 275        struct recent_command *prev;
 276        struct recent_command *next;
 277        char *buf;
 278};
 279
 280/* Configured limits on output */
 281static unsigned long max_depth = 10;
 282static off_t max_packsize = (1LL << 32) - 1;
 283static int force_update;
 284static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 285static int pack_compression_seen;
 286
 287/* Stats and misc. counters */
 288static uintmax_t alloc_count;
 289static uintmax_t marks_set_count;
 290static uintmax_t object_count_by_type[1 << TYPE_BITS];
 291static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 292static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 293static unsigned long object_count;
 294static unsigned long branch_count;
 295static unsigned long branch_load_count;
 296static int failure;
 297static FILE *pack_edges;
 298static unsigned int show_stats = 1;
 299
 300/* Memory pools */
 301static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 302static size_t total_allocd;
 303static struct mem_pool *mem_pool;
 304
 305/* Atom management */
 306static unsigned int atom_table_sz = 4451;
 307static unsigned int atom_cnt;
 308static struct atom_str **atom_table;
 309
 310/* The .pack file being generated */
 311static unsigned int pack_id;
 312static struct packed_git *pack_data;
 313static struct packed_git **all_packs;
 314static unsigned long pack_size;
 315
 316/* Table of objects we've written. */
 317static unsigned int object_entry_alloc = 5000;
 318static struct object_entry_pool *blocks;
 319static struct object_entry *object_table[1 << 16];
 320static struct mark_set *marks;
 321static const char *export_marks_file;
 322static const char *import_marks_file;
 323
 324/* Our last blob */
 325static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 326
 327/* Tree management */
 328static unsigned int tree_entry_alloc = 1000;
 329static void *avail_tree_entry;
 330static unsigned int avail_tree_table_sz = 100;
 331static struct avail_tree_content **avail_tree_table;
 332static struct strbuf old_tree = STRBUF_INIT;
 333static struct strbuf new_tree = STRBUF_INIT;
 334
 335/* Branch data */
 336static unsigned long max_active_branches = 5;
 337static unsigned long cur_active_branches;
 338static unsigned long branch_table_sz = 1039;
 339static struct branch **branch_table;
 340static struct branch *active_branches;
 341
 342/* Tag data */
 343static struct tag *first_tag;
 344static struct tag *last_tag;
 345
 346/* Input stream parsing */
 347static whenspec_type whenspec = WHENSPEC_RAW;
 348static struct strbuf command_buf = STRBUF_INIT;
 349static int unread_command_buf;
 350static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 351static struct recent_command *cmd_tail = &cmd_hist;
 352static struct recent_command *rc_free;
 353static unsigned int cmd_save = 100;
 354static uintmax_t next_mark;
 355static struct strbuf new_data = STRBUF_INIT;
 356
 357static void write_branch_report(FILE *rpt, struct branch *b)
 358{
 359        fprintf(rpt, "%s:\n", b->name);
 360
 361        fprintf(rpt, "  status      :");
 362        if (b->active)
 363                fputs(" active", rpt);
 364        if (b->branch_tree.tree)
 365                fputs(" loaded", rpt);
 366        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 367                fputs(" dirty", rpt);
 368        fputc('\n', rpt);
 369
 370        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 371        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 372        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 373        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 374
 375        fputs("  last pack   : ", rpt);
 376        if (b->pack_id < MAX_PACK_ID)
 377                fprintf(rpt, "%u", b->pack_id);
 378        fputc('\n', rpt);
 379
 380        fputc('\n', rpt);
 381}
 382
 383static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 384
 385static void write_crash_report(const char *err)
 386{
 387        char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 388        FILE *rpt = fopen(loc, "w");
 389        struct branch *b;
 390        unsigned long lu;
 391        struct recent_command *rc;
 392
 393        if (!rpt) {
 394                error("can't write crash report %s: %s", loc, strerror(errno));
 395                return;
 396        }
 397
 398        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 399
 400        fprintf(rpt, "fast-import crash report:\n");
 401        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 402        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 403        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 404        fputc('\n', rpt);
 405
 406        fputs("fatal: ", rpt);
 407        fputs(err, rpt);
 408        fputc('\n', rpt);
 409
 410        fputc('\n', rpt);
 411        fputs("Most Recent Commands Before Crash\n", rpt);
 412        fputs("---------------------------------\n", rpt);
 413        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 414                if (rc->next == &cmd_hist)
 415                        fputs("* ", rpt);
 416                else
 417                        fputs("  ", rpt);
 418                fputs(rc->buf, rpt);
 419                fputc('\n', rpt);
 420        }
 421
 422        fputc('\n', rpt);
 423        fputs("Active Branch LRU\n", rpt);
 424        fputs("-----------------\n", rpt);
 425        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 426                cur_active_branches,
 427                max_active_branches);
 428        fputc('\n', rpt);
 429        fputs("  pos  clock name\n", rpt);
 430        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 431        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 432                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 433                        ++lu, b->last_commit, b->name);
 434
 435        fputc('\n', rpt);
 436        fputs("Inactive Branches\n", rpt);
 437        fputs("-----------------\n", rpt);
 438        for (lu = 0; lu < branch_table_sz; lu++) {
 439                for (b = branch_table[lu]; b; b = b->table_next_branch)
 440                        write_branch_report(rpt, b);
 441        }
 442
 443        if (first_tag) {
 444                struct tag *tg;
 445                fputc('\n', rpt);
 446                fputs("Annotated Tags\n", rpt);
 447                fputs("--------------\n", rpt);
 448                for (tg = first_tag; tg; tg = tg->next_tag) {
 449                        fputs(sha1_to_hex(tg->sha1), rpt);
 450                        fputc(' ', rpt);
 451                        fputs(tg->name, rpt);
 452                        fputc('\n', rpt);
 453                }
 454        }
 455
 456        fputc('\n', rpt);
 457        fputs("Marks\n", rpt);
 458        fputs("-----\n", rpt);
 459        if (export_marks_file)
 460                fprintf(rpt, "  exported to %s\n", export_marks_file);
 461        else
 462                dump_marks_helper(rpt, 0, marks);
 463
 464        fputc('\n', rpt);
 465        fputs("-------------------\n", rpt);
 466        fputs("END OF CRASH REPORT\n", rpt);
 467        fclose(rpt);
 468}
 469
 470static void end_packfile(void);
 471static void unkeep_all_packs(void);
 472static void dump_marks(void);
 473
 474static NORETURN void die_nicely(const char *err, va_list params)
 475{
 476        static int zombie;
 477        char message[2 * PATH_MAX];
 478
 479        vsnprintf(message, sizeof(message), err, params);
 480        fputs("fatal: ", stderr);
 481        fputs(message, stderr);
 482        fputc('\n', stderr);
 483
 484        if (!zombie) {
 485                zombie = 1;
 486                write_crash_report(message);
 487                end_packfile();
 488                unkeep_all_packs();
 489                dump_marks();
 490        }
 491        exit(128);
 492}
 493
 494static void alloc_objects(unsigned int cnt)
 495{
 496        struct object_entry_pool *b;
 497
 498        b = xmalloc(sizeof(struct object_entry_pool)
 499                + cnt * sizeof(struct object_entry));
 500        b->next_pool = blocks;
 501        b->next_free = b->entries;
 502        b->end = b->entries + cnt;
 503        blocks = b;
 504        alloc_count += cnt;
 505}
 506
 507static struct object_entry *new_object(unsigned char *sha1)
 508{
 509        struct object_entry *e;
 510
 511        if (blocks->next_free == blocks->end)
 512                alloc_objects(object_entry_alloc);
 513
 514        e = blocks->next_free++;
 515        hashcpy(e->sha1, sha1);
 516        return e;
 517}
 518
 519static struct object_entry *find_object(unsigned char *sha1)
 520{
 521        unsigned int h = sha1[0] << 8 | sha1[1];
 522        struct object_entry *e;
 523        for (e = object_table[h]; e; e = e->next)
 524                if (!hashcmp(sha1, e->sha1))
 525                        return e;
 526        return NULL;
 527}
 528
 529static struct object_entry *insert_object(unsigned char *sha1)
 530{
 531        unsigned int h = sha1[0] << 8 | sha1[1];
 532        struct object_entry *e = object_table[h];
 533        struct object_entry *p = NULL;
 534
 535        while (e) {
 536                if (!hashcmp(sha1, e->sha1))
 537                        return e;
 538                p = e;
 539                e = e->next;
 540        }
 541
 542        e = new_object(sha1);
 543        e->next = NULL;
 544        e->offset = 0;
 545        if (p)
 546                p->next = e;
 547        else
 548                object_table[h] = e;
 549        return e;
 550}
 551
 552static unsigned int hc_str(const char *s, size_t len)
 553{
 554        unsigned int r = 0;
 555        while (len-- > 0)
 556                r = r * 31 + *s++;
 557        return r;
 558}
 559
 560static void *pool_alloc(size_t len)
 561{
 562        struct mem_pool *p;
 563        void *r;
 564
 565        /* round up to a 'uintmax_t' alignment */
 566        if (len & (sizeof(uintmax_t) - 1))
 567                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 568
 569        for (p = mem_pool; p; p = p->next_pool)
 570                if ((p->end - p->next_free >= len))
 571                        break;
 572
 573        if (!p) {
 574                if (len >= (mem_pool_alloc/2)) {
 575                        total_allocd += len;
 576                        return xmalloc(len);
 577                }
 578                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 579                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 580                p->next_pool = mem_pool;
 581                p->next_free = (char *) p->space;
 582                p->end = p->next_free + mem_pool_alloc;
 583                mem_pool = p;
 584        }
 585
 586        r = p->next_free;
 587        p->next_free += len;
 588        return r;
 589}
 590
 591static void *pool_calloc(size_t count, size_t size)
 592{
 593        size_t len = count * size;
 594        void *r = pool_alloc(len);
 595        memset(r, 0, len);
 596        return r;
 597}
 598
