fast-import.con commit Fix parsing of @{-1}@{1} (aa9c55b)
   1/*
   2Format of STDIN stream:
   3
   4  stream ::= cmd*;
   5
   6  cmd ::= new_blob
   7        | new_commit
   8        | new_tag
   9        | reset_branch
  10        | checkpoint
  11        | progress
  12        ;
  13
  14  new_blob ::= 'blob' lf
  15    mark?
  16    file_content;
  17  file_content ::= data;
  18
  19  new_commit ::= 'commit' sp ref_str lf
  20    mark?
  21    ('author' sp name '<' email '>' when lf)?
  22    'committer' sp name '<' email '>' when lf
  23    commit_msg
  24    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  25    ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
  26    file_change*
  27    lf?;
  28  commit_msg ::= data;
  29
  30  file_change ::= file_clr
  31    | file_del
  32    | file_rnm
  33    | file_cpy
  34    | file_obm
  35    | file_inm;
  36  file_clr ::= 'deleteall' lf;
  37  file_del ::= 'D' sp path_str lf;
  38  file_rnm ::= 'R' sp path_str sp path_str lf;
  39  file_cpy ::= 'C' sp path_str sp path_str lf;
  40  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  41  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  42    data;
  43
  44  new_tag ::= 'tag' sp tag_str lf
  45    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
  46    ('tagger' sp name '<' email '>' when lf)?
  47    tag_msg;
  48  tag_msg ::= data;
  49
  50  reset_branch ::= 'reset' sp ref_str lf
  51    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  52    lf?;
  53
  54  checkpoint ::= 'checkpoint' lf
  55    lf?;
  56
  57  progress ::= 'progress' sp not_lf* lf
  58    lf?;
  59
  60     # note: the first idnum in a stream should be 1 and subsequent
  61     # idnums should not have gaps between values as this will cause
  62     # the stream parser to reserve space for the gapped values.  An
  63     # idnum can be updated in the future to a new object by issuing
  64     # a new mark directive with the old idnum.
  65     #
  66  mark ::= 'mark' sp idnum lf;
  67  data ::= (delimited_data | exact_data)
  68    lf?;
  69
  70    # note: delim may be any string but must not contain lf.
  71    # data_line may contain any data but must not be exactly
  72    # delim.
  73  delimited_data ::= 'data' sp '<<' delim lf
  74    (data_line lf)*
  75    delim lf;
  76
  77     # note: declen indicates the length of binary_data in bytes.
  78     # declen does not include the lf preceeding the binary data.
  79     #
  80  exact_data ::= 'data' sp declen lf
  81    binary_data;
  82
  83     # note: quoted strings are C-style quoting supporting \c for
  84     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  85     # is the signed byte value in octal.  Note that the only
  86     # characters which must actually be escaped to protect the
  87     # stream formatting is: \, " and LF.  Otherwise these values
  88     # are UTF8.
  89     #
  90  ref_str     ::= ref;
  91  sha1exp_str ::= sha1exp;
  92  tag_str     ::= tag;
  93  path_str    ::= path    | '"' quoted(path)    '"' ;
  94  mode        ::= '100644' | '644'
  95                | '100755' | '755'
  96                | '120000'
  97                ;
  98
  99  declen ::= # unsigned 32 bit value, ascii base10 notation;
 100  bigint ::= # unsigned integer value, ascii base10 notation;
 101  binary_data ::= # file content, not interpreted;
 102
 103  when         ::= raw_when | rfc2822_when;
 104  raw_when     ::= ts sp tz;
 105  rfc2822_when ::= # Valid RFC 2822 date and time;
 106
 107  sp ::= # ASCII space character;
 108  lf ::= # ASCII newline (LF) character;
 109
 110     # note: a colon (':') must precede the numerical value assigned to
 111     # an idnum.  This is to distinguish it from a ref or tag name as
 112     # GIT does not permit ':' in ref or tag strings.
 113     #
 114  idnum   ::= ':' bigint;
 115  path    ::= # GIT style file path, e.g. "a/b/c";
 116  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 117  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 118  sha1exp ::= # Any valid GIT SHA1 expression;
 119  hexsha1 ::= # SHA1 in hexadecimal format;
 120
 121     # note: name and email are UTF8 strings, however name must not
 122     # contain '<' or lf and email must not contain any of the
 123     # following: '<', '>', lf.
 124     #
 125  name  ::= # valid GIT author/committer name;
 126  email ::= # valid GIT author/committer email;
 127  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 128  tz    ::= # GIT style timezone;
 129
 130     # note: comments may appear anywhere in the input, except
 131     # within a data command.  Any form of the data command
 132     # always escapes the related input from comment processing.
 133     #
 134     # In case it is not clear, the '#' that starts the comment
 135     # must be the first character on that the line (an lf have
 136     # preceeded it).
 137     #
 138  comment ::= '#' not_lf* lf;
 139  not_lf  ::= # Any byte that is not ASCII newline (LF);
 140*/
 141
 142#include "builtin.h"
 143#include "cache.h"
 144#include "object.h"
 145#include "blob.h"
 146#include "tree.h"
 147#include "commit.h"
 148#include "delta.h"
 149#include "pack.h"
 150#include "refs.h"
 151#include "csum-file.h"
 152#include "quote.h"
 153
 154#define PACK_ID_BITS 16
 155#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 156#define DEPTH_BITS 13
 157#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 158
 159struct object_entry
 160{
 161        struct object_entry *next;
 162        uint32_t offset;
 163        uint32_t type : TYPE_BITS,
 164                pack_id : PACK_ID_BITS,
 165                depth : DEPTH_BITS;
 166        unsigned char sha1[20];
 167};
 168
 169struct object_entry_pool
 170{
 171        struct object_entry_pool *next_pool;
 172        struct object_entry *next_free;
 173        struct object_entry *end;
 174        struct object_entry entries[FLEX_ARRAY]; /* more */
 175};
 176
 177struct mark_set
 178{
 179        union {
 180                struct object_entry *marked[1024];
 181                struct mark_set *sets[1024];
 182        } data;
 183        unsigned int shift;
 184};
 185
 186struct last_object
 187{
 188        struct strbuf data;
 189        uint32_t offset;
 190        unsigned int depth;
 191        unsigned no_swap : 1;
 192};
 193
 194struct mem_pool
 195{
 196        struct mem_pool *next_pool;
 197        char *next_free;
 198        char *end;
 199        uintmax_t space[FLEX_ARRAY]; /* more */
 200};
 201
 202struct atom_str
 203{
 204        struct atom_str *next_atom;
 205        unsigned short str_len;
 206        char str_dat[FLEX_ARRAY]; /* more */
 207};
 208
 209struct tree_content;
 210struct tree_entry
 211{
 212        struct tree_content *tree;
 213        struct atom_str* name;
 214        struct tree_entry_ms
 215        {
 216                uint16_t mode;
 217                unsigned char sha1[20];
 218        } versions[2];
 219};
 220
 221struct tree_content
 222{
 223        unsigned int entry_capacity; /* must match avail_tree_content */
 224        unsigned int entry_count;
 225        unsigned int delta_depth;
 226        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 227};
 228
 229struct avail_tree_content
 230{
 231        unsigned int entry_capacity; /* must match tree_content */
 232        struct avail_tree_content *next_avail;
 233};
 234
 235struct branch
 236{
 237        struct branch *table_next_branch;
 238        struct branch *active_next_branch;
 239        const char *name;
 240        struct tree_entry branch_tree;
 241        uintmax_t last_commit;
 242        unsigned active : 1;
 243        unsigned pack_id : PACK_ID_BITS;
 244        unsigned char sha1[20];
 245};
 246
 247struct tag
 248{
 249        struct tag *next_tag;
 250        const char *name;
 251        unsigned int pack_id;
 252        unsigned char sha1[20];
 253};
 254
 255struct hash_list
 256{
 257        struct hash_list *next;
 258        unsigned char sha1[20];
 259};
 260
 261typedef enum {
 262        WHENSPEC_RAW = 1,
 263        WHENSPEC_RFC2822,
 264        WHENSPEC_NOW,
 265} whenspec_type;
 266
 267struct recent_command
 268{
 269        struct recent_command *prev;
 270        struct recent_command *next;
 271        char *buf;
 272};
 273
 274/* Configured limits on output */
 275static unsigned long max_depth = 10;
 276static off_t max_packsize = (1LL << 32) - 1;
 277static int force_update;
 278static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 279static int pack_compression_seen;
 280
 281/* Stats and misc. counters */
 282static uintmax_t alloc_count;
 283static uintmax_t marks_set_count;
 284static uintmax_t object_count_by_type[1 << TYPE_BITS];
 285static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 286static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 287static unsigned long object_count;
 288static unsigned long branch_count;
 289static unsigned long branch_load_count;
 290static int failure;
 291static FILE *pack_edges;
 292
 293/* Memory pools */
 294static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 295static size_t total_allocd;
 296static struct mem_pool *mem_pool;
 297
 298/* Atom management */
 299static unsigned int atom_table_sz = 4451;
 300static unsigned int atom_cnt;
 301static struct atom_str **atom_table;
 302
 303/* The .pack file being generated */
 304static unsigned int pack_id;
 305static struct packed_git *pack_data;
 306static struct packed_git **all_packs;
 307static unsigned long pack_size;
 308
 309/* Table of objects we've written. */
 310static unsigned int object_entry_alloc = 5000;
 311static struct object_entry_pool *blocks;
 312static struct object_entry *object_table[1 << 16];
 313static struct mark_set *marks;
 314static const char* mark_file;
 315
 316/* Our last blob */
 317static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 318
 319/* Tree management */
 320static unsigned int tree_entry_alloc = 1000;
 321static void *avail_tree_entry;
 322static unsigned int avail_tree_table_sz = 100;
 323static struct avail_tree_content **avail_tree_table;
 324static struct strbuf old_tree = STRBUF_INIT;
 325static struct strbuf new_tree = STRBUF_INIT;
 326
 327/* Branch data */
 328static unsigned long max_active_branches = 5;
 329static unsigned long cur_active_branches;
 330static unsigned long branch_table_sz = 1039;
 331static struct branch **branch_table;
 332static struct branch *active_branches;
 333
 334/* Tag data */
