ab099b63382e67e8094f5e865451d7e24ce96526
   1/*
   2(See Documentation/git-fast-import.txt for maintained documentation.)
   3Format of STDIN stream:
   4
   5  stream ::= cmd*;
   6
   7  cmd ::= new_blob
   8        | new_commit
   9        | new_tag
  10        | reset_branch
  11        | checkpoint
  12        | progress
  13        ;
  14
  15  new_blob ::= 'blob' lf
  16    mark?
  17    file_content;
  18  file_content ::= data;
  19
  20  new_commit ::= 'commit' sp ref_str lf
  21    mark?
  22    ('author' sp name sp '<' email '>' sp when lf)?
  23    'committer' sp name sp '<' email '>' sp when lf
  24    commit_msg
  25    ('from' sp committish lf)?
  26    ('merge' sp committish lf)*
  27    file_change*
  28    lf?;
  29  commit_msg ::= data;
  30
  31  file_change ::= file_clr
  32    | file_del
  33    | file_rnm
  34    | file_cpy
  35    | file_obm
  36    | file_inm;
  37  file_clr ::= 'deleteall' lf;
  38  file_del ::= 'D' sp path_str lf;
  39  file_rnm ::= 'R' sp path_str sp path_str lf;
  40  file_cpy ::= 'C' sp path_str sp path_str lf;
  41  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
  42  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
  43    data;
  44  note_obm ::= 'N' sp (hexsha1 | idnum) sp committish lf;
  45  note_inm ::= 'N' sp 'inline' sp committish lf
  46    data;
  47
  48  new_tag ::= 'tag' sp tag_str lf
  49    'from' sp committish lf
  50    ('tagger' sp name sp '<' email '>' sp when lf)?
  51    tag_msg;
  52  tag_msg ::= data;
  53
  54  reset_branch ::= 'reset' sp ref_str lf
  55    ('from' sp committish lf)?
  56    lf?;
  57
  58  checkpoint ::= 'checkpoint' lf
  59    lf?;
  60
  61  progress ::= 'progress' sp not_lf* lf
  62    lf?;
  63
  64     # note: the first idnum in a stream should be 1 and subsequent
  65     # idnums should not have gaps between values as this will cause
  66     # the stream parser to reserve space for the gapped values.  An
  67     # idnum can be updated in the future to a new object by issuing
  68     # a new mark directive with the old idnum.
  69     #
  70  mark ::= 'mark' sp idnum lf;
  71  data ::= (delimited_data | exact_data)
  72    lf?;
  73
  74    # note: delim may be any string but must not contain lf.
  75    # data_line may contain any data but must not be exactly
  76    # delim.
  77  delimited_data ::= 'data' sp '<<' delim lf
  78    (data_line lf)*
  79    delim lf;
  80
  81     # note: declen indicates the length of binary_data in bytes.
  82     # declen does not include the lf preceding the binary data.
  83     #
  84  exact_data ::= 'data' sp declen lf
  85    binary_data;
  86
  87     # note: quoted strings are C-style quoting supporting \c for
  88     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  89     # is the signed byte value in octal.  Note that the only
  90     # characters which must actually be escaped to protect the
  91     # stream formatting is: \, " and LF.  Otherwise these values
  92     # are UTF8.
  93     #
  94  committish  ::= (ref_str | hexsha1 | sha1exp_str | idnum);
  95  ref_str     ::= ref;
  96  sha1exp_str ::= sha1exp;
  97  tag_str     ::= tag;
  98  path_str    ::= path    | '"' quoted(path)    '"' ;
  99  mode        ::= '100644' | '644'
 100                | '100755' | '755'
 101                | '120000'
 102                ;
 103
 104  declen ::= # unsigned 32 bit value, ascii base10 notation;
 105  bigint ::= # unsigned integer value, ascii base10 notation;
 106  binary_data ::= # file content, not interpreted;
 107
 108  when         ::= raw_when | rfc2822_when;
 109  raw_when     ::= ts sp tz;
 110  rfc2822_when ::= # Valid RFC 2822 date and time;
 111
 112  sp ::= # ASCII space character;
 113  lf ::= # ASCII newline (LF) character;
 114
 115     # note: a colon (':') must precede the numerical value assigned to
 116     # an idnum.  This is to distinguish it from a ref or tag name as
 117     # GIT does not permit ':' in ref or tag strings.