 599static char *pool_strdup(const char *s)
 600{
 601        char *r = pool_alloc(strlen(s) + 1);
 602        strcpy(r, s);
 603        return r;
 604}
 605
 606static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 607{
 608        struct mark_set *s = marks;
 609        while ((idnum >> s->shift) >= 1024) {
 610                s = pool_calloc(1, sizeof(struct mark_set));
 611                s->shift = marks->shift + 10;
 612                s->data.sets[0] = marks;
 613                marks = s;
 614        }
 615        while (s->shift) {
 616                uintmax_t i = idnum >> s->shift;
 617                idnum -= i << s->shift;
 618                if (!s->data.sets[i]) {
 619                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 620                        s->data.sets[i]->shift = s->shift - 10;
 621                }
 622                s = s->data.sets[i];
 623        }
 624        if (!s->data.marked[idnum])
 625                marks_set_count++;
 626        s->data.marked[idnum] = oe;
 627}
 628
 629static struct object_entry *find_mark(uintmax_t idnum)
 630{
 631        uintmax_t orig_idnum = idnum;
 632        struct mark_set *s = marks;
 633        struct object_entry *oe = NULL;
 634        if ((idnum >> s->shift) < 1024) {
 635                while (s && s->shift) {
 636                        uintmax_t i = idnum >> s->shift;
 637                        idnum -= i << s->shift;
 638                        s = s->data.sets[i];
 639                }
 640                if (s)
 641                        oe = s->data.marked[idnum];
 642        }
 643        if (!oe)
 644                die("mark :%" PRIuMAX " not declared", orig_idnum);
 645        return oe;
 646}
 647
 648static struct atom_str *to_atom(const char *s, unsigned short len)
 649{
 650        unsigned int hc = hc_str(s, len) % atom_table_sz;
 651        struct atom_str *c;
 652
 653        for (c = atom_table[hc]; c; c = c->next_atom)
 654                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 655                        return c;
 656
 657        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 658        c->str_len = len;
 659        strncpy(c->str_dat, s, len);
 660        c->str_dat[len] = 0;
 661        c->next_atom = atom_table[hc];
 662        atom_table[hc] = c;
 663        atom_cnt++;
 664        return c;
 665}
 666
 667static struct branch *lookup_branch(const char *name)
 668{
 669        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 670        struct branch *b;
 671
 672        for (b = branch_table[hc]; b; b = b->table_next_branch)
 673                if (!strcmp(name, b->name))
 674                        return b;
 675        return NULL;
 676}
 677
 678static struct branch *new_branch(const char *name)
 679{
 680        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 681        struct branch *b = lookup_branch(name);
 682
 683        if (b)
 684                die("Invalid attempt to create duplicate branch: %s", name);
 685        switch (check_ref_format(name)) {
 686        case 0: break; /* its valid */
 687        case CHECK_REF_FORMAT_ONELEVEL:
 688                break; /* valid, but too few '/', allow anyway */
 689        default:
 690                die("Branch name doesn't conform to GIT standards: %s", name);
 691        }
 692
 693        b = pool_calloc(1, sizeof(struct branch));
 694        b->name = pool_strdup(name);
 695        b->table_next_branch = branch_table[hc];
 696        b->branch_tree.versions[0].mode = S_IFDIR;
 697        b->branch_tree.versions[1].mode = S_IFDIR;
 698        b->active = 0;
 699        b->pack_id = MAX_PACK_ID;
 700        branch_table[hc] = b;
 701        branch_count++;
 702        return b;
 703}
 704
 705static unsigned int hc_entries(unsigned int cnt)
 706{
 707        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 708        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 709}
 710
 711static struct tree_content *new_tree_content(unsigned int cnt)
 712{
 713        struct avail_tree_content *f, *l = NULL;
 714        struct tree_content *t;
 715        unsigned int hc = hc_entries(cnt);
 716
 717        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 718                if (f->entry_capacity >= cnt)
 719                        break;
 720
 721        if (f) {
 722                if (l)
 723                        l->next_avail = f->next_avail;
 724                else
 725                        avail_tree_table[hc] = f->next_avail;
 726        } else {
 727                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 728                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 729                f->entry_capacity = cnt;
 730        }
 731
 732        t = (struct tree_content*)f;
 733        t->entry_count = 0;
 734        t->delta_depth = 0;
 735        return t;
 736}
 737
 738static void release_tree_entry(struct tree_entry *e);
 739static void release_tree_content(struct tree_content *t)
 740{
 741        struct avail_tree_content *f = (struct avail_tree_content*)t;
 742        unsigned int hc = hc_entries(f->entry_capacity);
 743        f->next_avail = avail_tree_table[hc];
 744        avail_tree_table[hc] = f;
 745}
 746
 747static void release_tree_content_recursive(struct tree_content *t)
 748{
 749        unsigned int i;
 750        for (i = 0; i < t->entry_count; i++)
 751                release_tree_entry(t->entries[i]);
 752        release_tree_content(t);
 753}
 754
 755static struct tree_content *grow_tree_content(
 756        struct tree_content *t,
 757        int amt)
 758{
 759        struct tree_content *r = new_tree_content(t->entry_count + amt);
 760        r->entry_count = t->entry_count;
 761        r->delta_depth = t->delta_depth;
 762        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 763        release_tree_content(t);
 764        return r;
 765}
 766
 767static struct tree_entry *new_tree_entry(void)
 768{
 769        struct tree_entry *e;
 770
 771        if (!avail_tree_entry) {
 772                unsigned int n = tree_entry_alloc;
 773                total_allocd += n * sizeof(struct tree_entry);
 774                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 775                while (n-- > 1) {
 776                        *((void**)e) = e + 1;
 777                        e++;
 778                }
 779                *((void**)e) = NULL;
 780        }
 781
 782        e = avail_tree_entry;
 783        avail_tree_entry = *((void**)e);
 784        return e;
 785}
 786
 787static void release_tree_entry(struct tree_entry *e)
 788{
 789        if (e->tree)
 790                release_tree_content_recursive(e->tree);
 791        *((void**)e) = avail_tree_entry;
 792        avail_tree_entry = e;
 793}
 794
 795static struct tree_content *dup_tree_content(struct tree_content *s)
 796{
 797        struct tree_content *d;
 798        struct tree_entry *a, *b;
 799        unsigned int i;
 800
 801        if (!s)
 802                return NULL;
 803        d = new_tree_content(s->entry_count);
 804        for (i = 0; i < s->entry_count; i++) {
 805                a = s->entries[i];
 806                b = new_tree_entry();
 807                memcpy(b, a, sizeof(*a));
 808                if (a->tree && is_null_sha1(b->versions[1].sha1))
 809                        b->tree = dup_tree_content(a->tree);
 810                else
 811                        b->tree = NULL;
 812                d->entries[i] = b;
 813        }
 814        d->entry_count = s->entry_count;
 815        d->delta_depth = s->delta_depth;
 816
 817        return d;
 818}
 819
 820static void start_packfile(void)
 821{
 822        static char tmpfile[PATH_MAX];
 823        struct packed_git *p;
 824        struct pack_header hdr;
 825        int pack_fd;
 826
 827        pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 828                              "pack/tmp_pack_XXXXXX");
 829        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 830        strcpy(p->pack_name, tmpfile);
 831        p->pack_fd = pack_fd;
 832
 833        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 834        hdr.hdr_version = htonl(2);
 835        hdr.hdr_entries = 0;
 836        write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 837
 838        pack_data = p;
 839        pack_size = sizeof(hdr);
 840        object_count = 0;
 841
 842        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 843        all_packs[pack_id] = p;
 844}
 845
 846static int oecmp (const void *a_, const void *b_)
 847{
 848        struct object_entry *a = *((struct object_entry**)a_);
 849        struct object_entry *b = *((struct object_entry**)b_);
 850        return hashcmp(a->sha1, b->sha1);
 851}
 852
 853static char *create_index(void)
 854{
 855        static char tmpfile[PATH_MAX];
 856        git_SHA_CTX ctx;
 857        struct sha1file *f;
 858        struct object_entry **idx, **c, **last, *e;
 859        struct object_entry_pool *o;
 860        uint32_t array[256];
 861        int i, idx_fd;
 862
 863        /* Build the sorted table of object IDs. */
 864        idx = xmalloc(object_count * sizeof(struct object_entry*));
 865        c = idx;
 866        for (o = blocks; o; o = o->next_pool)
 867                for (e = o->next_free; e-- != o->entries;)
 868                        if (pack_id == e->pack_id)
 869                                *c++ = e;
 870        last = idx + object_count;
 871        if (c != last)
 872                die("internal consistency error creating the index");
 873        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 874
 875        /* Generate the fan-out array. */
 876        c = idx;
 877        for (i = 0; i < 256; i++) {
 878                struct object_entry **next = c;
 879                while (next < last) {
 880                        if ((*next)->sha1[0] != i)
 881                                break;
 882                        next++;
 883                }
 884                array[i] = htonl(next - idx);
 885                c = next;
 886        }
 887
 888        idx_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 889                             "pack/tmp_idx_XXXXXX");
 890        f = sha1fd(idx_fd, tmpfile);
 891        sha1write(f, array, 256 * sizeof(int));
 892        git_SHA1_Init(&ctx);
 893        for (c = idx; c != last; c++) {
 894                uint32_t offset = htonl((*c)->offset);
 895                sha1write(f, &offset, 4);
 896                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 897                git_SHA1_Update(&ctx, (*c)->sha1, 20);
 898        }
 899        sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
 900        sha1close(f, NULL, CSUM_FSYNC);
 901        free(idx);
 902        git_SHA1_Final(pack_data->sha1, &ctx);
 903        return tmpfile;
 904}
 905
 906static char *keep_pack(char *curr_index_name)
 907{
 908        static char name[PATH_MAX];
 909        static const char *keep_msg = "fast-import";
 910        int keep_fd;
 911