 335static struct tag *first_tag;
 336static struct tag *last_tag;
 337
 338/* Input stream parsing */
 339static whenspec_type whenspec = WHENSPEC_RAW;
 340static struct strbuf command_buf = STRBUF_INIT;
 341static int unread_command_buf;
 342static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 343static struct recent_command *cmd_tail = &cmd_hist;
 344static struct recent_command *rc_free;
 345static unsigned int cmd_save = 100;
 346static uintmax_t next_mark;
 347static struct strbuf new_data = STRBUF_INIT;
 348
 349static void write_branch_report(FILE *rpt, struct branch *b)
 350{
 351        fprintf(rpt, "%s:\n", b->name);
 352
 353        fprintf(rpt, "  status      :");
 354        if (b->active)
 355                fputs(" active", rpt);
 356        if (b->branch_tree.tree)
 357                fputs(" loaded", rpt);
 358        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 359                fputs(" dirty", rpt);
 360        fputc('\n', rpt);
 361
 362        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 363        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 364        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 365        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 366
 367        fputs("  last pack   : ", rpt);
 368        if (b->pack_id < MAX_PACK_ID)
 369                fprintf(rpt, "%u", b->pack_id);
 370        fputc('\n', rpt);
 371
 372        fputc('\n', rpt);
 373}
 374
 375static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 376
 377static void write_crash_report(const char *err)
 378{
 379        char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 380        FILE *rpt = fopen(loc, "w");
 381        struct branch *b;
 382        unsigned long lu;
 383        struct recent_command *rc;
 384
 385        if (!rpt) {
 386                error("can't write crash report %s: %s", loc, strerror(errno));
 387                return;
 388        }
 389
 390        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 391
 392        fprintf(rpt, "fast-import crash report:\n");
 393        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 394        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 395        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 396        fputc('\n', rpt);
 397
 398        fputs("fatal: ", rpt);
 399        fputs(err, rpt);
 400        fputc('\n', rpt);
 401
 402        fputc('\n', rpt);
 403        fputs("Most Recent Commands Before Crash\n", rpt);
 404        fputs("---------------------------------\n", rpt);
 405        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 406                if (rc->next == &cmd_hist)
 407                        fputs("* ", rpt);
 408                else
 409                        fputs("  ", rpt);
 410                fputs(rc->buf, rpt);
 411                fputc('\n', rpt);
 412        }
 413
 414        fputc('\n', rpt);
 415        fputs("Active Branch LRU\n", rpt);
 416        fputs("-----------------\n", rpt);
 417        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 418                cur_active_branches,
 419                max_active_branches);
 420        fputc('\n', rpt);
 421        fputs("  pos  clock name\n", rpt);
 422        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 423        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 424                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 425                        ++lu, b->last_commit, b->name);
 426
 427        fputc('\n', rpt);
 428        fputs("Inactive Branches\n", rpt);
 429        fputs("-----------------\n", rpt);
 430        for (lu = 0; lu < branch_table_sz; lu++) {
 431                for (b = branch_table[lu]; b; b = b->table_next_branch)
 432                        write_branch_report(rpt, b);
 433        }
 434
 435        if (first_tag) {
 436                struct tag *tg;
 437                fputc('\n', rpt);
 438                fputs("Annotated Tags\n", rpt);
 439                fputs("--------------\n", rpt);
 440                for (tg = first_tag; tg; tg = tg->next_tag) {
 441                        fputs(sha1_to_hex(tg->sha1), rpt);
 442                        fputc(' ', rpt);
 443                        fputs(tg->name, rpt);
 444                        fputc('\n', rpt);
 445                }
 446        }
 447
 448        fputc('\n', rpt);
 449        fputs("Marks\n", rpt);
 450        fputs("-----\n", rpt);
 451        if (mark_file)
 452                fprintf(rpt, "  exported to %s\n", mark_file);
 453        else
 454                dump_marks_helper(rpt, 0, marks);
 455
 456        fputc('\n', rpt);
 457        fputs("-------------------\n", rpt);
 458        fputs("END OF CRASH REPORT\n", rpt);
 459        fclose(rpt);
 460}
 461
 462static void end_packfile(void);
 463static void unkeep_all_packs(void);
 464static void dump_marks(void);
 465
 466static NORETURN void die_nicely(const char *err, va_list params)
 467{
 468        static int zombie;
 469        char message[2 * PATH_MAX];
 470
 471        vsnprintf(message, sizeof(message), err, params);
 472        fputs("fatal: ", stderr);
 473        fputs(message, stderr);
 474        fputc('\n', stderr);
 475
 476        if (!zombie) {
 477                zombie = 1;
 478                write_crash_report(message);
 479                end_packfile();
 480                unkeep_all_packs();
 481                dump_marks();
 482        }
 483        exit(128);
 484}
 485
 486static void alloc_objects(unsigned int cnt)
 487{
 488        struct object_entry_pool *b;
 489
 490        b = xmalloc(sizeof(struct object_entry_pool)
 491                + cnt * sizeof(struct object_entry));
 492        b->next_pool = blocks;
 493        b->next_free = b->entries;
 494        b->end = b->entries + cnt;
 495        blocks = b;
 496        alloc_count += cnt;
 497}
 498
 499static struct object_entry *new_object(unsigned char *sha1)
 500{
 501        struct object_entry *e;
 502
 503        if (blocks->next_free == blocks->end)
 504                alloc_objects(object_entry_alloc);
 505
 506        e = blocks->next_free++;
 507        hashcpy(e->sha1, sha1);
 508        return e;
 509}
 510
 511static struct object_entry *find_object(unsigned char *sha1)
 512{
 513        unsigned int h = sha1[0] << 8 | sha1[1];
 514        struct object_entry *e;
 515        for (e = object_table[h]; e; e = e->next)
 516                if (!hashcmp(sha1, e->sha1))
 517                        return e;
 518        return NULL;
 519}
 520
 521static struct object_entry *insert_object(unsigned char *sha1)
 522{
 523        unsigned int h = sha1[0] << 8 | sha1[1];
 524        struct object_entry *e = object_table[h];
 525        struct object_entry *p = NULL;
 526
 527        while (e) {
 528                if (!hashcmp(sha1, e->sha1))
 529                        return e;
 530                p = e;
 531                e = e->next;
 532        }
 533
 534        e = new_object(sha1);
 535        e->next = NULL;
 536        e->offset = 0;
 537        if (p)
 538                p->next = e;
 539        else
 540                object_table[h] = e;
 541        return e;
 542}
 543
 544static unsigned int hc_str(const char *s, size_t len)
 545{
 546        unsigned int r = 0;
 547        while (len-- > 0)
 548                r = r * 31 + *s++;
 549        return r;
 550}
 551
 552static void *pool_alloc(size_t len)
 553{
 554        struct mem_pool *p;
 555        void *r;
 556
 557        /* round up to a 'uintmax_t' alignment */
 558        if (len & (sizeof(uintmax_t) - 1))
 559                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 560
 561        for (p = mem_pool; p; p = p->next_pool)
 562                if ((p->end - p->next_free >= len))
 563                        break;
 564
 565        if (!p) {
 566                if (len >= (mem_pool_alloc/2)) {
 567                        total_allocd += len;
 568                        return xmalloc(len);
 569                }
 570                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 571                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 572                p->next_pool = mem_pool;
 573                p->next_free = (char *) p->space;
 574                p->end = p->next_free + mem_pool_alloc;
 575                mem_pool = p;
 576        }
 577
 578        r = p->next_free;
 579        p->next_free += len;
 580        return r;
 581}
 582
 583static void *pool_calloc(size_t count, size_t size)
 584{
 585        size_t len = count * size;
 586        void *r = pool_alloc(len);
 587        memset(r, 0, len);
 588        return r;
 589}
 590
 591static char *pool_strdup(const char *s)
 592{
 593        char *r = pool_alloc(strlen(s) + 1);
 594        strcpy(r, s);
 595        return r;
 596}
 597
 598static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 599{
 600        struct mark_set *s = marks;
 601        while ((idnum >> s->shift) >= 1024) {
 602                s = pool_calloc(1, sizeof(struct mark_set));
 603                s->shift = marks->shift + 10;
 604                s->data.sets[0] = marks;
 605                marks = s;
 606        }
 607        while (s->shift) {
 608                uintmax_t i = idnum >> s->shift;
 609                idnum -= i << s->shift;
 610                if (!s->data.sets[i]) {
 611                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 612                        s->data.sets[i]->shift = s->shift - 10;
 613                }
 614                s = s->data.sets[i];
 615        }
 616        if (!s->data.marked[idnum])
 617                marks_set_count++;
 618        s->data.marked[idnum] = oe;
 619}
 620
 621static struct object_entry *find_mark(uintmax_t idnum)
 622{
 623        uintmax_t orig_idnum = idnum;
 624        struct mark_set *s = marks;
 625        struct object_entry *oe = NULL;
 626        if ((idnum >> s->shift) < 1024) {
 627                while (s && s->shift) {
 628                        uintmax_t i = idnum >> s->shift;
 629                        idnum -= i << s->shift;
 630                        s = s->data.sets[i];
 631                }
 632                if (s)
 633                        oe = s->data.marked[idnum];
 634        }
 635        if (!oe)
 636                die("mark :%" PRIuMAX " not declared", orig_idnum);
 637        return oe;
 638}
 639
 640static struct atom_str *to_atom(const char *s, unsigned short len)
 641{
 642        unsigned int hc = hc_str(s, len) % atom_table_sz;