 118     #
 119  idnum   ::= ':' bigint;
 120  path    ::= # GIT style file path, e.g. "a/b/c";
 121  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
 122  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
 123  sha1exp ::= # Any valid GIT SHA1 expression;
 124  hexsha1 ::= # SHA1 in hexadecimal format;
 125
 126     # note: name and email are UTF8 strings, however name must not
 127     # contain '<' or lf and email must not contain any of the
 128     # following: '<', '>', lf.
 129     #
 130  name  ::= # valid GIT author/committer name;
 131  email ::= # valid GIT author/committer email;
 132  ts    ::= # time since the epoch in seconds, ascii base10 notation;
 133  tz    ::= # GIT style timezone;
 134
 135     # note: comments may appear anywhere in the input, except
 136     # within a data command.  Any form of the data command
 137     # always escapes the related input from comment processing.
 138     #
 139     # In case it is not clear, the '#' that starts the comment
 140     # must be the first character on that line (an lf
 141     # preceded it).
 142     #
 143  comment ::= '#' not_lf* lf;
 144  not_lf  ::= # Any byte that is not ASCII newline (LF);
 145*/
 146
 147#include "builtin.h"
 148#include "cache.h"
 149#include "object.h"
 150#include "blob.h"
 151#include "tree.h"
 152#include "commit.h"
 153#include "delta.h"
 154#include "pack.h"
 155#include "refs.h"
 156#include "csum-file.h"
 157#include "quote.h"
 158#include "exec_cmd.h"
 159
 160#define PACK_ID_BITS 16
 161#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 162#define DEPTH_BITS 13
 163#define MAX_DEPTH ((1<<DEPTH_BITS)-1)
 164
 165struct object_entry
 166{
 167        struct object_entry *next;
 168        uint32_t offset;
 169        uint32_t type : TYPE_BITS,
 170                pack_id : PACK_ID_BITS,
 171                depth : DEPTH_BITS;
 172        unsigned char sha1[20];
 173};
 174
 175struct object_entry_pool
 176{
 177        struct object_entry_pool *next_pool;
 178        struct object_entry *next_free;
 179        struct object_entry *end;
 180        struct object_entry entries[FLEX_ARRAY]; /* more */
 181};
 182
 183struct mark_set
 184{
 185        union {
 186                struct object_entry *marked[1024];
 187                struct mark_set *sets[1024];
 188        } data;
 189        unsigned int shift;
 190};
 191
 192struct last_object
 193{
 194        struct strbuf data;
 195        uint32_t offset;
 196        unsigned int depth;
 197        unsigned no_swap : 1;
 198};
 199
 200struct mem_pool
 201{
 202        struct mem_pool *next_pool;
 203        char *next_free;
 204        char *end;
 205        uintmax_t space[FLEX_ARRAY]; /* more */
 206};
 207
 208struct atom_str
 209{
 210        struct atom_str *next_atom;
 211        unsigned short str_len;
 212        char str_dat[FLEX_ARRAY]; /* more */
 213};
 214
 215struct tree_content;
 216struct tree_entry
 217{
 218        struct tree_content *tree;
 219        struct atom_str *name;
 220        struct tree_entry_ms
 221        {
 222                uint16_t mode;
 223                unsigned char sha1[20];
 224        } versions[2];
 225};
 226
 227struct tree_content
 228{
 229        unsigned int entry_capacity; /* must match avail_tree_content */
 230        unsigned int entry_count;
 231        unsigned int delta_depth;
 232        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 233};
 234
 235struct avail_tree_content
 236{
 237        unsigned int entry_capacity; /* must match tree_content */
 238        struct avail_tree_content *next_avail;
 239};
 240
 241struct branch
 242{
 243        struct branch *table_next_branch;
 244        struct branch *active_next_branch;
 245        const char *name;
 246        struct tree_entry branch_tree;
 247        uintmax_t last_commit;
 248        unsigned active : 1;
 249        unsigned pack_id : PACK_ID_BITS;
 250        unsigned char sha1[20];
 251};
 252
 253struct tag
 254{
 255        struct tag *next_tag;
 256        const char *name;