 912        keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
 913        if (keep_fd < 0)
 914                die_errno("cannot create keep file");
 915        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 916        if (close(keep_fd))
 917                die_errno("failed to write keep file");
 918
 919        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 920                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 921        if (move_temp_to_file(pack_data->pack_name, name))
 922                die("cannot store pack file");
 923
 924        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 925                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 926        if (move_temp_to_file(curr_index_name, name))
 927                die("cannot store index file");
 928        return name;
 929}
 930
 931static void unkeep_all_packs(void)
 932{
 933        static char name[PATH_MAX];
 934        int k;
 935
 936        for (k = 0; k < pack_id; k++) {
 937                struct packed_git *p = all_packs[k];
 938                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 939                         get_object_directory(), sha1_to_hex(p->sha1));
 940                unlink_or_warn(name);
 941        }
 942}
 943
 944static void end_packfile(void)
 945{
 946        struct packed_git *old_p = pack_data, *new_p;
 947
 948        clear_delta_base_cache();
 949        if (object_count) {
 950                char *idx_name;
 951                int i;
 952                struct branch *b;
 953                struct tag *t;
 954
 955                close_pack_windows(pack_data);
 956                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 957                                    pack_data->pack_name, object_count,
 958                                    NULL, 0);
 959                close(pack_data->pack_fd);
 960                idx_name = keep_pack(create_index());
 961
 962                /* Register the packfile with core git's machinery. */
 963                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 964                if (!new_p)
 965                        die("core git rejected index %s", idx_name);
 966                all_packs[pack_id] = new_p;
 967                install_packed_git(new_p);
 968
 969                /* Print the boundary */
 970                if (pack_edges) {
 971                        fprintf(pack_edges, "%s:", new_p->pack_name);
 972                        for (i = 0; i < branch_table_sz; i++) {
 973                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 974                                        if (b->pack_id == pack_id)
 975                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 976                                }
 977                        }
 978                        for (t = first_tag; t; t = t->next_tag) {
 979                                if (t->pack_id == pack_id)
 980                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 981                        }
 982                        fputc('\n', pack_edges);
 983                        fflush(pack_edges);
 984                }
 985
 986                pack_id++;
 987        }
 988        else {
 989                close(old_p->pack_fd);
 990                unlink_or_warn(old_p->pack_name);
 991        }
 992        free(old_p);
 993
 994        /* We can't carry a delta across packfiles. */
 995        strbuf_release(&last_blob.data);
 996        last_blob.offset = 0;
 997        last_blob.depth = 0;
 998}
 999
1000static void cycle_packfile(void)
1001{
1002        end_packfile();
1003        start_packfile();
1004}
1005
1006static size_t encode_header(
1007        enum object_type type,
1008        size_t size,
1009        unsigned char *hdr)
1010{
1011        int n = 1;
1012        unsigned char c;
1013
1014        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
1015                die("bad type %d", type);
1016
1017        c = (type << 4) | (size & 15);
1018        size >>= 4;
1019        while (size) {
1020                *hdr++ = c | 0x80;
1021                c = size & 0x7f;
1022                size >>= 7;
1023                n++;
1024        }
1025        *hdr = c;
1026        return n;
1027}
1028
1029static int store_object(
1030        enum object_type type,
1031        struct strbuf *dat,
1032        struct last_object *last,
1033        unsigned char *sha1out,
1034        uintmax_t mark)
1035{
1036        void *out, *delta;
1037        struct object_entry *e;
1038        unsigned char hdr[96];
1039        unsigned char sha1[20];
1040        unsigned long hdrlen, deltalen;
1041        git_SHA_CTX c;
1042        z_stream s;
1043
1044        hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1045                (unsigned long)dat->len) + 1;
1046        git_SHA1_Init(&c);
1047        git_SHA1_Update(&c, hdr, hdrlen);
1048        git_SHA1_Update(&c, dat->buf, dat->len);
1049        git_SHA1_Final(sha1, &c);
1050        if (sha1out)
1051                hashcpy(sha1out, sha1);
1052
1053        e = insert_object(sha1);
1054        if (mark)
1055                insert_mark(mark, e);
1056        if (e->offset) {
1057                duplicate_count_by_type[type]++;
1058                return 1;
1059        } else if (find_sha1_pack(sha1, packed_git)) {
1060                e->type = type;
1061                e->pack_id = MAX_PACK_ID;
1062                e->offset = 1; /* just not zero! */
1063                duplicate_count_by_type[type]++;
1064                return 1;
1065        }
1066
1067        if (last && last->data.buf && last->depth < max_depth) {
1068                delta = diff_delta(last->data.buf, last->data.len,
1069                        dat->buf, dat->len,
1070                        &deltalen, 0);
1071                if (delta && deltalen >= dat->len) {
1072                        free(delta);
1073                        delta = NULL;
1074                }
1075        } else
1076                delta = NULL;
1077
1078        memset(&s, 0, sizeof(s));
1079        deflateInit(&s, pack_compression_level);
1080        if (delta) {
1081                s.next_in = delta;
1082                s.avail_in = deltalen;
1083        } else {
1084                s.next_in = (void *)dat->buf;
1085                s.avail_in = dat->len;
1086        }
1087        s.avail_out = deflateBound(&s, s.avail_in);
1088        s.next_out = out = xmalloc(s.avail_out);
1089        while (deflate(&s, Z_FINISH) == Z_OK)
1090                /* nothing */;
1091        deflateEnd(&s);
1092
1093        /* Determine if we should auto-checkpoint. */
1094        if ((pack_size + 60 + s.total_out) > max_packsize
1095                || (pack_size + 60 + s.total_out) < pack_size) {
1096
1097                /* This new object needs to *not* have the current pack_id. */
1098                e->pack_id = pack_id + 1;
1099                cycle_packfile();
1100
1101                /* We cannot carry a delta into the new pack. */
1102                if (delta) {
1103                        free(delta);
1104                        delta = NULL;
1105
1106                        memset(&s, 0, sizeof(s));
1107                        deflateInit(&s, pack_compression_level);
1108                        s.next_in = (void *)dat->buf;
1109                        s.avail_in = dat->len;
1110                        s.avail_out = deflateBound(&s, s.avail_in);
1111                        s.next_out = out = xrealloc(out, s.avail_out);
1112                        while (deflate(&s, Z_FINISH) == Z_OK)
1113                                /* nothing */;
1114                        deflateEnd(&s);
1115                }
1116        }
1117
1118        e->type = type;
1119        e->pack_id = pack_id;
1120        e->offset = pack_size;
1121        object_count++;
1122        object_count_by_type[type]++;
1123
1124        if (delta) {
1125                unsigned long ofs = e->offset - last->offset;
1126                unsigned pos = sizeof(hdr) - 1;
1127
1128                delta_count_by_type[type]++;
1129                e->depth = last->depth + 1;
1130
1131                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1132                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1133                pack_size += hdrlen;
1134
1135                hdr[pos] = ofs & 127;
1136                while (ofs >>= 7)
1137                        hdr[--pos] = 128 | (--ofs & 127);
1138                write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1139                pack_size += sizeof(hdr) - pos;
1140        } else {
1141                e->depth = 0;
1142                hdrlen = encode_header(type, dat->len, hdr);
1143                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1144                pack_size += hdrlen;
1145        }
1146
1147        write_or_die(pack_data->pack_fd, out, s.total_out);
1148        pack_size += s.total_out;
1149
1150        free(out);
1151        free(delta);
1152        if (last) {
1153                if (last->no_swap) {
1154                        last->data = *dat;
1155                } else {
1156                        strbuf_swap(&last->data, dat);
1157                }
1158                last->offset = e->offset;
1159                last->depth = e->depth;
1160        }
1161        return 0;
1162}
1163
1164/* All calls must be guarded by find_object() or find_mark() to
1165 * ensure the 'struct object_entry' passed was written by this
1166 * process instance.  We unpack the entry by the offset, avoiding
1167 * the need for the corresponding .idx file.  This unpacking rule
1168 * works because we only use OBJ_REF_DELTA within the packfiles
1169 * created by fast-import.
1170 *
1171 * oe must not be NULL.  Such an oe usually comes from giving
1172 * an unknown SHA-1 to find_object() or an undefined mark to
1173 * find_mark().  Callers must test for this condition and use
1174 * the standard read_sha1_file() when it happens.
1175 *
1176 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1177 * find_mark(), where the mark was reloaded from an existing marks
1178 * file and is referencing an object that this fast-import process
1179 * instance did not write out to a packfile.  Callers must test for
1180 * this condition and use read_sha1_file() instead.
1181 */
1182static void *gfi_unpack_entry(
1183        struct object_entry *oe,
1184        unsigned long *sizep)
1185{
1186        enum object_type type;
1187        struct packed_git *p = all_packs[oe->pack_id];
1188        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1189                /* The object is stored in the packfile we are writing to
1190                 * and we have modified it since the last time we scanned
1191                 * back to read a previously written object.  If an old
1192                 * window covered [p->pack_size, p->pack_size + 20) its
1193                 * data is stale and is not valid.  Closing all windows
1194                 * and updating the packfile length ensures we can read
1195                 * the newly written data.