 643        struct atom_str *c;
 644
 645        for (c = atom_table[hc]; c; c = c->next_atom)
 646                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 647                        return c;
 648
 649        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 650        c->str_len = len;
 651        strncpy(c->str_dat, s, len);
 652        c->str_dat[len] = 0;
 653        c->next_atom = atom_table[hc];
 654        atom_table[hc] = c;
 655        atom_cnt++;
 656        return c;
 657}
 658
 659static struct branch *lookup_branch(const char *name)
 660{
 661        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 662        struct branch *b;
 663
 664        for (b = branch_table[hc]; b; b = b->table_next_branch)
 665                if (!strcmp(name, b->name))
 666                        return b;
 667        return NULL;
 668}
 669
 670static struct branch *new_branch(const char *name)
 671{
 672        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 673        struct branch* b = lookup_branch(name);
 674
 675        if (b)
 676                die("Invalid attempt to create duplicate branch: %s", name);
 677        switch (check_ref_format(name)) {
 678        case 0: break; /* its valid */
 679        case CHECK_REF_FORMAT_ONELEVEL:
 680                break; /* valid, but too few '/', allow anyway */
 681        default:
 682                die("Branch name doesn't conform to GIT standards: %s", name);
 683        }
 684
 685        b = pool_calloc(1, sizeof(struct branch));
 686        b->name = pool_strdup(name);
 687        b->table_next_branch = branch_table[hc];
 688        b->branch_tree.versions[0].mode = S_IFDIR;
 689        b->branch_tree.versions[1].mode = S_IFDIR;
 690        b->active = 0;
 691        b->pack_id = MAX_PACK_ID;
 692        branch_table[hc] = b;
 693        branch_count++;
 694        return b;
 695}
 696
 697static unsigned int hc_entries(unsigned int cnt)
 698{
 699        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 700        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 701}
 702
 703static struct tree_content *new_tree_content(unsigned int cnt)
 704{
 705        struct avail_tree_content *f, *l = NULL;
 706        struct tree_content *t;
 707        unsigned int hc = hc_entries(cnt);
 708
 709        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 710                if (f->entry_capacity >= cnt)
 711                        break;
 712
 713        if (f) {
 714                if (l)
 715                        l->next_avail = f->next_avail;
 716                else
 717                        avail_tree_table[hc] = f->next_avail;
 718        } else {
 719                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 720                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 721                f->entry_capacity = cnt;
 722        }
 723
 724        t = (struct tree_content*)f;
 725        t->entry_count = 0;
 726        t->delta_depth = 0;
 727        return t;
 728}
 729
 730static void release_tree_entry(struct tree_entry *e);
 731static void release_tree_content(struct tree_content *t)
 732{
 733        struct avail_tree_content *f = (struct avail_tree_content*)t;
 734        unsigned int hc = hc_entries(f->entry_capacity);
 735        f->next_avail = avail_tree_table[hc];
 736        avail_tree_table[hc] = f;
 737}
 738
 739static void release_tree_content_recursive(struct tree_content *t)
 740{
 741        unsigned int i;
 742        for (i = 0; i < t->entry_count; i++)
 743                release_tree_entry(t->entries[i]);
 744        release_tree_content(t);
 745}
 746
 747static struct tree_content *grow_tree_content(
 748        struct tree_content *t,
 749        int amt)
 750{
 751        struct tree_content *r = new_tree_content(t->entry_count + amt);
 752        r->entry_count = t->entry_count;
 753        r->delta_depth = t->delta_depth;
 754        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 755        release_tree_content(t);
 756        return r;
 757}
 758
 759static struct tree_entry *new_tree_entry(void)
 760{
 761        struct tree_entry *e;
 762
 763        if (!avail_tree_entry) {
 764                unsigned int n = tree_entry_alloc;
 765                total_allocd += n * sizeof(struct tree_entry);
 766                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 767                while (n-- > 1) {
 768                        *((void**)e) = e + 1;
 769                        e++;
 770                }
 771                *((void**)e) = NULL;
 772        }
 773
 774        e = avail_tree_entry;
 775        avail_tree_entry = *((void**)e);
 776        return e;
 777}
 778
 779static void release_tree_entry(struct tree_entry *e)
 780{
 781        if (e->tree)
 782                release_tree_content_recursive(e->tree);
 783        *((void**)e) = avail_tree_entry;
 784        avail_tree_entry = e;
 785}
 786
 787static struct tree_content *dup_tree_content(struct tree_content *s)
 788{
 789        struct tree_content *d;
 790        struct tree_entry *a, *b;
 791        unsigned int i;
 792
 793        if (!s)
 794                return NULL;
 795        d = new_tree_content(s->entry_count);
 796        for (i = 0; i < s->entry_count; i++) {
 797                a = s->entries[i];
 798                b = new_tree_entry();
 799                memcpy(b, a, sizeof(*a));
 800                if (a->tree && is_null_sha1(b->versions[1].sha1))
 801                        b->tree = dup_tree_content(a->tree);
 802                else
 803                        b->tree = NULL;
 804                d->entries[i] = b;
 805        }
 806        d->entry_count = s->entry_count;
 807        d->delta_depth = s->delta_depth;
 808
 809        return d;
 810}
 811
 812static void start_packfile(void)
 813{
 814        static char tmpfile[PATH_MAX];
 815        struct packed_git *p;
 816        struct pack_header hdr;
 817        int pack_fd;
 818
 819        snprintf(tmpfile, sizeof(tmpfile),
 820                "%s/pack/tmp_pack_XXXXXX", get_object_directory());
 821        pack_fd = xmkstemp(tmpfile);
 822        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 823        strcpy(p->pack_name, tmpfile);
 824        p->pack_fd = pack_fd;
 825
 826        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 827        hdr.hdr_version = htonl(2);
 828        hdr.hdr_entries = 0;
 829        write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 830
 831        pack_data = p;
 832        pack_size = sizeof(hdr);
 833        object_count = 0;
 834
 835        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 836        all_packs[pack_id] = p;
 837}
 838
 839static int oecmp (const void *a_, const void *b_)
 840{
 841        struct object_entry *a = *((struct object_entry**)a_);
 842        struct object_entry *b = *((struct object_entry**)b_);
 843        return hashcmp(a->sha1, b->sha1);
 844}
 845
 846static char *create_index(void)
 847{
 848        static char tmpfile[PATH_MAX];
 849        git_SHA_CTX ctx;
 850        struct sha1file *f;
 851        struct object_entry **idx, **c, **last, *e;
 852        struct object_entry_pool *o;
 853        uint32_t array[256];
 854        int i, idx_fd;
 855
 856        /* Build the sorted table of object IDs. */
 857        idx = xmalloc(object_count * sizeof(struct object_entry*));
 858        c = idx;
 859        for (o = blocks; o; o = o->next_pool)
 860                for (e = o->next_free; e-- != o->entries;)
 861                        if (pack_id == e->pack_id)
 862                                *c++ = e;
 863        last = idx + object_count;
 864        if (c != last)
 865                die("internal consistency error creating the index");
 866        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 867
 868        /* Generate the fan-out array. */
 869        c = idx;
 870        for (i = 0; i < 256; i++) {
 871                struct object_entry **next = c;;
 872                while (next < last) {
 873                        if ((*next)->sha1[0] != i)
 874                                break;
 875                        next++;
 876                }
 877                array[i] = htonl(next - idx);
 878                c = next;
 879        }
 880
 881        snprintf(tmpfile, sizeof(tmpfile),
 882                "%s/pack/tmp_idx_XXXXXX", get_object_directory());
 883        idx_fd = xmkstemp(tmpfile);
 884        f = sha1fd(idx_fd, tmpfile);
 885        sha1write(f, array, 256 * sizeof(int));
 886        git_SHA1_Init(&ctx);
 887        for (c = idx; c != last; c++) {
 888                uint32_t offset = htonl((*c)->offset);
 889                sha1write(f, &offset, 4);
 890                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 891                git_SHA1_Update(&ctx, (*c)->sha1, 20);
 892        }
 893        sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
 894        sha1close(f, NULL, CSUM_FSYNC);
 895        free(idx);
 896        git_SHA1_Final(pack_data->sha1, &ctx);
 897        return tmpfile;
 898}
 899
 900static char *keep_pack(char *curr_index_name)
 901{
 902        static char name[PATH_MAX];
 903        static const char *keep_msg = "fast-import";
 904        int keep_fd;
 905
 906        chmod(pack_data->pack_name, 0444);
 907        chmod(curr_index_name, 0444);
 908
 909        snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 910                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 911        keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
 912        if (keep_fd < 0)
 913                die("cannot create keep file");
 914        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 915        if (close(keep_fd))
 916                die("failed to write keep file");
 917
 918        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 919                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 920        if (move_temp_to_file(pack_data->pack_name, name))
 921                die("cannot store pack file");
 922
 923        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 924                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 925        if (move_temp_to_file(curr_index_name, name))
 926                die("cannot store index file");
 927        return name;
 928}
 929
 930static void unkeep_all_packs(void)
 931{
 932        static char name[PATH_MAX];
 933        int k;
 934
 935        for (k = 0; k < pack_id; k++) {