 257        unsigned int pack_id;
 258        unsigned char sha1[20];
 259};
 260
 261struct hash_list
 262{
 263        struct hash_list *next;
 264        unsigned char sha1[20];
 265};
 266
 267typedef enum {
 268        WHENSPEC_RAW = 1,
 269        WHENSPEC_RFC2822,
 270        WHENSPEC_NOW,
 271} whenspec_type;
 272
 273struct recent_command
 274{
 275        struct recent_command *prev;
 276        struct recent_command *next;
 277        char *buf;
 278};
 279
 280/* Configured limits on output */
 281static unsigned long max_depth = 10;
 282static off_t max_packsize = (1LL << 32) - 1;
 283static int force_update;
 284static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 285static int pack_compression_seen;
 286
 287/* Stats and misc. counters */
 288static uintmax_t alloc_count;
 289static uintmax_t marks_set_count;
 290static uintmax_t object_count_by_type[1 << TYPE_BITS];
 291static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
 292static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 293static unsigned long object_count;
 294static unsigned long branch_count;
 295static unsigned long branch_load_count;
 296static int failure;
 297static FILE *pack_edges;
 298static unsigned int show_stats = 1;
 299static int global_argc;
 300static const char **global_argv;
 301
 302/* Memory pools */
 303static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 304static size_t total_allocd;
 305static struct mem_pool *mem_pool;
 306
 307/* Atom management */
 308static unsigned int atom_table_sz = 4451;
 309static unsigned int atom_cnt;
 310static struct atom_str **atom_table;
 311
 312/* The .pack file being generated */
 313static unsigned int pack_id;
 314static struct packed_git *pack_data;
 315static struct packed_git **all_packs;
 316static unsigned long pack_size;
 317
 318/* Table of objects we've written. */
 319static unsigned int object_entry_alloc = 5000;
 320static struct object_entry_pool *blocks;
 321static struct object_entry *object_table[1 << 16];
 322static struct mark_set *marks;
 323static const char *export_marks_file;
 324static const char *import_marks_file;
 325
 326/* Our last blob */
 327static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
 328
 329/* Tree management */
 330static unsigned int tree_entry_alloc = 1000;
 331static void *avail_tree_entry;
 332static unsigned int avail_tree_table_sz = 100;
 333static struct avail_tree_content **avail_tree_table;
 334static struct strbuf old_tree = STRBUF_INIT;
 335static struct strbuf new_tree = STRBUF_INIT;
 336
 337/* Branch data */
 338static unsigned long max_active_branches = 5;
 339static unsigned long cur_active_branches;
 340static unsigned long branch_table_sz = 1039;
 341static struct branch **branch_table;
 342static struct branch *active_branches;
 343
 344/* Tag data */
 345static struct tag *first_tag;
 346static struct tag *last_tag;
 347
 348/* Input stream parsing */
 349static whenspec_type whenspec = WHENSPEC_RAW;
 350static struct strbuf command_buf = STRBUF_INIT;
 351static int unread_command_buf;
 352static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
 353static struct recent_command *cmd_tail = &cmd_hist;
 354static struct recent_command *rc_free;
 355static unsigned int cmd_save = 100;
 356static uintmax_t next_mark;
 357static struct strbuf new_data = STRBUF_INIT;
 358static int seen_data_command;
 359
 360static void parse_argv(void);
 361
 362static void write_branch_report(FILE *rpt, struct branch *b)
 363{
 364        fprintf(rpt, "%s:\n", b->name);
 365
 366        fprintf(rpt, "  status      :");
 367        if (b->active)
 368                fputs(" active", rpt);
 369        if (b->branch_tree.tree)
 370                fputs(" loaded", rpt);
 371        if (is_null_sha1(b->branch_tree.versions[1].sha1))
 372                fputs(" dirty", rpt);
 373        fputc('\n', rpt);
 374
 375        fprintf(rpt, "  tip commit  : %s\n", sha1_to_hex(b->sha1));
 376        fprintf(rpt, "  old tree    : %s\n", sha1_to_hex(b->branch_tree.versions[0].sha1));