1196                 */
1197                close_pack_windows(p);
1198
1199                /* We have to offer 20 bytes additional on the end of
1200                 * the packfile as the core unpacker code assumes the
1201                 * footer is present at the file end and must promise
1202                 * at least 20 bytes within any window it maps.  But
1203                 * we don't actually create the footer here.
1204                 */
1205                p->pack_size = pack_size + 20;
1206        }
1207        return unpack_entry(p, oe->offset, &type, sizep);
1208}
1209
1210static const char *get_mode(const char *str, uint16_t *modep)
1211{
1212        unsigned char c;
1213        uint16_t mode = 0;
1214
1215        while ((c = *str++) != ' ') {
1216                if (c < '0' || c > '7')
1217                        return NULL;
1218                mode = (mode << 3) + (c - '0');
1219        }
1220        *modep = mode;
1221        return str;
1222}
1223
1224static void load_tree(struct tree_entry *root)
1225{
1226        unsigned char *sha1 = root->versions[1].sha1;
1227        struct object_entry *myoe;
1228        struct tree_content *t;
1229        unsigned long size;
1230        char *buf;
1231        const char *c;
1232
1233        root->tree = t = new_tree_content(8);
1234        if (is_null_sha1(sha1))
1235                return;
1236
1237        myoe = find_object(sha1);
1238        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1239                if (myoe->type != OBJ_TREE)
1240                        die("Not a tree: %s", sha1_to_hex(sha1));
1241                t->delta_depth = myoe->depth;
1242                buf = gfi_unpack_entry(myoe, &size);
1243                if (!buf)
1244                        die("Can't load tree %s", sha1_to_hex(sha1));
1245        } else {
1246                enum object_type type;
1247                buf = read_sha1_file(sha1, &type, &size);
1248                if (!buf || type != OBJ_TREE)
1249                        die("Can't load tree %s", sha1_to_hex(sha1));
1250        }
1251
1252        c = buf;
1253        while (c != (buf + size)) {
1254                struct tree_entry *e = new_tree_entry();
1255
1256                if (t->entry_count == t->entry_capacity)
1257                        root->tree = t = grow_tree_content(t, t->entry_count);
1258                t->entries[t->entry_count++] = e;
1259
1260                e->tree = NULL;
1261                c = get_mode(c, &e->versions[1].mode);
1262                if (!c)
1263                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1264                e->versions[0].mode = e->versions[1].mode;
1265                e->name = to_atom(c, strlen(c));
1266                c += e->name->str_len + 1;
1267                hashcpy(e->versions[0].sha1, (unsigned char *)c);
1268                hashcpy(e->versions[1].sha1, (unsigned char *)c);
1269                c += 20;
1270        }
1271        free(buf);
1272}
1273
1274static int tecmp0 (const void *_a, const void *_b)
1275{
1276        struct tree_entry *a = *((struct tree_entry**)_a);
1277        struct tree_entry *b = *((struct tree_entry**)_b);
1278        return base_name_compare(
1279                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1280                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1281}
1282
1283static int tecmp1 (const void *_a, const void *_b)
1284{
1285        struct tree_entry *a = *((struct tree_entry**)_a);
1286        struct tree_entry *b = *((struct tree_entry**)_b);
1287        return base_name_compare(
1288                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1289                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1290}
1291
1292static void mktree(struct tree_content *t, int v, struct strbuf *b)
1293{
1294        size_t maxlen = 0;
1295        unsigned int i;
1296
1297        if (!v)
1298                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1299        else
1300                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1301
1302        for (i = 0; i < t->entry_count; i++) {
1303                if (t->entries[i]->versions[v].mode)
1304                        maxlen += t->entries[i]->name->str_len + 34;
1305        }
1306
1307        strbuf_reset(b);
1308        strbuf_grow(b, maxlen);
1309        for (i = 0; i < t->entry_count; i++) {
1310                struct tree_entry *e = t->entries[i];
1311                if (!e->versions[v].mode)
1312                        continue;
1313                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1314                                        e->name->str_dat, '\0');
1315                strbuf_add(b, e->versions[v].sha1, 20);
1316        }
1317}
1318
1319static void store_tree(struct tree_entry *root)
1320{
1321        struct tree_content *t = root->tree;
1322        unsigned int i, j, del;
1323        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1324        struct object_entry *le;
1325
1326        if (!is_null_sha1(root->versions[1].sha1))
1327                return;
1328
1329        for (i = 0; i < t->entry_count; i++) {
1330                if (t->entries[i]->tree)
1331                        store_tree(t->entries[i]);
1332        }
1333
1334        le = find_object(root->versions[0].sha1);
1335        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1336                mktree(t, 0, &old_tree);
1337                lo.data = old_tree;
1338                lo.offset = le->offset;
1339                lo.depth = t->delta_depth;
1340        }
1341
1342        mktree(t, 1, &new_tree);
1343        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1344
1345        t->delta_depth = lo.depth;
1346        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1347                struct tree_entry *e = t->entries[i];
1348                if (e->versions[1].mode) {
1349                        e->versions[0].mode = e->versions[1].mode;
1350                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1351                        t->entries[j++] = e;
1352                } else {
1353                        release_tree_entry(e);
1354                        del++;
1355                }
1356        }
1357        t->entry_count -= del;
1358}
1359
1360static int tree_content_set(
1361        struct tree_entry *root,
1362        const char *p,
1363        const unsigned char *sha1,
1364        const uint16_t mode,
1365        struct tree_content *subtree)
1366{
1367        struct tree_content *t = root->tree;
1368        const char *slash1;
1369        unsigned int i, n;
1370        struct tree_entry *e;
1371
1372        slash1 = strchr(p, '/');
1373        if (slash1)
1374                n = slash1 - p;
1375        else
1376                n = strlen(p);
1377        if (!n)
1378                die("Empty path component found in input");
1379        if (!slash1 && !S_ISDIR(mode) && subtree)
1380                die("Non-directories cannot have subtrees");
1381
1382        for (i = 0; i < t->entry_count; i++) {
1383                e = t->entries[i];
1384                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1385                        if (!slash1) {
1386                                if (!S_ISDIR(mode)
1387                                                && e->versions[1].mode == mode
1388                                                && !hashcmp(e->versions[1].sha1, sha1))
1389                                        return 0;
1390                                e->versions[1].mode = mode;
1391                                hashcpy(e->versions[1].sha1, sha1);
1392                                if (e->tree)
1393                                        release_tree_content_recursive(e->tree);
1394                                e->tree = subtree;
1395                                hashclr(root->versions[1].sha1);
1396                                return 1;
1397                        }
1398                        if (!S_ISDIR(e->versions[1].mode)) {
1399                                e->tree = new_tree_content(8);
1400                                e->versions[1].mode = S_IFDIR;
1401                        }
1402                        if (!e->tree)
1403                                load_tree(e);
1404                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1405                                hashclr(root->versions[1].sha1);
1406                                return 1;
1407                        }
1408                        return 0;
1409                }
1410        }
1411
1412        if (t->entry_count == t->entry_capacity)
1413                root->tree = t = grow_tree_content(t, t->entry_count);
1414        e = new_tree_entry();
1415        e->name = to_atom(p, n);
1416        e->versions[0].mode = 0;
1417        hashclr(e->versions[0].sha1);
1418        t->entries[t->entry_count++] = e;
1419        if (slash1) {
1420                e->tree = new_tree_content(8);
1421                e->versions[1].mode = S_IFDIR;
1422                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1423        } else {
1424                e->tree = subtree;
1425                e->versions[1].mode = mode;
1426                hashcpy(e->versions[1].sha1, sha1);
1427        }
1428        hashclr(root->versions[1].sha1);
1429        return 1;
1430}
1431
1432static int tree_content_remove(
1433        struct tree_entry *root,
1434        const char *p,
1435        struct tree_entry *backup_leaf)
1436{
1437        struct tree_content *t = root->tree;
1438        const char *slash1;
1439        unsigned int i, n;
1440        struct tree_entry *e;
1441
1442        slash1 = strchr(p, '/');
1443        if (slash1)
1444                n = slash1 - p;
1445        else
1446                n = strlen(p);
1447
1448        for (i = 0; i < t->entry_count; i++) {
1449                e = t->entries[i];
1450                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1451                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1452                                goto del_entry;
1453                        if (!e->tree)
1454                                load_tree(e);
1455                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1456                                for (n = 0; n < e->tree->entry_count; n++) {
1457                                        if (e->tree->entries[n]->versions[1].mode) {
1458                                                hashclr(root->versions[1].sha1);
1459                                                return 1;
1460                                        }
1461                                }
1462                                backup_leaf = NULL;
1463                                goto del_entry;
1464                        }
1465                        return 0;
1466                }
1467        }
1468        return 0;
1469
1470del_entry:
1471        if (backup_leaf)
1472                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1473        else if (e->tree)
1474                release_tree_content_recursive(e->tree);
1475        e->tree = NULL;
1476        e->versions[1].mode = 0;
1477        hashclr(e->versions[1].sha1);