 936                struct packed_git *p = all_packs[k];
 937                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 938                         get_object_directory(), sha1_to_hex(p->sha1));
 939                unlink(name);
 940        }
 941}
 942
 943static void end_packfile(void)
 944{
 945        struct packed_git *old_p = pack_data, *new_p;
 946
 947        if (object_count) {
 948                char *idx_name;
 949                int i;
 950                struct branch *b;
 951                struct tag *t;
 952
 953                close_pack_windows(pack_data);
 954                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 955                                    pack_data->pack_name, object_count,
 956                                    NULL, 0);
 957                close(pack_data->pack_fd);
 958                idx_name = keep_pack(create_index());
 959
 960                /* Register the packfile with core git's machinary. */
 961                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 962                if (!new_p)
 963                        die("core git rejected index %s", idx_name);
 964                all_packs[pack_id] = new_p;
 965                install_packed_git(new_p);
 966
 967                /* Print the boundary */
 968                if (pack_edges) {
 969                        fprintf(pack_edges, "%s:", new_p->pack_name);
 970                        for (i = 0; i < branch_table_sz; i++) {
 971                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 972                                        if (b->pack_id == pack_id)
 973                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 974                                }
 975                        }
 976                        for (t = first_tag; t; t = t->next_tag) {
 977                                if (t->pack_id == pack_id)
 978                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 979                        }
 980                        fputc('\n', pack_edges);
 981                        fflush(pack_edges);
 982                }
 983
 984                pack_id++;
 985        }
 986        else {
 987                close(old_p->pack_fd);
 988                unlink(old_p->pack_name);
 989        }
 990        free(old_p);
 991
 992        /* We can't carry a delta across packfiles. */
 993        strbuf_release(&last_blob.data);
 994        last_blob.offset = 0;
 995        last_blob.depth = 0;
 996}
 997
 998static void cycle_packfile(void)
 999{
1000        end_packfile();
1001        start_packfile();
1002}
1003
1004static size_t encode_header(
1005        enum object_type type,
1006        size_t size,
1007        unsigned char *hdr)
1008{
1009        int n = 1;
1010        unsigned char c;
1011
1012        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
1013                die("bad type %d", type);
1014
1015        c = (type << 4) | (size & 15);
1016        size >>= 4;
1017        while (size) {
1018                *hdr++ = c | 0x80;
1019                c = size & 0x7f;
1020                size >>= 7;
1021                n++;
1022        }
1023        *hdr = c;
1024        return n;
1025}
1026
1027static int store_object(
1028        enum object_type type,
1029        struct strbuf *dat,
1030        struct last_object *last,
1031        unsigned char *sha1out,
1032        uintmax_t mark)
1033{
1034        void *out, *delta;
1035        struct object_entry *e;
1036        unsigned char hdr[96];
1037        unsigned char sha1[20];
1038        unsigned long hdrlen, deltalen;
1039        git_SHA_CTX c;
1040        z_stream s;
1041
1042        hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
1043                (unsigned long)dat->len) + 1;
1044        git_SHA1_Init(&c);
1045        git_SHA1_Update(&c, hdr, hdrlen);
1046        git_SHA1_Update(&c, dat->buf, dat->len);
1047        git_SHA1_Final(sha1, &c);
1048        if (sha1out)
1049                hashcpy(sha1out, sha1);
1050
1051        e = insert_object(sha1);
1052        if (mark)
1053                insert_mark(mark, e);
1054        if (e->offset) {
1055                duplicate_count_by_type[type]++;
1056                return 1;
1057        } else if (find_sha1_pack(sha1, packed_git)) {
1058                e->type = type;
1059                e->pack_id = MAX_PACK_ID;
1060                e->offset = 1; /* just not zero! */
1061                duplicate_count_by_type[type]++;
1062                return 1;
1063        }
1064
1065        if (last && last->data.buf && last->depth < max_depth) {
1066                delta = diff_delta(last->data.buf, last->data.len,
1067                        dat->buf, dat->len,
1068                        &deltalen, 0);
1069                if (delta && deltalen >= dat->len) {
1070                        free(delta);
1071                        delta = NULL;
1072                }
1073        } else
1074                delta = NULL;
1075
1076        memset(&s, 0, sizeof(s));
1077        deflateInit(&s, pack_compression_level);
1078        if (delta) {
1079                s.next_in = delta;
1080                s.avail_in = deltalen;
1081        } else {
1082                s.next_in = (void *)dat->buf;
1083                s.avail_in = dat->len;
1084        }
1085        s.avail_out = deflateBound(&s, s.avail_in);
1086        s.next_out = out = xmalloc(s.avail_out);
1087        while (deflate(&s, Z_FINISH) == Z_OK)
1088                /* nothing */;
1089        deflateEnd(&s);
1090
1091        /* Determine if we should auto-checkpoint. */
1092        if ((pack_size + 60 + s.total_out) > max_packsize
1093                || (pack_size + 60 + s.total_out) < pack_size) {
1094
1095                /* This new object needs to *not* have the current pack_id. */
1096                e->pack_id = pack_id + 1;
1097                cycle_packfile();
1098
1099                /* We cannot carry a delta into the new pack. */
1100                if (delta) {
1101                        free(delta);
1102                        delta = NULL;
1103
1104                        memset(&s, 0, sizeof(s));
1105                        deflateInit(&s, pack_compression_level);
1106                        s.next_in = (void *)dat->buf;
1107                        s.avail_in = dat->len;
1108                        s.avail_out = deflateBound(&s, s.avail_in);
1109                        s.next_out = out = xrealloc(out, s.avail_out);
1110                        while (deflate(&s, Z_FINISH) == Z_OK)
1111                                /* nothing */;
1112                        deflateEnd(&s);
1113                }
1114        }
1115
1116        e->type = type;
1117        e->pack_id = pack_id;
1118        e->offset = pack_size;
1119        object_count++;
1120        object_count_by_type[type]++;
1121
1122        if (delta) {
1123                unsigned long ofs = e->offset - last->offset;
1124                unsigned pos = sizeof(hdr) - 1;
1125
1126                delta_count_by_type[type]++;
1127                e->depth = last->depth + 1;
1128
1129                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1130                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1131                pack_size += hdrlen;
1132
1133                hdr[pos] = ofs & 127;
1134                while (ofs >>= 7)
1135                        hdr[--pos] = 128 | (--ofs & 127);
1136                write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1137                pack_size += sizeof(hdr) - pos;
1138        } else {
1139                e->depth = 0;
1140                hdrlen = encode_header(type, dat->len, hdr);
1141                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1142                pack_size += hdrlen;
1143        }
1144
1145        write_or_die(pack_data->pack_fd, out, s.total_out);
1146        pack_size += s.total_out;
1147
1148        free(out);
1149        free(delta);
1150        if (last) {
1151                if (last->no_swap) {
1152                        last->data = *dat;
1153                } else {
1154                        strbuf_swap(&last->data, dat);
1155                }
1156                last->offset = e->offset;
1157                last->depth = e->depth;
1158        }
1159        return 0;
1160}
1161
1162/* All calls must be guarded by find_object() or find_mark() to
1163 * ensure the 'struct object_entry' passed was written by this
1164 * process instance.  We unpack the entry by the offset, avoiding
1165 * the need for the corresponding .idx file.  This unpacking rule
1166 * works because we only use OBJ_REF_DELTA within the packfiles
1167 * created by fast-import.
1168 *
1169 * oe must not be NULL.  Such an oe usually comes from giving
1170 * an unknown SHA-1 to find_object() or an undefined mark to
1171 * find_mark().  Callers must test for this condition and use
1172 * the standard read_sha1_file() when it happens.
1173 *
1174 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1175 * find_mark(), where the mark was reloaded from an existing marks
1176 * file and is referencing an object that this fast-import process
1177 * instance did not write out to a packfile.  Callers must test for
1178 * this condition and use read_sha1_file() instead.
1179 */
1180static void *gfi_unpack_entry(
1181        struct object_entry *oe,
1182        unsigned long *sizep)
1183{
1184        enum object_type type;
1185        struct packed_git *p = all_packs[oe->pack_id];
1186        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1187                /* The object is stored in the packfile we are writing to
1188                 * and we have modified it since the last time we scanned
1189                 * back to read a previously written object.  If an old
1190                 * window covered [p->pack_size, p->pack_size + 20) its
1191                 * data is stale and is not valid.  Closing all windows
1192                 * and updating the packfile length ensures we can read
1193                 * the newly written data.
1194                 */
1195                close_pack_windows(p);
1196
1197                /* We have to offer 20 bytes additional on the end of
1198                 * the packfile as the core unpacker code assumes the
1199                 * footer is present at the file end and must promise
1200                 * at least 20 bytes within any window it maps.  But
1201                 * we don't actually create the footer here.