 377        fprintf(rpt, "  cur tree    : %s\n", sha1_to_hex(b->branch_tree.versions[1].sha1));
 378        fprintf(rpt, "  commit clock: %" PRIuMAX "\n", b->last_commit);
 379
 380        fputs("  last pack   : ", rpt);
 381        if (b->pack_id < MAX_PACK_ID)
 382                fprintf(rpt, "%u", b->pack_id);
 383        fputc('\n', rpt);
 384
 385        fputc('\n', rpt);
 386}
 387
 388static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
 389
 390static void write_crash_report(const char *err)
 391{
 392        char *loc = git_path("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
 393        FILE *rpt = fopen(loc, "w");
 394        struct branch *b;
 395        unsigned long lu;
 396        struct recent_command *rc;
 397
 398        if (!rpt) {
 399                error("can't write crash report %s: %s", loc, strerror(errno));
 400                return;
 401        }
 402
 403        fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
 404
 405        fprintf(rpt, "fast-import crash report:\n");
 406        fprintf(rpt, "    fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
 407        fprintf(rpt, "    parent process     : %"PRIuMAX"\n", (uintmax_t) getppid());
 408        fprintf(rpt, "    at %s\n", show_date(time(NULL), 0, DATE_LOCAL));
 409        fputc('\n', rpt);
 410
 411        fputs("fatal: ", rpt);
 412        fputs(err, rpt);
 413        fputc('\n', rpt);
 414
 415        fputc('\n', rpt);
 416        fputs("Most Recent Commands Before Crash\n", rpt);
 417        fputs("---------------------------------\n", rpt);
 418        for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
 419                if (rc->next == &cmd_hist)
 420                        fputs("* ", rpt);
 421                else
 422                        fputs("  ", rpt);
 423                fputs(rc->buf, rpt);
 424                fputc('\n', rpt);
 425        }
 426
 427        fputc('\n', rpt);
 428        fputs("Active Branch LRU\n", rpt);
 429        fputs("-----------------\n", rpt);
 430        fprintf(rpt, "    active_branches = %lu cur, %lu max\n",
 431                cur_active_branches,
 432                max_active_branches);
 433        fputc('\n', rpt);
 434        fputs("  pos  clock name\n", rpt);
 435        fputs("  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
 436        for (b = active_branches, lu = 0; b; b = b->active_next_branch)
 437                fprintf(rpt, "  %2lu) %6" PRIuMAX" %s\n",
 438                        ++lu, b->last_commit, b->name);
 439
 440        fputc('\n', rpt);
 441        fputs("Inactive Branches\n", rpt);
 442        fputs("-----------------\n", rpt);
 443        for (lu = 0; lu < branch_table_sz; lu++) {
 444                for (b = branch_table[lu]; b; b = b->table_next_branch)
 445                        write_branch_report(rpt, b);
 446        }
 447
 448        if (first_tag) {
 449                struct tag *tg;
 450                fputc('\n', rpt);
 451                fputs("Annotated Tags\n", rpt);
 452                fputs("--------------\n", rpt);
 453                for (tg = first_tag; tg; tg = tg->next_tag) {
 454                        fputs(sha1_to_hex(tg->sha1), rpt);
 455                        fputc(' ', rpt);
 456                        fputs(tg->name, rpt);
 457                        fputc('\n', rpt);
 458                }
 459        }
 460
 461        fputc('\n', rpt);
 462        fputs("Marks\n", rpt);
 463        fputs("-----\n", rpt);
 464        if (export_marks_file)
 465                fprintf(rpt, "  exported to %s\n", export_marks_file);
 466        else
 467                dump_marks_helper(rpt, 0, marks);
 468
 469        fputc('\n', rpt);
 470        fputs("-------------------\n", rpt);
 471        fputs("END OF CRASH REPORT\n", rpt);
 472        fclose(rpt);
 473}
 474
 475static void end_packfile(void);
 476static void unkeep_all_packs(void);
 477static void dump_marks(void);
 478
 479static NORETURN void die_nicely(const char *err, va_list params)
 480{
 481        static int zombie;
 482        char message[2 * PATH_MAX];