1478        hashclr(root->versions[1].sha1);
1479        return 1;
1480}
1481
1482static int tree_content_get(
1483        struct tree_entry *root,
1484        const char *p,
1485        struct tree_entry *leaf)
1486{
1487        struct tree_content *t = root->tree;
1488        const char *slash1;
1489        unsigned int i, n;
1490        struct tree_entry *e;
1491
1492        slash1 = strchr(p, '/');
1493        if (slash1)
1494                n = slash1 - p;
1495        else
1496                n = strlen(p);
1497
1498        for (i = 0; i < t->entry_count; i++) {
1499                e = t->entries[i];
1500                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1501                        if (!slash1) {
1502                                memcpy(leaf, e, sizeof(*leaf));
1503                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1504                                        leaf->tree = dup_tree_content(e->tree);
1505                                else
1506                                        leaf->tree = NULL;
1507                                return 1;
1508                        }
1509                        if (!S_ISDIR(e->versions[1].mode))
1510                                return 0;
1511                        if (!e->tree)
1512                                load_tree(e);
1513                        return tree_content_get(e, slash1 + 1, leaf);
1514                }
1515        }
1516        return 0;
1517}
1518
1519static int update_branch(struct branch *b)
1520{
1521        static const char *msg = "fast-import";
1522        struct ref_lock *lock;
1523        unsigned char old_sha1[20];
1524
1525        if (is_null_sha1(b->sha1))
1526                return 0;
1527        if (read_ref(b->name, old_sha1))
1528                hashclr(old_sha1);
1529        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1530        if (!lock)
1531                return error("Unable to lock %s", b->name);
1532        if (!force_update && !is_null_sha1(old_sha1)) {
1533                struct commit *old_cmit, *new_cmit;
1534
1535                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1536                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1537                if (!old_cmit || !new_cmit) {
1538                        unlock_ref(lock);
1539                        return error("Branch %s is missing commits.", b->name);
1540                }
1541
1542                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1543                        unlock_ref(lock);
1544                        warning("Not updating %s"
1545                                " (new tip %s does not contain %s)",
1546                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1547                        return -1;
1548                }
1549        }
1550        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1551                return error("Unable to update %s", b->name);
1552        return 0;
1553}
1554
1555static void dump_branches(void)
1556{
1557        unsigned int i;
1558        struct branch *b;
1559
1560        for (i = 0; i < branch_table_sz; i++) {
1561                for (b = branch_table[i]; b; b = b->table_next_branch)
1562                        failure |= update_branch(b);
1563        }
1564}
1565
1566static void dump_tags(void)
1567{
1568        static const char *msg = "fast-import";
1569        struct tag *t;
1570        struct ref_lock *lock;
1571        char ref_name[PATH_MAX];
1572
1573        for (t = first_tag; t; t = t->next_tag) {
1574                sprintf(ref_name, "tags/%s", t->name);
1575                lock = lock_ref_sha1(ref_name, NULL);
1576                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1577                        failure |= error("Unable to update %s", ref_name);
1578        }
1579}
1580
1581static void dump_marks_helper(FILE *f,
1582        uintmax_t base,
1583        struct mark_set *m)
1584{
1585        uintmax_t k;
1586        if (m->shift) {
1587                for (k = 0; k < 1024; k++) {
1588                        if (m->data.sets[k])
1589                                dump_marks_helper(f, (base + k) << m->shift,
1590                                        m->data.sets[k]);
1591                }
1592        } else {
1593                for (k = 0; k < 1024; k++) {
1594                        if (m->data.marked[k])
1595                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1596                                        sha1_to_hex(m->data.marked[k]->sha1));
1597                }
1598        }
1599}
1600
1601static void dump_marks(void)
1602{
1603        static struct lock_file mark_lock;
1604        int mark_fd;
1605        FILE *f;
1606
1607        if (!export_marks_file)
1608                return;
1609
1610        mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1611        if (mark_fd < 0) {
1612                failure |= error("Unable to write marks file %s: %s",
1613                        export_marks_file, strerror(errno));
1614                return;
1615        }
1616
1617        f = fdopen(mark_fd, "w");
1618        if (!f) {
1619                int saved_errno = errno;
1620                rollback_lock_file(&mark_lock);
1621                failure |= error("Unable to write marks file %s: %s",
1622                        export_marks_file, strerror(saved_errno));
1623                return;
1624        }
1625
1626        /*
1627         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1628         * Assign -1 to the lock file descriptor so that commit_lock_file()
1629         * won't try to close() it.
1630         */
1631        mark_lock.fd = -1;
1632
1633        dump_marks_helper(f, 0, marks);
1634        if (ferror(f) || fclose(f)) {
1635                int saved_errno = errno;
1636                rollback_lock_file(&mark_lock);
1637                failure |= error("Unable to write marks file %s: %s",
1638                        export_marks_file, strerror(saved_errno));
1639                return;
1640        }
1641
1642        if (commit_lock_file(&mark_lock)) {
1643                int saved_errno = errno;
1644                rollback_lock_file(&mark_lock);
1645                failure |= error("Unable to commit marks file %s: %s",
1646                        export_marks_file, strerror(saved_errno));
1647                return;
1648        }
1649}
1650
1651static void read_marks(void)
1652{
1653        char line[512];
1654        FILE *f = fopen(import_marks_file, "r");
1655        if (!f)
1656                die_errno("cannot read '%s'", import_marks_file);
1657        while (fgets(line, sizeof(line), f)) {
1658                uintmax_t mark;
1659                char *end;
1660                unsigned char sha1[20];
1661                struct object_entry *e;
1662
1663                end = strchr(line, '\n');
1664                if (line[0] != ':' || !end)
1665                        die("corrupt mark line: %s", line);
1666                *end = 0;
1667                mark = strtoumax(line + 1, &end, 10);
1668                if (!mark || end == line + 1
1669                        || *end != ' ' || get_sha1(end + 1, sha1))
1670                        die("corrupt mark line: %s", line);
1671                e = find_object(sha1);
1672                if (!e) {
1673                        enum object_type type = sha1_object_info(sha1, NULL);
1674                        if (type < 0)
1675                                die("object not found: %s", sha1_to_hex(sha1));
1676                        e = insert_object(sha1);
1677                        e->type = type;
1678                        e->pack_id = MAX_PACK_ID;
1679                        e->offset = 1; /* just not zero! */
1680                }
1681                insert_mark(mark, e);
1682        }
1683        fclose(f);
1684}
1685
1686
1687static int read_next_command(void)
1688{
1689        static int stdin_eof = 0;
1690
1691        if (stdin_eof) {
1692                unread_command_buf = 0;
1693                return EOF;
1694        }
1695
1696        do {
1697                if (unread_command_buf) {
1698                        unread_command_buf = 0;
1699                } else {
1700                        struct recent_command *rc;
1701
1702                        strbuf_detach(&command_buf, NULL);
1703                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1704                        if (stdin_eof)
1705                                return EOF;
1706
1707                        rc = rc_free;
1708                        if (rc)
1709                                rc_free = rc->next;
1710                        else {
1711                                rc = cmd_hist.next;
1712                                cmd_hist.next = rc->next;
1713                                cmd_hist.next->prev = &cmd_hist;
1714                                free(rc->buf);
1715                        }
1716
1717                        rc->buf = command_buf.buf;
1718                        rc->prev = cmd_tail;
1719                        rc->next = cmd_hist.prev;
1720                        rc->prev->next = rc;
1721                        cmd_tail = rc;
1722                }
1723        } while (command_buf.buf[0] == '#');
1724
1725        return 0;
1726}
1727
1728static void skip_optional_lf(void)
1729{
1730        int term_char = fgetc(stdin);
1731        if (term_char != '\n' && term_char != EOF)
1732                ungetc(term_char, stdin);
1733}
1734
1735static void parse_mark(void)
1736{
1737        if (!prefixcmp(command_buf.buf, "mark :")) {
1738                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1739                read_next_command();
1740        }
1741        else
1742                next_mark = 0;
1743}
1744
1745static void parse_data(struct strbuf *sb)
1746{
1747        strbuf_reset(sb);
1748
1749        if (prefixcmp(command_buf.buf, "data "))
1750                die("Expected 'data n' command, found: %s", command_buf.buf);
1751
1752        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1753                char *term = xstrdup(command_buf.buf + 5 + 2);
1754                size_t term_len = command_buf.len - 5 - 2;
1755
1756                strbuf_detach(&command_buf, NULL);
1757                for (;;) {
1758                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1759                                die("EOF in data (terminator '%s' not found)", term);
1760                        if (term_len == command_buf.len
1761                                && !strcmp(term, command_buf.buf))
1762                                break;
1763                        strbuf_addbuf(sb, &command_buf);
1764                        strbuf_addch(sb, '\n');
1765                }
1766                free(term);
1767        }
1768        else {
1769                size_t n = 0, length;
1770
1771                length = strtoul(command_buf.buf + 5, NULL, 10);
1772
1773                while (n < length) {
1774                        size_t s = strbuf_fread(sb, length - n, stdin);
1775                        if (!s && feof(stdin))
1776                                die("EOF in data (%lu bytes remaining)",