1202                 */
1203                p->pack_size = pack_size + 20;
1204        }
1205        return unpack_entry(p, oe->offset, &type, sizep);
1206}
1207
1208static const char *get_mode(const char *str, uint16_t *modep)
1209{
1210        unsigned char c;
1211        uint16_t mode = 0;
1212
1213        while ((c = *str++) != ' ') {
1214                if (c < '0' || c > '7')
1215                        return NULL;
1216                mode = (mode << 3) + (c - '0');
1217        }
1218        *modep = mode;
1219        return str;
1220}
1221
1222static void load_tree(struct tree_entry *root)
1223{
1224        unsigned char* sha1 = root->versions[1].sha1;
1225        struct object_entry *myoe;
1226        struct tree_content *t;
1227        unsigned long size;
1228        char *buf;
1229        const char *c;
1230
1231        root->tree = t = new_tree_content(8);
1232        if (is_null_sha1(sha1))
1233                return;
1234
1235        myoe = find_object(sha1);
1236        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1237                if (myoe->type != OBJ_TREE)
1238                        die("Not a tree: %s", sha1_to_hex(sha1));
1239                t->delta_depth = myoe->depth;
1240                buf = gfi_unpack_entry(myoe, &size);
1241                if (!buf)
1242                        die("Can't load tree %s", sha1_to_hex(sha1));
1243        } else {
1244                enum object_type type;
1245                buf = read_sha1_file(sha1, &type, &size);
1246                if (!buf || type != OBJ_TREE)
1247                        die("Can't load tree %s", sha1_to_hex(sha1));
1248        }
1249
1250        c = buf;
1251        while (c != (buf + size)) {
1252                struct tree_entry *e = new_tree_entry();
1253
1254                if (t->entry_count == t->entry_capacity)
1255                        root->tree = t = grow_tree_content(t, t->entry_count);
1256                t->entries[t->entry_count++] = e;
1257
1258                e->tree = NULL;
1259                c = get_mode(c, &e->versions[1].mode);
1260                if (!c)
1261                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1262                e->versions[0].mode = e->versions[1].mode;
1263                e->name = to_atom(c, strlen(c));
1264                c += e->name->str_len + 1;
1265                hashcpy(e->versions[0].sha1, (unsigned char*)c);
1266                hashcpy(e->versions[1].sha1, (unsigned char*)c);
1267                c += 20;
1268        }
1269        free(buf);
1270}
1271
1272static int tecmp0 (const void *_a, const void *_b)
1273{
1274        struct tree_entry *a = *((struct tree_entry**)_a);
1275        struct tree_entry *b = *((struct tree_entry**)_b);
1276        return base_name_compare(
1277                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1278                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1279}
1280
1281static int tecmp1 (const void *_a, const void *_b)
1282{
1283        struct tree_entry *a = *((struct tree_entry**)_a);
1284        struct tree_entry *b = *((struct tree_entry**)_b);
1285        return base_name_compare(
1286                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1287                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1288}
1289
1290static void mktree(struct tree_content *t, int v, struct strbuf *b)
1291{
1292        size_t maxlen = 0;
1293        unsigned int i;
1294
1295        if (!v)
1296                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1297        else
1298                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1299
1300        for (i = 0; i < t->entry_count; i++) {
1301                if (t->entries[i]->versions[v].mode)
1302                        maxlen += t->entries[i]->name->str_len + 34;
1303        }
1304
1305        strbuf_reset(b);
1306        strbuf_grow(b, maxlen);
1307        for (i = 0; i < t->entry_count; i++) {
1308                struct tree_entry *e = t->entries[i];
1309                if (!e->versions[v].mode)
1310                        continue;
1311                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1312                                        e->name->str_dat, '\0');
1313                strbuf_add(b, e->versions[v].sha1, 20);
1314        }
1315}
1316
1317static void store_tree(struct tree_entry *root)
1318{
1319        struct tree_content *t = root->tree;
1320        unsigned int i, j, del;
1321        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1322        struct object_entry *le;
1323
1324        if (!is_null_sha1(root->versions[1].sha1))
1325                return;
1326
1327        for (i = 0; i < t->entry_count; i++) {
1328                if (t->entries[i]->tree)
1329                        store_tree(t->entries[i]);
1330        }
1331
1332        le = find_object(root->versions[0].sha1);
1333        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1334                mktree(t, 0, &old_tree);
1335                lo.data = old_tree;
1336                lo.offset = le->offset;
1337                lo.depth = t->delta_depth;
1338        }
1339
1340        mktree(t, 1, &new_tree);
1341        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1342
1343        t->delta_depth = lo.depth;
1344        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1345                struct tree_entry *e = t->entries[i];
1346                if (e->versions[1].mode) {
1347                        e->versions[0].mode = e->versions[1].mode;
1348                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1349                        t->entries[j++] = e;
1350                } else {
1351                        release_tree_entry(e);
1352                        del++;
1353                }
1354        }
1355        t->entry_count -= del;
1356}
1357
1358static int tree_content_set(
1359        struct tree_entry *root,
1360        const char *p,
1361        const unsigned char *sha1,
1362        const uint16_t mode,
1363        struct tree_content *subtree)
1364{
1365        struct tree_content *t = root->tree;
1366        const char *slash1;
1367        unsigned int i, n;
1368        struct tree_entry *e;
1369
1370        slash1 = strchr(p, '/');
1371        if (slash1)
1372                n = slash1 - p;
1373        else
1374                n = strlen(p);
1375        if (!n)
1376                die("Empty path component found in input");
1377        if (!slash1 && !S_ISDIR(mode) && subtree)
1378                die("Non-directories cannot have subtrees");
1379
1380        for (i = 0; i < t->entry_count; i++) {
1381                e = t->entries[i];
1382                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1383                        if (!slash1) {
1384                                if (!S_ISDIR(mode)
1385                                                && e->versions[1].mode == mode
1386                                                && !hashcmp(e->versions[1].sha1, sha1))
1387                                        return 0;
1388                                e->versions[1].mode = mode;
1389                                hashcpy(e->versions[1].sha1, sha1);
1390                                if (e->tree)
1391                                        release_tree_content_recursive(e->tree);
1392                                e->tree = subtree;
1393                                hashclr(root->versions[1].sha1);
1394                                return 1;
1395                        }
1396                        if (!S_ISDIR(e->versions[1].mode)) {
1397                                e->tree = new_tree_content(8);
1398                                e->versions[1].mode = S_IFDIR;
1399                        }
1400                        if (!e->tree)
1401                                load_tree(e);
1402                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1403                                hashclr(root->versions[1].sha1);
1404                                return 1;
1405                        }
1406                        return 0;
1407                }
1408        }
1409
1410        if (t->entry_count == t->entry_capacity)
1411                root->tree = t = grow_tree_content(t, t->entry_count);
1412        e = new_tree_entry();
1413        e->name = to_atom(p, n);
1414        e->versions[0].mode = 0;
1415        hashclr(e->versions[0].sha1);
1416        t->entries[t->entry_count++] = e;
1417        if (slash1) {
1418                e->tree = new_tree_content(8);
1419                e->versions[1].mode = S_IFDIR;
1420                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1421        } else {
1422                e->tree = subtree;
1423                e->versions[1].mode = mode;
1424                hashcpy(e->versions[1].sha1, sha1);
1425        }
1426        hashclr(root->versions[1].sha1);
1427        return 1;
1428}
1429
1430static int tree_content_remove(
1431        struct tree_entry *root,
1432        const char *p,
1433        struct tree_entry *backup_leaf)
1434{
1435        struct tree_content *t = root->tree;
1436        const char *slash1;
1437        unsigned int i, n;
1438        struct tree_entry *e;
1439
1440        slash1 = strchr(p, '/');
1441        if (slash1)
1442                n = slash1 - p;
1443        else
1444                n = strlen(p);
1445
1446        for (i = 0; i < t->entry_count; i++) {
1447                e = t->entries[i];
1448                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1449                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1450                                goto del_entry;
1451                        if (!e->tree)
1452                                load_tree(e);
1453                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1454                                for (n = 0; n < e->tree->entry_count; n++) {
1455                                        if (e->tree->entries[n]->versions[1].mode) {
1456                                                hashclr(root->versions[1].sha1);
1457                                                return 1;
1458                                        }
1459                                }
1460                                backup_leaf = NULL;
1461                                goto del_entry;
1462                        }
1463                        return 0;
1464                }
1465        }
1466        return 0;
1467
1468del_entry:
1469        if (backup_leaf)
1470                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1471        else if (e->tree)
1472                release_tree_content_recursive(e->tree);
1473        e->tree = NULL;
1474        e->versions[1].mode = 0;
1475        hashclr(e->versions[1].sha1);
1476        hashclr(root->versions[1].sha1);
1477        return 1;
1478}
1479
1480static int tree_content_get(
1481        struct tree_entry *root,
1482        const char *p,
1483        struct tree_entry *leaf)
1484{
1485        struct tree_content *t = root->tree;
1486        const char *slash1;
1487        unsigned int i, n;
1488        struct tree_entry *e;
1489
1490        slash1 = strchr(p, '/');
1491        if (slash1)
1492                n = slash1 - p;
1493        else
1494                n = strlen(p);
1495
1496        for (i = 0; i < t->entry_count; i++) {
1497                e = t->entries[i];
1498                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1499                        if (!slash1) {
1500                                memcpy(leaf, e, sizeof(*leaf));
1501                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1502                                        leaf->tree = dup_tree_content(e->tree);
1503                                else
1504                                        leaf->tree = NULL;
1505                                return 1;
1506                        }
1507                        if (!S_ISDIR(e->versions[1].mode))
1508                                return 0;
1509                        if (!e->tree)
1510                                load_tree(e);
1511                        return tree_content_get(e, slash1 + 1, leaf);
1512                }
1513        }
1514        return 0;
1515}
1516
1517static int update_branch(struct branch *b)
1518{
1519        static const char *msg = "fast-import";
1520        struct ref_lock *lock;
1521        unsigned char old_sha1[20];
1522
1523        if (is_null_sha1(b->sha1))
1524                return 0;
1525        if (read_ref(b->name, old_sha1))
1526                hashclr(old_sha1);
1527        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1528        if (!lock)
1529                return error("Unable to lock %s", b->name);
1530        if (!force_update && !is_null_sha1(old_sha1)) {
1531                struct commit *old_cmit, *new_cmit;
1532
1533                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1534                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1535                if (!old_cmit || !new_cmit) {
1536                        unlock_ref(lock);
1537                        return error("Branch %s is missing commits.", b->name);
1538                }
1539
1540                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1541                        unlock_ref(lock);
1542                        warning("Not updating %s"
1543                                " (new tip %s does not contain %s)",
1544                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1545                        return -1;
1546                }
1547        }
1548        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1549                return error("Unable to update %s", b->name);
1550        return 0;
1551}
1552
1553static void dump_branches(void)
1554{
1555        unsigned int i;
1556        struct branch *b;
1557
1558        for (i = 0; i < branch_table_sz; i++) {
1559                for (b = branch_table[i]; b; b = b->table_next_branch)
1560                        failure |= update_branch(b);
1561        }
1562}
1563
1564static void dump_tags(void)
1565{
1566        static const char *msg = "fast-import";
1567        struct tag *t;
1568        struct ref_lock *lock;
1569        char ref_name[PATH_MAX];
1570
1571        for (t = first_tag; t; t = t->next_tag) {
1572                sprintf(ref_name, "tags/%s", t->name);
1573                lock = lock_ref_sha1(ref_name, NULL);
1574                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1575                        failure |= error("Unable to update %s", ref_name);
1576        }
1577}
1578
1579static void dump_marks_helper(FILE *f,
1580        uintmax_t base,
1581        struct mark_set *m)
1582{
1583        uintmax_t k;
1584        if (m->shift) {
1585                for (k = 0; k < 1024; k++) {
1586                        if (m->data.sets[k])
1587                                dump_marks_helper(f, (base + k) << m->shift,
1588                                        m->data.sets[k]);
1589                }
1590        } else {
1591                for (k = 0; k < 1024; k++) {
1592                        if (m->data.marked[k])
1593                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1594                                        sha1_to_hex(m->data.marked[k]->sha1));
1595                }
1596        }
1597}
1598
1599static void dump_marks(void)
1600{
1601        static struct lock_file mark_lock;
1602        int mark_fd;
1603        FILE *f;
1604
1605        if (!mark_file)
1606                return;
1607
1608        mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1609        if (mark_fd < 0) {
1610                failure |= error("Unable to write marks file %s: %s",
1611                        mark_file, strerror(errno));
1612                return;
1613        }
1614
1615        f = fdopen(mark_fd, "w");
1616        if (!f) {
1617                int saved_errno = errno;
1618                rollback_lock_file(&mark_lock);
1619                failure |= error("Unable to write marks file %s: %s",
1620                        mark_file, strerror(saved_errno));
1621                return;
1622        }
1623
1624        /*
1625         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1626         * Assign -1 to the lock file descriptor so that commit_lock_file()
1627         * won't try to close() it.