 483
 484        vsnprintf(message, sizeof(message), err, params);
 485        fputs("fatal: ", stderr);
 486        fputs(message, stderr);
 487        fputc('\n', stderr);
 488
 489        if (!zombie) {
 490                zombie = 1;
 491                write_crash_report(message);
 492                end_packfile();
 493                unkeep_all_packs();
 494                dump_marks();
 495        }
 496        exit(128);
 497}
 498
 499static void alloc_objects(unsigned int cnt)
 500{
 501        struct object_entry_pool *b;
 502
 503        b = xmalloc(sizeof(struct object_entry_pool)
 504                + cnt * sizeof(struct object_entry));
 505        b->next_pool = blocks;
 506        b->next_free = b->entries;
 507        b->end = b->entries + cnt;
 508        blocks = b;
 509        alloc_count += cnt;
 510}
 511
 512static struct object_entry *new_object(unsigned char *sha1)
 513{
 514        struct object_entry *e;
 515
 516        if (blocks->next_free == blocks->end)
 517                alloc_objects(object_entry_alloc);
 518
 519        e = blocks->next_free++;
 520        hashcpy(e->sha1, sha1);
 521        return e;
 522}
 523
 524static struct object_entry *find_object(unsigned char *sha1)
 525{
 526        unsigned int h = sha1[0] << 8 | sha1[1];
 527        struct object_entry *e;
 528        for (e = object_table[h]; e; e = e->next)
 529                if (!hashcmp(sha1, e->sha1))
 530                        return e;
 531        return NULL;
 532}
 533
 534static struct object_entry *insert_object(unsigned char *sha1)
 535{
 536        unsigned int h = sha1[0] << 8 | sha1[1];
 537        struct object_entry *e = object_table[h];
 538        struct object_entry *p = NULL;
 539
 540        while (e) {
 541                if (!hashcmp(sha1, e->sha1))
 542                        return e;
 543                p = e;
 544                e = e->next;
 545        }
 546
 547        e = new_object(sha1);
 548        e->next = NULL;
 549        e->offset = 0;
 550        if (p)
 551                p->next = e;
 552        else
 553                object_table[h] = e;
 554        return e;
 555}
 556
 557static unsigned int hc_str(const char *s, size_t len)
 558{
 559        unsigned int r = 0;
 560        while (len-- > 0)
 561                r = r * 31 + *s++;
 562        return r;
 563}
 564
 565static void *pool_alloc(size_t len)
 566{
 567        struct mem_pool *p;
 568        void *r;
 569
 570        /* round up to a 'uintmax_t' alignment */
 571        if (len & (sizeof(uintmax_t) - 1))
 572                len += sizeof(uintmax_t) - (len & (sizeof(uintmax_t) - 1));
 573
 574        for (p = mem_pool; p; p = p->next_pool)
 575                if ((p->end - p->next_free >= len))
 576                        break;
 577
 578        if (!p) {
 579                if (len >= (mem_pool_alloc/2)) {
 580                        total_allocd += len;
 581                        return xmalloc(len);
 582                }
 583                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 584                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 585                p->next_pool = mem_pool;
 586                p->next_free = (char *) p->space;
 587                p->end = p->next_free + mem_pool_alloc;
 588                mem_pool = p;
 589        }
 590
 591        r = p->next_free;
 592        p->next_free += len;
 593        return r;
 594}
 595
 596static void *pool_calloc(size_t count, size_t size)
 597{
 598        size_t len = count * size;
 599        void *r = pool_alloc(len);
 600        memset(r, 0, len);
 601        return r;
 602}
 603
 604static char *pool_strdup(const char *s)
 605{
 606        char *r = pool_alloc(strlen(s) + 1);
 607        strcpy(r, s);
 608        return r;
 609}
 610
 611static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 612{
 613        struct mark_set *s = marks;
 614        while ((idnum >> s->shift) >= 1024) {
 615                s = pool_calloc(1, sizeof(struct mark_set));
 616                s->shift = marks->shift + 10;