1777                                        (unsigned long)(length - n));
1778                        n += s;
1779                }
1780        }
1781
1782        skip_optional_lf();
1783}
1784
1785static int validate_raw_date(const char *src, char *result, int maxlen)
1786{
1787        const char *orig_src = src;
1788        char *endp;
1789        unsigned long num;
1790
1791        errno = 0;
1792
1793        num = strtoul(src, &endp, 10);
1794        /* NEEDSWORK: perhaps check for reasonable values? */
1795        if (errno || endp == src || *endp != ' ')
1796                return -1;
1797
1798        src = endp + 1;
1799        if (*src != '-' && *src != '+')
1800                return -1;
1801
1802        num = strtoul(src + 1, &endp, 10);
1803        if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1804            1400 < num)
1805                return -1;
1806
1807        strcpy(result, orig_src);
1808        return 0;
1809}
1810
1811static char *parse_ident(const char *buf)
1812{
1813        const char *gt;
1814        size_t name_len;
1815        char *ident;
1816
1817        gt = strrchr(buf, '>');
1818        if (!gt)
1819                die("Missing > in ident string: %s", buf);
1820        gt++;
1821        if (*gt != ' ')
1822                die("Missing space after > in ident string: %s", buf);
1823        gt++;
1824        name_len = gt - buf;
1825        ident = xmalloc(name_len + 24);
1826        strncpy(ident, buf, name_len);
1827
1828        switch (whenspec) {
1829        case WHENSPEC_RAW:
1830                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1831                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1832                break;
1833        case WHENSPEC_RFC2822:
1834                if (parse_date(gt, ident + name_len, 24) < 0)
1835                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1836                break;
1837        case WHENSPEC_NOW:
1838                if (strcmp("now", gt))
1839                        die("Date in ident must be 'now': %s", buf);
1840                datestamp(ident + name_len, 24);
1841                break;
1842        }
1843
1844        return ident;
1845}
1846
1847static void parse_new_blob(void)
1848{
1849        static struct strbuf buf = STRBUF_INIT;
1850
1851        read_next_command();
1852        parse_mark();
1853        parse_data(&buf);
1854        store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1855}
1856
1857static void unload_one_branch(void)
1858{
1859        while (cur_active_branches
1860                && cur_active_branches >= max_active_branches) {
1861                uintmax_t min_commit = ULONG_MAX;
1862                struct branch *e, *l = NULL, *p = NULL;
1863
1864                for (e = active_branches; e; e = e->active_next_branch) {
1865                        if (e->last_commit < min_commit) {
1866                                p = l;
1867                                min_commit = e->last_commit;
1868                        }
1869                        l = e;
1870                }
1871
1872                if (p) {
1873                        e = p->active_next_branch;
1874                        p->active_next_branch = e->active_next_branch;
1875                } else {
1876                        e = active_branches;
1877                        active_branches = e->active_next_branch;
1878                }
1879                e->active = 0;
1880                e->active_next_branch = NULL;
1881                if (e->branch_tree.tree) {
1882                        release_tree_content_recursive(e->branch_tree.tree);
1883                        e->branch_tree.tree = NULL;
1884                }
1885                cur_active_branches--;
1886        }
1887}
1888
1889static void load_branch(struct branch *b)
1890{
1891        load_tree(&b->branch_tree);
1892        if (!b->active) {
1893                b->active = 1;
1894                b->active_next_branch = active_branches;
1895                active_branches = b;
1896                cur_active_branches++;
1897                branch_load_count++;
1898        }
1899}
1900
1901static void file_change_m(struct branch *b)
1902{
1903        const char *p = command_buf.buf + 2;
1904        static struct strbuf uq = STRBUF_INIT;
1905        const char *endp;
1906        struct object_entry *oe = oe;
1907        unsigned char sha1[20];
1908        uint16_t mode, inline_data = 0;
1909
1910        p = get_mode(p, &mode);
1911        if (!p)
1912                die("Corrupt mode: %s", command_buf.buf);
1913        switch (mode) {
1914        case 0644:
1915        case 0755:
1916                mode |= S_IFREG;
1917        case S_IFREG | 0644:
1918        case S_IFREG | 0755:
1919        case S_IFLNK:
1920        case S_IFGITLINK:
1921                /* ok */
1922                break;
1923        default:
1924                die("Corrupt mode: %s", command_buf.buf);
1925        }
1926
1927        if (*p == ':') {
1928                char *x;
1929                oe = find_mark(strtoumax(p + 1, &x, 10));
1930                hashcpy(sha1, oe->sha1);
1931                p = x;
1932        } else if (!prefixcmp(p, "inline")) {
1933                inline_data = 1;
1934                p += 6;
1935        } else {
1936                if (get_sha1_hex(p, sha1))
1937                        die("Invalid SHA1: %s", command_buf.buf);
1938                oe = find_object(sha1);
1939                p += 40;
1940        }
1941        if (*p++ != ' ')
1942                die("Missing space after SHA1: %s", command_buf.buf);
1943
1944        strbuf_reset(&uq);
1945        if (!unquote_c_style(&uq, p, &endp)) {
1946                if (*endp)
1947                        die("Garbage after path in: %s", command_buf.buf);
1948                p = uq.buf;
1949        }
1950
1951        if (S_ISGITLINK(mode)) {
1952                if (inline_data)
1953                        die("Git links cannot be specified 'inline': %s",
1954                                command_buf.buf);
1955                else if (oe) {
1956                        if (oe->type != OBJ_COMMIT)
1957                                die("Not a commit (actually a %s): %s",
1958                                        typename(oe->type), command_buf.buf);
1959                }
1960                /*
1961                 * Accept the sha1 without checking; it expected to be in
1962                 * another repository.
1963                 */
1964        } else if (inline_data) {
1965                static struct strbuf buf = STRBUF_INIT;
1966
1967                if (p != uq.buf) {
1968                        strbuf_addstr(&uq, p);
1969                        p = uq.buf;
1970                }
1971                read_next_command();
1972                parse_data(&buf);
1973                store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1974        } else if (oe) {
1975                if (oe->type != OBJ_BLOB)
1976                        die("Not a blob (actually a %s): %s",
1977                                typename(oe->type), command_buf.buf);
1978        } else {
1979                enum object_type type = sha1_object_info(sha1, NULL);
1980                if (type < 0)
1981                        die("Blob not found: %s", command_buf.buf);
1982                if (type != OBJ_BLOB)
1983                        die("Not a blob (actually a %s): %s",
1984                            typename(type), command_buf.buf);
1985        }
1986
1987        tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
1988}
1989
1990static void file_change_d(struct branch *b)
1991{
1992        const char *p = command_buf.buf + 2;
1993        static struct strbuf uq = STRBUF_INIT;
1994        const char *endp;
1995
1996        strbuf_reset(&uq);
1997        if (!unquote_c_style(&uq, p, &endp)) {
1998                if (*endp)
1999                        die("Garbage after path in: %s", command_buf.buf);
2000                p = uq.buf;
2001        }
2002        tree_content_remove(&b->branch_tree, p, NULL);
2003}
2004
2005static void file_change_cr(struct branch *b, int rename)
2006{
2007        const char *s, *d;
2008        static struct strbuf s_uq = STRBUF_INIT;
2009        static struct strbuf d_uq = STRBUF_INIT;
2010        const char *endp;
2011        struct tree_entry leaf;
2012
2013        s = command_buf.buf + 2;
2014        strbuf_reset(&s_uq);
2015        if (!unquote_c_style(&s_uq, s, &endp)) {
2016                if (*endp != ' ')
2017                        die("Missing space after source: %s", command_buf.buf);
2018        } else {
2019                endp = strchr(s, ' ');
2020                if (!endp)
2021                        die("Missing space after source: %s", command_buf.buf);
2022                strbuf_add(&s_uq, s, endp - s);
2023        }
2024        s = s_uq.buf;
2025
2026        endp++;
2027        if (!*endp)
2028                die("Missing dest: %s", command_buf.buf);
2029
2030        d = endp;
2031        strbuf_reset(&d_uq);
2032        if (!unquote_c_style(&d_uq, d, &endp)) {
2033                if (*endp)
2034                        die("Garbage after dest in: %s", command_buf.buf);
2035                d = d_uq.buf;
2036        }
2037
2038        memset(&leaf, 0, sizeof(leaf));
2039        if (rename)
2040                tree_content_remove(&b->branch_tree, s, &leaf);
2041        else
2042                tree_content_get(&b->branch_tree, s, &leaf);
2043        if (!leaf.versions[1].mode)
2044                die("Path %s not in branch", s);
2045        tree_content_set(&b->branch_tree, d,
2046                leaf.versions[1].sha1,
2047                leaf.versions[1].mode,
2048                leaf.tree);
2049}
2050
2051static void note_change_n(struct branch *b)
2052{
2053        const char *p = command_buf.buf + 2;
2054        static struct strbuf uq = STRBUF_INIT;
2055        struct object_entry *oe = oe;
2056        struct branch *s;
2057        unsigned char sha1[20], commit_sha1[20];
2058        uint16_t inline_data = 0;
2059
2060        /* <dataref> or 'inline' */
2061        if (*p == ':') {
2062                char *x;
2063                oe = find_mark(strtoumax(p + 1, &x, 10));
2064                hashcpy(sha1, oe->sha1);
2065                p = x;
2066        } else if (!prefixcmp(p, "inline")) {
2067                inline_data = 1;
2068                p += 6;
2069        } else {
2070                if (get_sha1_hex(p, sha1))
2071                        die("Invalid SHA1: %s", command_buf.buf);
2072                oe = find_object(sha1);
2073                p += 40;
2074        }
2075        if (*p++ != ' ')
2076                die("Missing space after SHA1: %s", command_buf.buf);
2077
2078        /* <committish> */
2079        s = lookup_branch(p);
2080        if (s) {
2081                hashcpy(commit_sha1, s->sha1);
2082        } else if (*p == ':') {
2083                uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2084                struct object_entry *commit_oe = find_mark(commit_mark);
2085                if (commit_oe->type != OBJ_COMMIT)
2086                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2087                hashcpy(commit_sha1, commit_oe->sha1);