1628         */
1629        mark_lock.fd = -1;
1630
1631        dump_marks_helper(f, 0, marks);
1632        if (ferror(f) || fclose(f)) {
1633                int saved_errno = errno;
1634                rollback_lock_file(&mark_lock);
1635                failure |= error("Unable to write marks file %s: %s",
1636                        mark_file, strerror(saved_errno));
1637                return;
1638        }
1639
1640        if (commit_lock_file(&mark_lock)) {
1641                int saved_errno = errno;
1642                rollback_lock_file(&mark_lock);
1643                failure |= error("Unable to commit marks file %s: %s",
1644                        mark_file, strerror(saved_errno));
1645                return;
1646        }
1647}
1648
1649static int read_next_command(void)
1650{
1651        static int stdin_eof = 0;
1652
1653        if (stdin_eof) {
1654                unread_command_buf = 0;
1655                return EOF;
1656        }
1657
1658        do {
1659                if (unread_command_buf) {
1660                        unread_command_buf = 0;
1661                } else {
1662                        struct recent_command *rc;
1663
1664                        strbuf_detach(&command_buf, NULL);
1665                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1666                        if (stdin_eof)
1667                                return EOF;
1668
1669                        rc = rc_free;
1670                        if (rc)
1671                                rc_free = rc->next;
1672                        else {
1673                                rc = cmd_hist.next;
1674                                cmd_hist.next = rc->next;
1675                                cmd_hist.next->prev = &cmd_hist;
1676                                free(rc->buf);
1677                        }
1678
1679                        rc->buf = command_buf.buf;
1680                        rc->prev = cmd_tail;
1681                        rc->next = cmd_hist.prev;
1682                        rc->prev->next = rc;
1683                        cmd_tail = rc;
1684                }
1685        } while (command_buf.buf[0] == '#');
1686
1687        return 0;
1688}
1689
1690static void skip_optional_lf(void)
1691{
1692        int term_char = fgetc(stdin);
1693        if (term_char != '\n' && term_char != EOF)
1694                ungetc(term_char, stdin);
1695}
1696
1697static void parse_mark(void)
1698{
1699        if (!prefixcmp(command_buf.buf, "mark :")) {
1700                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1701                read_next_command();
1702        }
1703        else
1704                next_mark = 0;
1705}
1706
1707static void parse_data(struct strbuf *sb)
1708{
1709        strbuf_reset(sb);
1710
1711        if (prefixcmp(command_buf.buf, "data "))
1712                die("Expected 'data n' command, found: %s", command_buf.buf);
1713
1714        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1715                char *term = xstrdup(command_buf.buf + 5 + 2);
1716                size_t term_len = command_buf.len - 5 - 2;
1717
1718                strbuf_detach(&command_buf, NULL);
1719                for (;;) {
1720                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1721                                die("EOF in data (terminator '%s' not found)", term);
1722                        if (term_len == command_buf.len
1723                                && !strcmp(term, command_buf.buf))
1724                                break;
1725                        strbuf_addbuf(sb, &command_buf);
1726                        strbuf_addch(sb, '\n');
1727                }
1728                free(term);
1729        }
1730        else {
1731                size_t n = 0, length;
1732
1733                length = strtoul(command_buf.buf + 5, NULL, 10);
1734
1735                while (n < length) {
1736                        size_t s = strbuf_fread(sb, length - n, stdin);
1737                        if (!s && feof(stdin))
1738                                die("EOF in data (%lu bytes remaining)",
1739                                        (unsigned long)(length - n));
1740                        n += s;
1741                }
1742        }
1743
1744        skip_optional_lf();
1745}
1746
1747static int validate_raw_date(const char *src, char *result, int maxlen)
1748{
1749        const char *orig_src = src;
1750        char *endp, sign;
1751        unsigned long date;
1752
1753        errno = 0;
1754
1755        date = strtoul(src, &endp, 10);
1756        if (errno || endp == src || *endp != ' ')
1757                return -1;
1758
1759        src = endp + 1;
1760        if (*src != '-' && *src != '+')
1761                return -1;
1762        sign = *src;
1763
1764        date = strtoul(src + 1, &endp, 10);
1765        if (errno || endp == src || *endp || (endp - orig_src) >= maxlen)
1766                return -1;
1767
1768        strcpy(result, orig_src);
1769        return 0;
1770}
1771
1772static char *parse_ident(const char *buf)
1773{
1774        const char *gt;
1775        size_t name_len;
1776        char *ident;
1777
1778        gt = strrchr(buf, '>');
1779        if (!gt)
1780                die("Missing > in ident string: %s", buf);
1781        gt++;
1782        if (*gt != ' ')
1783                die("Missing space after > in ident string: %s", buf);
1784        gt++;
1785        name_len = gt - buf;
1786        ident = xmalloc(name_len + 24);
1787        strncpy(ident, buf, name_len);
1788
1789        switch (whenspec) {
1790        case WHENSPEC_RAW:
1791                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1792                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1793                break;
1794        case WHENSPEC_RFC2822:
1795                if (parse_date(gt, ident + name_len, 24) < 0)
1796                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1797                break;
1798        case WHENSPEC_NOW:
1799                if (strcmp("now", gt))
1800                        die("Date in ident must be 'now': %s", buf);
1801                datestamp(ident + name_len, 24);
1802                break;
1803        }
1804
1805        return ident;
1806}
1807
1808static void parse_new_blob(void)
1809{
1810        static struct strbuf buf = STRBUF_INIT;
1811
1812        read_next_command();
1813        parse_mark();
1814        parse_data(&buf);
1815        store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1816}
1817
1818static void unload_one_branch(void)
1819{
1820        while (cur_active_branches
1821                && cur_active_branches >= max_active_branches) {
1822                uintmax_t min_commit = ULONG_MAX;
1823                struct branch *e, *l = NULL, *p = NULL;
1824
1825                for (e = active_branches; e; e = e->active_next_branch) {
1826                        if (e->last_commit < min_commit) {
1827                                p = l;
1828                                min_commit = e->last_commit;
1829                        }
1830                        l = e;
1831                }
1832
1833                if (p) {
1834                        e = p->active_next_branch;
1835                        p->active_next_branch = e->active_next_branch;
1836                } else {
1837                        e = active_branches;
1838                        active_branches = e->active_next_branch;
1839                }
1840                e->active = 0;
1841                e->active_next_branch = NULL;
1842                if (e->branch_tree.tree) {
1843                        release_tree_content_recursive(e->branch_tree.tree);
1844                        e->branch_tree.tree = NULL;
1845                }
1846                cur_active_branches--;
1847        }
1848}
1849
1850static void load_branch(struct branch *b)
1851{
1852        load_tree(&b->branch_tree);
1853        if (!b->active) {
1854                b->active = 1;
1855                b->active_next_branch = active_branches;
1856                active_branches = b;
1857                cur_active_branches++;
1858                branch_load_count++;
1859        }
1860}
1861
1862static void file_change_m(struct branch *b)
1863{
1864        const char *p = command_buf.buf + 2;
1865        static struct strbuf uq = STRBUF_INIT;
1866        const char *endp;
1867        struct object_entry *oe = oe;
1868        unsigned char sha1[20];
1869        uint16_t mode, inline_data = 0;
1870
1871        p = get_mode(p, &mode);
1872        if (!p)
1873                die("Corrupt mode: %s", command_buf.buf);
1874        switch (mode) {
1875        case 0644:
1876        case 0755:
1877                mode |= S_IFREG;
1878        case S_IFREG | 0644:
1879        case S_IFREG | 0755:
1880        case S_IFLNK:
1881        case S_IFGITLINK:
1882                /* ok */
1883                break;
1884        default:
1885                die("Corrupt mode: %s", command_buf.buf);
1886        }
1887
1888        if (*p == ':') {
1889                char *x;
1890                oe = find_mark(strtoumax(p + 1, &x, 10));
1891                hashcpy(sha1, oe->sha1);
1892                p = x;
1893        } else if (!prefixcmp(p, "inline")) {
1894                inline_data = 1;
1895                p += 6;
1896        } else {
1897                if (get_sha1_hex(p, sha1))
1898                        die("Invalid SHA1: %s", command_buf.buf);
1899                oe = find_object(sha1);
1900                p += 40;
1901        }
1902        if (*p++ != ' ')
1903                die("Missing space after SHA1: %s", command_buf.buf);
1904
1905        strbuf_reset(&uq);
1906        if (!unquote_c_style(&uq, p, &endp)) {
1907                if (*endp)
1908                        die("Garbage after path in: %s", command_buf.buf);
1909                p = uq.buf;
1910        }
1911
1912        if (S_ISGITLINK(mode)) {
1913                if (inline_data)
1914                        die("Git links cannot be specified 'inline': %s",
1915                                command_buf.buf);
1916                else if (oe) {
1917                        if (oe->type != OBJ_COMMIT)
1918                                die("Not a commit (actually a %s): %s",
1919                                        typename(oe->type), command_buf.buf);
1920                }
1921                /*
1922                 * Accept the sha1 without checking; it expected to be in
1923                 * another repository.