 617                s->data.sets[0] = marks;
 618                marks = s;
 619        }
 620        while (s->shift) {
 621                uintmax_t i = idnum >> s->shift;
 622                idnum -= i << s->shift;
 623                if (!s->data.sets[i]) {
 624                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 625                        s->data.sets[i]->shift = s->shift - 10;
 626                }
 627                s = s->data.sets[i];
 628        }
 629        if (!s->data.marked[idnum])
 630                marks_set_count++;
 631        s->data.marked[idnum] = oe;
 632}
 633
 634static struct object_entry *find_mark(uintmax_t idnum)
 635{
 636        uintmax_t orig_idnum = idnum;
 637        struct mark_set *s = marks;
 638        struct object_entry *oe = NULL;
 639        if ((idnum >> s->shift) < 1024) {
 640                while (s && s->shift) {
 641                        uintmax_t i = idnum >> s->shift;
 642                        idnum -= i << s->shift;
 643                        s = s->data.sets[i];
 644                }
 645                if (s)
 646                        oe = s->data.marked[idnum];
 647        }
 648        if (!oe)
 649                die("mark :%" PRIuMAX " not declared", orig_idnum);
 650        return oe;
 651}
 652
 653static struct atom_str *to_atom(const char *s, unsigned short len)
 654{
 655        unsigned int hc = hc_str(s, len) % atom_table_sz;
 656        struct atom_str *c;
 657
 658        for (c = atom_table[hc]; c; c = c->next_atom)
 659                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 660                        return c;
 661
 662        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 663        c->str_len = len;
 664        strncpy(c->str_dat, s, len);
 665        c->str_dat[len] = 0;
 666        c->next_atom = atom_table[hc];
 667        atom_table[hc] = c;
 668        atom_cnt++;
 669        return c;
 670}
 671
 672static struct branch *lookup_branch(const char *name)
 673{
 674        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 675        struct branch *b;
 676
 677        for (b = branch_table[hc]; b; b = b->table_next_branch)
 678                if (!strcmp(name, b->name))
 679                        return b;
 680        return NULL;
 681}
 682
 683static struct branch *new_branch(const char *name)
 684{
 685        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 686        struct branch *b = lookup_branch(name);
 687
 688        if (b)
 689                die("Invalid attempt to create duplicate branch: %s", name);
 690        switch (check_ref_format(name)) {
 691        case 0: break; /* its valid */
 692        case CHECK_REF_FORMAT_ONELEVEL:
 693                break; /* valid, but too few '/', allow anyway */
 694        default:
 695                die("Branch name doesn't conform to GIT standards: %s", name);
 696        }
 697
 698        b = pool_calloc(1, sizeof(struct branch));
 699        b->name = pool_strdup(name);
 700        b->table_next_branch = branch_table[hc];
 701        b->branch_tree.versions[0].mode = S_IFDIR;
 702        b->branch_tree.versions[1].mode = S_IFDIR;
 703        b->active = 0;
 704        b->pack_id = MAX_PACK_ID;
 705        branch_table[hc] = b;
 706        branch_count++;
 707        return b;
 708}
 709
 710static unsigned int hc_entries(unsigned int cnt)
 711{
 712        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 713        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 714}
 715
 716static struct tree_content *new_tree_content(unsigned int cnt)
 717{
 718        struct avail_tree_content *f, *l = NULL;
 719        struct tree_content *t;
 720        unsigned int hc = hc_entries(cnt);
 721
 722        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 723                if (f->entry_capacity >= cnt)
 724                        break;
 725
 726        if (f) {
 727                if (l)
 728                        l->next_avail = f->next_avail;
 729                else