2088        } else if (!get_sha1(p, commit_sha1)) {
2089                unsigned long size;
2090                char *buf = read_object_with_reference(commit_sha1,
2091                        commit_type, &size, commit_sha1);
2092                if (!buf || size < 46)
2093                        die("Not a valid commit: %s", p);
2094                free(buf);
2095        } else
2096                die("Invalid ref name or SHA1 expression: %s", p);
2097
2098        if (inline_data) {
2099                static struct strbuf buf = STRBUF_INIT;
2100
2101                if (p != uq.buf) {
2102                        strbuf_addstr(&uq, p);
2103                        p = uq.buf;
2104                }
2105                read_next_command();
2106                parse_data(&buf);
2107                store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
2108        } else if (oe) {
2109                if (oe->type != OBJ_BLOB)
2110                        die("Not a blob (actually a %s): %s",
2111                                typename(oe->type), command_buf.buf);
2112        } else {
2113                enum object_type type = sha1_object_info(sha1, NULL);
2114                if (type < 0)
2115                        die("Blob not found: %s", command_buf.buf);
2116                if (type != OBJ_BLOB)
2117                        die("Not a blob (actually a %s): %s",
2118                            typename(type), command_buf.buf);
2119        }
2120
2121        tree_content_set(&b->branch_tree, sha1_to_hex(commit_sha1), sha1,
2122                S_IFREG | 0644, NULL);
2123}
2124
2125static void file_change_deleteall(struct branch *b)
2126{
2127        release_tree_content_recursive(b->branch_tree.tree);
2128        hashclr(b->branch_tree.versions[0].sha1);
2129        hashclr(b->branch_tree.versions[1].sha1);
2130        load_tree(&b->branch_tree);
2131}
2132
2133static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2134{
2135        if (!buf || size < 46)
2136                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2137        if (memcmp("tree ", buf, 5)
2138                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2139                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2140        hashcpy(b->branch_tree.versions[0].sha1,
2141                b->branch_tree.versions[1].sha1);
2142}
2143
2144static void parse_from_existing(struct branch *b)
2145{
2146        if (is_null_sha1(b->sha1)) {
2147                hashclr(b->branch_tree.versions[0].sha1);
2148                hashclr(b->branch_tree.versions[1].sha1);
2149        } else {
2150                unsigned long size;
2151                char *buf;
2152
2153                buf = read_object_with_reference(b->sha1,
2154                        commit_type, &size, b->sha1);
2155                parse_from_commit(b, buf, size);
2156                free(buf);
2157        }
2158}
2159
2160static int parse_from(struct branch *b)
2161{
2162        const char *from;
2163        struct branch *s;
2164
2165        if (prefixcmp(command_buf.buf, "from "))
2166                return 0;
2167
2168        if (b->branch_tree.tree) {
2169                release_tree_content_recursive(b->branch_tree.tree);
2170                b->branch_tree.tree = NULL;
2171        }
2172
2173        from = strchr(command_buf.buf, ' ') + 1;
2174        s = lookup_branch(from);
2175        if (b == s)
2176                die("Can't create a branch from itself: %s", b->name);
2177        else if (s) {
2178                unsigned char *t = s->branch_tree.versions[1].sha1;
2179                hashcpy(b->sha1, s->sha1);
2180                hashcpy(b->branch_tree.versions[0].sha1, t);
2181                hashcpy(b->branch_tree.versions[1].sha1, t);
2182        } else if (*from == ':') {
2183                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2184                struct object_entry *oe = find_mark(idnum);
2185                if (oe->type != OBJ_COMMIT)
2186                        die("Mark :%" PRIuMAX " not a commit", idnum);
2187                hashcpy(b->sha1, oe->sha1);
2188                if (oe->pack_id != MAX_PACK_ID) {
2189                        unsigned long size;
2190                        char *buf = gfi_unpack_entry(oe, &size);
2191                        parse_from_commit(b, buf, size);
2192                        free(buf);
2193                } else
2194                        parse_from_existing(b);
2195        } else if (!get_sha1(from, b->sha1))
2196                parse_from_existing(b);
2197        else
2198                die("Invalid ref name or SHA1 expression: %s", from);
2199
2200        read_next_command();
2201        return 1;
2202}
2203
2204static struct hash_list *parse_merge(unsigned int *count)
2205{
2206        struct hash_list *list = NULL, *n, *e = e;
2207        const char *from;
2208        struct branch *s;
2209
2210        *count = 0;
2211        while (!prefixcmp(command_buf.buf, "merge ")) {
2212                from = strchr(command_buf.buf, ' ') + 1;
2213                n = xmalloc(sizeof(*n));
2214                s = lookup_branch(from);
2215                if (s)
2216                        hashcpy(n->sha1, s->sha1);
2217                else if (*from == ':') {
2218                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2219                        struct object_entry *oe = find_mark(idnum);
2220                        if (oe->type != OBJ_COMMIT)
2221                                die("Mark :%" PRIuMAX " not a commit", idnum);
2222                        hashcpy(n->sha1, oe->sha1);
2223                } else if (!get_sha1(from, n->sha1)) {
2224                        unsigned long size;
2225                        char *buf = read_object_with_reference(n->sha1,
2226                                commit_type, &size, n->sha1);
2227                        if (!buf || size < 46)
2228                                die("Not a valid commit: %s", from);
2229                        free(buf);
2230                } else
2231                        die("Invalid ref name or SHA1 expression: %s", from);
2232
2233                n->next = NULL;
2234                if (list)
2235                        e->next = n;
2236                else
2237                        list = n;
2238                e = n;
2239                (*count)++;
2240                read_next_command();
2241        }
2242        return list;
2243}
2244
2245static void parse_new_commit(void)
2246{
2247        static struct strbuf msg = STRBUF_INIT;
2248        struct branch *b;
2249        char *sp;
2250        char *author = NULL;
2251        char *committer = NULL;
2252        struct hash_list *merge_list = NULL;
2253        unsigned int merge_count;
2254
2255        /* Obtain the branch name from the rest of our command */
2256        sp = strchr(command_buf.buf, ' ') + 1;
2257        b = lookup_branch(sp);
2258        if (!b)
2259                b = new_branch(sp);
2260
2261        read_next_command();
2262        parse_mark();
2263        if (!prefixcmp(command_buf.buf, "author ")) {
2264                author = parse_ident(command_buf.buf + 7);
2265                read_next_command();
2266        }
2267        if (!prefixcmp(command_buf.buf, "committer ")) {
2268                committer = parse_ident(command_buf.buf + 10);
2269                read_next_command();
2270        }
2271        if (!committer)
2272                die("Expected committer but didn't get one");
2273        parse_data(&msg);
2274        read_next_command();
2275        parse_from(b);
2276        merge_list = parse_merge(&merge_count);
2277
2278        /* ensure the branch is active/loaded */
2279        if (!b->branch_tree.tree || !max_active_branches) {
2280                unload_one_branch();
2281                load_branch(b);
2282        }
2283
2284        /* file_change* */
2285        while (command_buf.len > 0) {
2286                if (!prefixcmp(command_buf.buf, "M "))
2287                        file_change_m(b);
2288                else if (!prefixcmp(command_buf.buf, "D "))
2289                        file_change_d(b);
2290                else if (!prefixcmp(command_buf.buf, "R "))
2291                        file_change_cr(b, 1);
2292                else if (!prefixcmp(command_buf.buf, "C "))
2293                        file_change_cr(b, 0);
2294                else if (!prefixcmp(command_buf.buf, "N "))
2295                        note_change_n(b);
2296                else if (!strcmp("deleteall", command_buf.buf))
2297                        file_change_deleteall(b);
2298                else {
2299                        unread_command_buf = 1;
2300                        break;
2301                }
2302                if (read_next_command() == EOF)
2303                        break;
2304        }
2305
2306        /* build the tree and the commit */
2307        store_tree(&b->branch_tree);
2308        hashcpy(b->branch_tree.versions[0].sha1,
2309                b->branch_tree.versions[1].sha1);
2310
2311        strbuf_reset(&new_data);
2312        strbuf_addf(&new_data, "tree %s\n",
2313                sha1_to_hex(b->branch_tree.versions[1].sha1));
2314        if (!is_null_sha1(b->sha1))
2315                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2316        while (merge_list) {
2317                struct hash_list *next = merge_list->next;
2318                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2319                free(merge_list);
2320                merge_list = next;
2321        }
2322        strbuf_addf(&new_data,
2323                "author %s\n"
2324                "committer %s\n"
2325                "\n",
2326                author ? author : committer, committer);
2327        strbuf_addbuf(&new_data, &msg);
2328        free(author);
2329        free(committer);
2330
2331        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2332                b->pack_id = pack_id;
2333        b->last_commit = object_count_by_type[OBJ_COMMIT];
2334}
2335
2336static void parse_new_tag(void)
2337{
2338        static struct strbuf msg = STRBUF_INIT;
2339        char *sp;
2340        const char *from;
2341        char *tagger;
2342        struct branch *s;
2343        struct tag *t;
2344        uintmax_t from_mark = 0;
2345        unsigned char sha1[20];
2346
2347        /* Obtain the new tag name from the rest of our command */
2348        sp = strchr(command_buf.buf, ' ') + 1;
2349        t = pool_alloc(sizeof(struct tag));
2350        t->next_tag = NULL;
2351        t->name = pool_strdup(sp);
2352        if (last_tag)
2353                last_tag->next_tag = t;
2354        else
2355                first_tag = t;
2356        last_tag = t;
2357        read_next_command();
2358
2359        /* from ... */
2360        if (prefixcmp(command_buf.buf, "from "))
2361                die("Expected from command, got %s", command_buf.buf);
2362        from = strchr(command_buf.buf, ' ') + 1;
2363        s = lookup_branch(from);
2364        if (s) {
2365                hashcpy(sha1, s->sha1);
2366        } else if (*from == ':') {
2367                struct object_entry *oe;