1924                 */
1925        } else if (inline_data) {
1926                static struct strbuf buf = STRBUF_INIT;
1927
1928                if (p != uq.buf) {
1929                        strbuf_addstr(&uq, p);
1930                        p = uq.buf;
1931                }
1932                read_next_command();
1933                parse_data(&buf);
1934                store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1935        } else if (oe) {
1936                if (oe->type != OBJ_BLOB)
1937                        die("Not a blob (actually a %s): %s",
1938                                typename(oe->type), command_buf.buf);
1939        } else {
1940                enum object_type type = sha1_object_info(sha1, NULL);
1941                if (type < 0)
1942                        die("Blob not found: %s", command_buf.buf);
1943                if (type != OBJ_BLOB)
1944                        die("Not a blob (actually a %s): %s",
1945                            typename(type), command_buf.buf);
1946        }
1947
1948        tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
1949}
1950
1951static void file_change_d(struct branch *b)
1952{
1953        const char *p = command_buf.buf + 2;
1954        static struct strbuf uq = STRBUF_INIT;
1955        const char *endp;
1956
1957        strbuf_reset(&uq);
1958        if (!unquote_c_style(&uq, p, &endp)) {
1959                if (*endp)
1960                        die("Garbage after path in: %s", command_buf.buf);
1961                p = uq.buf;
1962        }
1963        tree_content_remove(&b->branch_tree, p, NULL);
1964}
1965
1966static void file_change_cr(struct branch *b, int rename)
1967{
1968        const char *s, *d;
1969        static struct strbuf s_uq = STRBUF_INIT;
1970        static struct strbuf d_uq = STRBUF_INIT;
1971        const char *endp;
1972        struct tree_entry leaf;
1973
1974        s = command_buf.buf + 2;
1975        strbuf_reset(&s_uq);
1976        if (!unquote_c_style(&s_uq, s, &endp)) {
1977                if (*endp != ' ')
1978                        die("Missing space after source: %s", command_buf.buf);
1979        } else {
1980                endp = strchr(s, ' ');
1981                if (!endp)
1982                        die("Missing space after source: %s", command_buf.buf);
1983                strbuf_add(&s_uq, s, endp - s);
1984        }
1985        s = s_uq.buf;
1986
1987        endp++;
1988        if (!*endp)
1989                die("Missing dest: %s", command_buf.buf);
1990
1991        d = endp;
1992        strbuf_reset(&d_uq);
1993        if (!unquote_c_style(&d_uq, d, &endp)) {
1994                if (*endp)
1995                        die("Garbage after dest in: %s", command_buf.buf);
1996                d = d_uq.buf;
1997        }
1998
1999        memset(&leaf, 0, sizeof(leaf));
2000        if (rename)
2001                tree_content_remove(&b->branch_tree, s, &leaf);
2002        else
2003                tree_content_get(&b->branch_tree, s, &leaf);
2004        if (!leaf.versions[1].mode)
2005                die("Path %s not in branch", s);
2006        tree_content_set(&b->branch_tree, d,
2007                leaf.versions[1].sha1,
2008                leaf.versions[1].mode,
2009                leaf.tree);
2010}
2011
2012static void file_change_deleteall(struct branch *b)
2013{
2014        release_tree_content_recursive(b->branch_tree.tree);
2015        hashclr(b->branch_tree.versions[0].sha1);
2016        hashclr(b->branch_tree.versions[1].sha1);
2017        load_tree(&b->branch_tree);
2018}
2019
2020static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2021{
2022        if (!buf || size < 46)
2023                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2024        if (memcmp("tree ", buf, 5)
2025                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2026                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2027        hashcpy(b->branch_tree.versions[0].sha1,
2028                b->branch_tree.versions[1].sha1);
2029}
2030
2031static void parse_from_existing(struct branch *b)
2032{
2033        if (is_null_sha1(b->sha1)) {
2034                hashclr(b->branch_tree.versions[0].sha1);
2035                hashclr(b->branch_tree.versions[1].sha1);
2036        } else {
2037                unsigned long size;
2038                char *buf;
2039
2040                buf = read_object_with_reference(b->sha1,
2041                        commit_type, &size, b->sha1);
2042                parse_from_commit(b, buf, size);
2043                free(buf);
2044        }
2045}
2046
2047static int parse_from(struct branch *b)
2048{
2049        const char *from;
2050        struct branch *s;
2051
2052        if (prefixcmp(command_buf.buf, "from "))
2053                return 0;
2054
2055        if (b->branch_tree.tree) {
2056                release_tree_content_recursive(b->branch_tree.tree);
2057                b->branch_tree.tree = NULL;
2058        }
2059
2060        from = strchr(command_buf.buf, ' ') + 1;
2061        s = lookup_branch(from);
2062        if (b == s)
2063                die("Can't create a branch from itself: %s", b->name);
2064        else if (s) {
2065                unsigned char *t = s->branch_tree.versions[1].sha1;
2066                hashcpy(b->sha1, s->sha1);
2067                hashcpy(b->branch_tree.versions[0].sha1, t);
2068                hashcpy(b->branch_tree.versions[1].sha1, t);
2069        } else if (*from == ':') {
2070                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2071                struct object_entry *oe = find_mark(idnum);
2072                if (oe->type != OBJ_COMMIT)
2073                        die("Mark :%" PRIuMAX " not a commit", idnum);
2074                hashcpy(b->sha1, oe->sha1);
2075                if (oe->pack_id != MAX_PACK_ID) {
2076                        unsigned long size;
2077                        char *buf = gfi_unpack_entry(oe, &size);
2078                        parse_from_commit(b, buf, size);
2079                        free(buf);
2080                } else
2081                        parse_from_existing(b);
2082        } else if (!get_sha1(from, b->sha1))
2083                parse_from_existing(b);
2084        else
2085                die("Invalid ref name or SHA1 expression: %s", from);
2086
2087        read_next_command();
2088        return 1;
2089}
2090
2091static struct hash_list *parse_merge(unsigned int *count)
2092{
2093        struct hash_list *list = NULL, *n, *e = e;
2094        const char *from;
2095        struct branch *s;
2096
2097        *count = 0;
2098        while (!prefixcmp(command_buf.buf, "merge ")) {
2099                from = strchr(command_buf.buf, ' ') + 1;
2100                n = xmalloc(sizeof(*n));
2101                s = lookup_branch(from);
2102                if (s)
2103                        hashcpy(n->sha1, s->sha1);
2104                else if (*from == ':') {
2105                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2106                        struct object_entry *oe = find_mark(idnum);
2107                        if (oe->type != OBJ_COMMIT)
2108                                die("Mark :%" PRIuMAX " not a commit", idnum);
2109                        hashcpy(n->sha1, oe->sha1);
2110                } else if (!get_sha1(from, n->sha1)) {
2111                        unsigned long size;
2112                        char *buf = read_object_with_reference(n->sha1,
2113                                commit_type, &size, n->sha1);
2114                        if (!buf || size < 46)
2115                                die("Not a valid commit: %s", from);
2116                        free(buf);
2117                } else
2118                        die("Invalid ref name or SHA1 expression: %s", from);
2119
2120                n->next = NULL;
2121                if (list)
2122                        e->next = n;
2123                else
2124                        list = n;
2125                e = n;
2126                (*count)++;
2127                read_next_command();
2128        }
2129        return list;
2130}
2131
2132static void parse_new_commit(void)
2133{
2134        static struct strbuf msg = STRBUF_INIT;
2135        struct branch *b;
2136        char *sp;
2137        char *author = NULL;
2138        char *committer = NULL;
2139        struct hash_list *merge_list = NULL;
2140        unsigned int merge_count;
2141
2142        /* Obtain the branch name from the rest of our command */
2143        sp = strchr(command_buf.buf, ' ') + 1;
2144        b = lookup_branch(sp);
2145        if (!b)
2146                b = new_branch(sp);
2147
2148        read_next_command();
2149        parse_mark();
2150        if (!prefixcmp(command_buf.buf, "author ")) {
2151                author = parse_ident(command_buf.buf + 7);
2152                read_next_command();
2153        }
2154        if (!prefixcmp(command_buf.buf, "committer ")) {
2155                committer = parse_ident(command_buf.buf + 10);
2156                read_next_command();
2157        }
2158        if (!committer)
2159                die("Expected committer but didn't get one");
2160        parse_data(&msg);
2161        read_next_command();
2162        parse_from(b);
2163        merge_list = parse_merge(&merge_count);
2164
2165        /* ensure the branch is active/loaded */
2166        if (!b->branch_tree.tree || !max_active_branches) {
2167                unload_one_branch();
2168                load_branch(b);
2169        }
2170
2171        /* file_change* */
2172        while (command_buf.len > 0) {
2173                if (!prefixcmp(command_buf.buf, "M "))
2174                        file_change_m(b);
2175                else if (!prefixcmp(command_buf.buf, "D "))
2176                        file_change_d(b);
2177                else if (!prefixcmp(command_buf.buf, "R "))
2178                        file_change_cr(b, 1);
2179                else if (!prefixcmp(command_buf.buf, "C "))
2180                        file_change_cr(b, 0);
2181                else if (!strcmp("deleteall", command_buf.buf))
2182                        file_change_deleteall(b);
2183                else {
2184                        unread_command_buf = 1;
2185                        break;
2186                }
2187                if (read_next_command() == EOF)
2188                        break;
2189        }
2190
2191        /* build the tree and the commit */
2192        store_tree(&b->branch_tree);
2193        hashcpy(b->branch_tree.versions[0].sha1,
2194                b->branch_tree.versions[1].sha1);
2195
2196        strbuf_reset(&new_data);
2197        strbuf_addf(&new_data, "tree %s\n",
2198                sha1_to_hex(b->branch_tree.versions[1].sha1));
2199        if (!is_null_sha1(b->sha1))
2200                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2201        while (merge_list) {
2202                struct hash_list *next = merge_list->next;
2203                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2204                free(merge_list);
2205                merge_list = next;
2206        }
2207        strbuf_addf(&new_data,
2208                "author %s\n"
2209                "committer %s\n"
2210                "\n",
2211                author ? author : committer, committer);
2212        strbuf_addbuf(&new_data, &msg);
2213        free(author);
2214        free(committer);
2215
2216        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2217                b->pack_id = pack_id;
2218        b->last_commit = object_count_by_type[OBJ_COMMIT];
2219}
2220
2221static void parse_new_tag(void)
2222{
2223        static struct strbuf msg = STRBUF_INIT;
2224        char *sp;
2225        const char *from;
2226        char *tagger;
2227        struct branch *s;
2228        struct tag *t;
2229        uintmax_t from_mark = 0;
2230        unsigned char sha1[20];
2231
2232        /* Obtain the new tag name from the rest of our command */
2233        sp = strchr(command_buf.buf, ' ') + 1;
2234        t = pool_alloc(sizeof(struct tag));
2235        t->next_tag = NULL;
2236        t->name = pool_strdup(sp);
2237        if (last_tag)
2238                last_tag->next_tag = t;
2239        else
2240                first_tag = t;
2241        last_tag = t;
2242        read_next_command();