2368                from_mark = strtoumax(from + 1, NULL, 10);
2369                oe = find_mark(from_mark);
2370                if (oe->type != OBJ_COMMIT)
2371                        die("Mark :%" PRIuMAX " not a commit", from_mark);
2372                hashcpy(sha1, oe->sha1);
2373        } else if (!get_sha1(from, sha1)) {
2374                unsigned long size;
2375                char *buf;
2376
2377                buf = read_object_with_reference(sha1,
2378                        commit_type, &size, sha1);
2379                if (!buf || size < 46)
2380                        die("Not a valid commit: %s", from);
2381                free(buf);
2382        } else
2383                die("Invalid ref name or SHA1 expression: %s", from);
2384        read_next_command();
2385
2386        /* tagger ... */
2387        if (!prefixcmp(command_buf.buf, "tagger ")) {
2388                tagger = parse_ident(command_buf.buf + 7);
2389                read_next_command();
2390        } else
2391                tagger = NULL;
2392
2393        /* tag payload/message */
2394        parse_data(&msg);
2395
2396        /* build the tag object */
2397        strbuf_reset(&new_data);
2398
2399        strbuf_addf(&new_data,
2400                    "object %s\n"
2401                    "type %s\n"
2402                    "tag %s\n",
2403                    sha1_to_hex(sha1), commit_type, t->name);
2404        if (tagger)
2405                strbuf_addf(&new_data,
2406                            "tagger %s\n", tagger);
2407        strbuf_addch(&new_data, '\n');
2408        strbuf_addbuf(&new_data, &msg);
2409        free(tagger);
2410
2411        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2412                t->pack_id = MAX_PACK_ID;
2413        else
2414                t->pack_id = pack_id;
2415}
2416
2417static void parse_reset_branch(void)
2418{
2419        struct branch *b;
2420        char *sp;
2421
2422        /* Obtain the branch name from the rest of our command */
2423        sp = strchr(command_buf.buf, ' ') + 1;
2424        b = lookup_branch(sp);
2425        if (b) {
2426                hashclr(b->sha1);
2427                hashclr(b->branch_tree.versions[0].sha1);
2428                hashclr(b->branch_tree.versions[1].sha1);
2429                if (b->branch_tree.tree) {
2430                        release_tree_content_recursive(b->branch_tree.tree);
2431                        b->branch_tree.tree = NULL;
2432                }
2433        }
2434        else
2435                b = new_branch(sp);
2436        read_next_command();
2437        parse_from(b);
2438        if (command_buf.len > 0)
2439                unread_command_buf = 1;
2440}
2441
2442static void parse_checkpoint(void)
2443{
2444        if (object_count) {
2445                cycle_packfile();
2446                dump_branches();
2447                dump_tags();
2448                dump_marks();
2449        }
2450        skip_optional_lf();
2451}
2452
2453static void parse_progress(void)
2454{
2455        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2456        fputc('\n', stdout);
2457        fflush(stdout);
2458        skip_optional_lf();
2459}
2460
2461static void option_import_marks(const char *marks)
2462{
2463        import_marks_file = xstrdup(marks);
2464}
2465
2466static void option_date_format(const char *fmt)
2467{
2468        if (!strcmp(fmt, "raw"))
2469                whenspec = WHENSPEC_RAW;
2470        else if (!strcmp(fmt, "rfc2822"))
2471                whenspec = WHENSPEC_RFC2822;
2472        else if (!strcmp(fmt, "now"))
2473                whenspec = WHENSPEC_NOW;
2474        else
2475                die("unknown --date-format argument %s", fmt);
2476}
2477
2478static void option_max_pack_size(const char *packsize)
2479{
2480        max_packsize = strtoumax(packsize, NULL, 0) * 1024 * 1024;
2481}
2482
2483static void option_depth(const char *depth)
2484{
2485        max_depth = strtoul(depth, NULL, 0);
2486        if (max_depth > MAX_DEPTH)
2487                die("--depth cannot exceed %u", MAX_DEPTH);
2488}
2489
2490static void option_active_branches(const char *branches)
2491{
2492        max_active_branches = strtoul(branches, NULL, 0);
2493}
2494
2495static void option_export_marks(const char *marks)
2496{
2497        export_marks_file = xstrdup(marks);
2498}
2499
2500static void option_export_pack_edges(const char *edges)
2501{
2502        if (pack_edges)
2503                fclose(pack_edges);
2504        pack_edges = fopen(edges, "a");
2505        if (!pack_edges)
2506                die_errno("Cannot open '%s'", edges);
2507}
2508
2509static void parse_one_option(const char *option)
2510{
2511        if (!prefixcmp(option, "date-format=")) {
2512                option_date_format(option + 12);
2513        } else if (!prefixcmp(option, "max-pack-size=")) {
2514                option_max_pack_size(option + 14);
2515        } else if (!prefixcmp(option, "depth=")) {
2516                option_depth(option + 6);
2517        } else if (!prefixcmp(option, "active-branches=")) {
2518                option_active_branches(option + 16);
2519        } else if (!prefixcmp(option, "import-marks=")) {
2520                option_import_marks(option + 13);
2521        } else if (!prefixcmp(option, "export-marks=")) {
2522                option_export_marks(option + 13);
2523        } else if (!prefixcmp(option, "export-pack-edges=")) {
2524                option_export_pack_edges(option + 18);
2525        } else if (!prefixcmp(option, "force")) {
2526                force_update = 1;
2527        } else if (!prefixcmp(option, "quiet")) {
2528                show_stats = 0;
2529        } else if (!prefixcmp(option, "stats")) {
2530                show_stats = 1;
2531        } else {
2532                die("Unsupported option: %s", option);
2533        }
2534}
2535
2536static int git_pack_config(const char *k, const char *v, void *cb)
2537{
2538        if (!strcmp(k, "pack.depth")) {
2539                max_depth = git_config_int(k, v);
2540                if (max_depth > MAX_DEPTH)
2541                        max_depth = MAX_DEPTH;
2542                return 0;
2543        }
2544        if (!strcmp(k, "pack.compression")) {
2545                int level = git_config_int(k, v);
2546                if (level == -1)
2547                        level = Z_DEFAULT_COMPRESSION;
2548                else if (level < 0 || level > Z_BEST_COMPRESSION)
2549                        die("bad pack compression level %d", level);
2550                pack_compression_level = level;
2551                pack_compression_seen = 1;
2552                return 0;
2553        }
2554        return git_default_config(k, v, cb);
2555}
2556
2557static const char fast_import_usage[] =
2558"git fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2559
2560int main(int argc, const char **argv)
2561{
2562        unsigned int i;
2563
2564        git_extract_argv0_path(argv[0]);
2565
2566        if (argc == 2 && !strcmp(argv[1], "-h"))
2567                usage(fast_import_usage);
2568
2569        setup_git_directory();
2570        git_config(git_pack_config, NULL);
2571        if (!pack_compression_seen && core_compression_seen)
2572                pack_compression_level = core_compression_level;
2573
2574        alloc_objects(object_entry_alloc);
2575        strbuf_init(&command_buf, 0);
2576        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2577        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2578        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2579        marks = pool_calloc(1, sizeof(struct mark_set));
2580
2581        for (i = 1; i < argc; i++) {
2582                const char *a = argv[i];
2583
2584                if (*a != '-' || !strcmp(a, "--"))
2585                        break;
2586
2587                parse_one_option(a + 2);
2588        }
2589        if (i != argc)
2590                usage(fast_import_usage);
2591        if (import_marks_file)
2592                read_marks();
2593
2594        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2595        for (i = 0; i < (cmd_save - 1); i++)
2596                rc_free[i].next = &rc_free[i + 1];
2597        rc_free[cmd_save - 1].next = NULL;
2598
2599        prepare_packed_git();
2600        start_packfile();
2601        set_die_routine(die_nicely);
2602        while (read_next_command() != EOF) {
2603                if (!strcmp("blob", command_buf.buf))
2604                        parse_new_blob();
2605                else if (!prefixcmp(command_buf.buf, "commit "))
2606                        parse_new_commit();
2607                else if (!prefixcmp(command_buf.buf, "tag "))
2608                        parse_new_tag();
2609                else if (!prefixcmp(command_buf.buf, "reset "))
2610                        parse_reset_branch();
2611                else if (!strcmp("checkpoint", command_buf.buf))
2612                        parse_checkpoint();
2613                else if (!prefixcmp(command_buf.buf, "progress "))
2614                        parse_progress();
2615                else
2616                        die("Unsupported command: %s", command_buf.buf);
2617        }
2618        end_packfile();
2619
2620        dump_branches();
2621        dump_tags();
2622        unkeep_all_packs();
2623        dump_marks();
2624
2625        if (pack_edges)
2626                fclose(pack_edges);
2627
2628        if (show_stats) {
2629                uintmax_t total_count = 0, duplicate_count = 0;
2630                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2631                        total_count += object_count_by_type[i];
2632                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2633                        duplicate_count += duplicate_count_by_type[i];
2634
2635                fprintf(stderr, "%s statistics:\n", argv[0]);
2636                fprintf(stderr, "---------------------------------------------------------------------\n");
2637                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2638                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2639                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]);
2640                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]);
2641                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]);
2642                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]);
2643                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2644                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2645                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2646                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2647                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2648                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2649                fprintf(stderr, "---------------------------------------------------------------------\n");
2650                pack_report();
2651                fprintf(stderr, "---------------------------------------------------------------------\n");
2652                fprintf(stderr, "\n");
2653        }
2654
2655        return failure ? 1 : 0;
2656}