2243
2244        /* from ... */
2245        if (prefixcmp(command_buf.buf, "from "))
2246                die("Expected from command, got %s", command_buf.buf);
2247        from = strchr(command_buf.buf, ' ') + 1;
2248        s = lookup_branch(from);
2249        if (s) {
2250                hashcpy(sha1, s->sha1);
2251        } else if (*from == ':') {
2252                struct object_entry *oe;
2253                from_mark = strtoumax(from + 1, NULL, 10);
2254                oe = find_mark(from_mark);
2255                if (oe->type != OBJ_COMMIT)
2256                        die("Mark :%" PRIuMAX " not a commit", from_mark);
2257                hashcpy(sha1, oe->sha1);
2258        } else if (!get_sha1(from, sha1)) {
2259                unsigned long size;
2260                char *buf;
2261
2262                buf = read_object_with_reference(sha1,
2263                        commit_type, &size, sha1);
2264                if (!buf || size < 46)
2265                        die("Not a valid commit: %s", from);
2266                free(buf);
2267        } else
2268                die("Invalid ref name or SHA1 expression: %s", from);
2269        read_next_command();
2270
2271        /* tagger ... */
2272        if (!prefixcmp(command_buf.buf, "tagger ")) {
2273                tagger = parse_ident(command_buf.buf + 7);
2274                read_next_command();
2275        } else
2276                tagger = NULL;
2277
2278        /* tag payload/message */
2279        parse_data(&msg);
2280
2281        /* build the tag object */
2282        strbuf_reset(&new_data);
2283
2284        strbuf_addf(&new_data,
2285                    "object %s\n"
2286                    "type %s\n"
2287                    "tag %s\n",
2288                    sha1_to_hex(sha1), commit_type, t->name);
2289        if (tagger)
2290                strbuf_addf(&new_data,
2291                            "tagger %s\n", tagger);
2292        strbuf_addch(&new_data, '\n');
2293        strbuf_addbuf(&new_data, &msg);
2294        free(tagger);
2295
2296        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2297                t->pack_id = MAX_PACK_ID;
2298        else
2299                t->pack_id = pack_id;
2300}
2301
2302static void parse_reset_branch(void)
2303{
2304        struct branch *b;
2305        char *sp;
2306
2307        /* Obtain the branch name from the rest of our command */
2308        sp = strchr(command_buf.buf, ' ') + 1;
2309        b = lookup_branch(sp);
2310        if (b) {
2311                hashclr(b->sha1);
2312                hashclr(b->branch_tree.versions[0].sha1);
2313                hashclr(b->branch_tree.versions[1].sha1);
2314                if (b->branch_tree.tree) {
2315                        release_tree_content_recursive(b->branch_tree.tree);
2316                        b->branch_tree.tree = NULL;
2317                }
2318        }
2319        else
2320                b = new_branch(sp);
2321        read_next_command();
2322        parse_from(b);
2323        if (command_buf.len > 0)
2324                unread_command_buf = 1;
2325}
2326
2327static void parse_checkpoint(void)
2328{
2329        if (object_count) {
2330                cycle_packfile();
2331                dump_branches();
2332                dump_tags();
2333                dump_marks();
2334        }
2335        skip_optional_lf();
2336}
2337
2338static void parse_progress(void)
2339{
2340        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2341        fputc('\n', stdout);
2342        fflush(stdout);
2343        skip_optional_lf();
2344}
2345
2346static void import_marks(const char *input_file)
2347{
2348        char line[512];
2349        FILE *f = fopen(input_file, "r");
2350        if (!f)
2351                die("cannot read %s: %s", input_file, strerror(errno));
2352        while (fgets(line, sizeof(line), f)) {
2353                uintmax_t mark;
2354                char *end;
2355                unsigned char sha1[20];
2356                struct object_entry *e;
2357
2358                end = strchr(line, '\n');
2359                if (line[0] != ':' || !end)
2360                        die("corrupt mark line: %s", line);
2361                *end = 0;
2362                mark = strtoumax(line + 1, &end, 10);
2363                if (!mark || end == line + 1
2364                        || *end != ' ' || get_sha1(end + 1, sha1))
2365                        die("corrupt mark line: %s", line);
2366                e = find_object(sha1);
2367                if (!e) {
2368                        enum object_type type = sha1_object_info(sha1, NULL);
2369                        if (type < 0)
2370                                die("object not found: %s", sha1_to_hex(sha1));
2371                        e = insert_object(sha1);
2372                        e->type = type;
2373                        e->pack_id = MAX_PACK_ID;
2374                        e->offset = 1; /* just not zero! */
2375                }
2376                insert_mark(mark, e);
2377        }
2378        fclose(f);
2379}
2380
2381static int git_pack_config(const char *k, const char *v, void *cb)
2382{
2383        if (!strcmp(k, "pack.depth")) {
2384                max_depth = git_config_int(k, v);
2385                if (max_depth > MAX_DEPTH)
2386                        max_depth = MAX_DEPTH;
2387                return 0;
2388        }
2389        if (!strcmp(k, "pack.compression")) {
2390                int level = git_config_int(k, v);
2391                if (level == -1)
2392                        level = Z_DEFAULT_COMPRESSION;
2393                else if (level < 0 || level > Z_BEST_COMPRESSION)
2394                        die("bad pack compression level %d", level);
2395                pack_compression_level = level;
2396                pack_compression_seen = 1;
2397                return 0;
2398        }
2399        return git_default_config(k, v, cb);
2400}
2401
2402static const char fast_import_usage[] =
2403"git fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2404
2405int main(int argc, const char **argv)
2406{
2407        unsigned int i, show_stats = 1;
2408
2409        setup_git_directory();
2410        git_config(git_pack_config, NULL);
2411        if (!pack_compression_seen && core_compression_seen)
2412                pack_compression_level = core_compression_level;
2413
2414        alloc_objects(object_entry_alloc);
2415        strbuf_init(&command_buf, 0);
2416        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2417        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2418        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2419        marks = pool_calloc(1, sizeof(struct mark_set));
2420
2421        for (i = 1; i < argc; i++) {
2422                const char *a = argv[i];
2423
2424                if (*a != '-' || !strcmp(a, "--"))
2425                        break;
2426                else if (!prefixcmp(a, "--date-format=")) {
2427                        const char *fmt = a + 14;
2428                        if (!strcmp(fmt, "raw"))
2429                                whenspec = WHENSPEC_RAW;
2430                        else if (!strcmp(fmt, "rfc2822"))
2431                                whenspec = WHENSPEC_RFC2822;
2432                        else if (!strcmp(fmt, "now"))
2433                                whenspec = WHENSPEC_NOW;
2434                        else
2435                                die("unknown --date-format argument %s", fmt);
2436                }
2437                else if (!prefixcmp(a, "--max-pack-size="))
2438                        max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2439                else if (!prefixcmp(a, "--depth=")) {
2440                        max_depth = strtoul(a + 8, NULL, 0);
2441                        if (max_depth > MAX_DEPTH)
2442                                die("--depth cannot exceed %u", MAX_DEPTH);
2443                }
2444                else if (!prefixcmp(a, "--active-branches="))
2445                        max_active_branches = strtoul(a + 18, NULL, 0);
2446                else if (!prefixcmp(a, "--import-marks="))
2447                        import_marks(a + 15);
2448                else if (!prefixcmp(a, "--export-marks="))
2449                        mark_file = a + 15;
2450                else if (!prefixcmp(a, "--export-pack-edges=")) {
2451                        if (pack_edges)
2452                                fclose(pack_edges);
2453                        pack_edges = fopen(a + 20, "a");
2454                        if (!pack_edges)
2455                                die("Cannot open %s: %s", a + 20, strerror(errno));
2456                } else if (!strcmp(a, "--force"))
2457                        force_update = 1;
2458                else if (!strcmp(a, "--quiet"))
2459                        show_stats = 0;
2460                else if (!strcmp(a, "--stats"))
2461                        show_stats = 1;
2462                else
2463                        die("unknown option %s", a);
2464        }
2465        if (i != argc)
2466                usage(fast_import_usage);
2467
2468        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2469        for (i = 0; i < (cmd_save - 1); i++)
2470                rc_free[i].next = &rc_free[i + 1];
2471        rc_free[cmd_save - 1].next = NULL;
2472
2473        prepare_packed_git();
2474        start_packfile();
2475        set_die_routine(die_nicely);
2476        while (read_next_command() != EOF) {
2477                if (!strcmp("blob", command_buf.buf))
2478                        parse_new_blob();
2479                else if (!prefixcmp(command_buf.buf, "commit "))
2480                        parse_new_commit();
2481                else if (!prefixcmp(command_buf.buf, "tag "))
2482                        parse_new_tag();
2483                else if (!prefixcmp(command_buf.buf, "reset "))
2484                        parse_reset_branch();
2485                else if (!strcmp("checkpoint", command_buf.buf))
2486                        parse_checkpoint();
2487                else if (!prefixcmp(command_buf.buf, "progress "))
2488                        parse_progress();
2489                else
2490                        die("Unsupported command: %s", command_buf.buf);
2491        }
2492        end_packfile();
2493
2494        dump_branches();
2495        dump_tags();
2496        unkeep_all_packs();
2497        dump_marks();
2498
2499        if (pack_edges)
2500                fclose(pack_edges);
2501
2502        if (show_stats) {
2503                uintmax_t total_count = 0, duplicate_count = 0;
2504                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2505                        total_count += object_count_by_type[i];
2506                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2507                        duplicate_count += duplicate_count_by_type[i];
2508
2509                fprintf(stderr, "%s statistics:\n", argv[0]);
2510                fprintf(stderr, "---------------------------------------------------------------------\n");
2511                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2512                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2513                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]);
2514                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]);
2515                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]);
2516                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]);
2517                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2518                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2519                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2520                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2521                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2522                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2523                fprintf(stderr, "---------------------------------------------------------------------\n");
2524                pack_report();
2525                fprintf(stderr, "---------------------------------------------------------------------\n");
2526                fprintf(stderr, "\n");
2527        }
2528
2529        return failure ? 1 : 0;
2530}