 730                        avail_tree_table[hc] = f->next_avail;
 731        } else {
 732                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 733                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 734                f->entry_capacity = cnt;
 735        }
 736
 737        t = (struct tree_content*)f;
 738        t->entry_count = 0;
 739        t->delta_depth = 0;
 740        return t;
 741}
 742
 743static void release_tree_entry(struct tree_entry *e);
 744static void release_tree_content(struct tree_content *t)
 745{
 746        struct avail_tree_content *f = (struct avail_tree_content*)t;
 747        unsigned int hc = hc_entries(f->entry_capacity);
 748        f->next_avail = avail_tree_table[hc];
 749        avail_tree_table[hc] = f;
 750}
 751
 752static void release_tree_content_recursive(struct tree_content *t)
 753{
 754        unsigned int i;
 755        for (i = 0; i < t->entry_count; i++)
 756                release_tree_entry(t->entries[i]);
 757        release_tree_content(t);
 758}
 759
 760static struct tree_content *grow_tree_content(
 761        struct tree_content *t,
 762        int amt)
 763{
 764        struct tree_content *r = new_tree_content(t->entry_count + amt);
 765        r->entry_count = t->entry_count;
 766        r->delta_depth = t->delta_depth;
 767        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 768        release_tree_content(t);
 769        return r;
 770}
 771
 772static struct tree_entry *new_tree_entry(void)
 773{
 774        struct tree_entry *e;
 775
 776        if (!avail_tree_entry) {
 777                unsigned int n = tree_entry_alloc;
 778                total_allocd += n * sizeof(struct tree_entry);
 779                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 780                while (n-- > 1) {
 781                        *((void**)e) = e + 1;
 782                        e++;
 783                }
 784                *((void**)e) = NULL;
 785        }
 786
 787        e = avail_tree_entry;
 788        avail_tree_entry = *((void**)e);
 789        return e;
 790}
 791
 792static void release_tree_entry(struct tree_entry *e)
 793{
 794        if (e->tree)
 795                release_tree_content_recursive(e->tree);
 796        *((void**)e) = avail_tree_entry;
 797        avail_tree_entry = e;
 798}
 799
 800static struct tree_content *dup_tree_content(struct tree_content *s)
 801{
 802        struct tree_content *d;
 803        struct tree_entry *a, *b;
 804        unsigned int i;
 805
 806        if (!s)
 807                return NULL;
 808        d = new_tree_content(s->entry_count);
 809        for (i = 0; i < s->entry_count; i++) {
 810                a = s->entries[i];
 811                b = new_tree_entry();
 812                memcpy(b, a, sizeof(*a));
 813                if (a->tree && is_null_sha1(b->versions[1].sha1))
 814                        b->tree = dup_tree_content(a->tree);
 815                else
 816                        b->tree = NULL;
 817                d->entries[i] = b;
 818        }
 819        d->entry_count = s->entry_count;
 820        d->delta_depth = s->delta_depth;
 821
 822        return d;
 823}
 824
 825static void start_packfile(void)
 826{
 827        static char tmpfile[PATH_MAX];
 828        struct packed_git *p;
 829        struct pack_header hdr;
 830        int pack_fd;
 831
 832        pack_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 833                              "pack/tmp_pack_XXXXXX");
 834        p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
 835        strcpy(p->pack_name, tmpfile);
 836        p->pack_fd = pack_fd;
 837
 838        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 839        hdr.hdr_version = htonl(2);
 840        hdr.hdr_entries = 0;
 841        write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 842
 843        pack_data = p;
 844        pack_size = sizeof(hdr);
 845        object_count = 0;
 846
 847        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 848        all_packs[pack_id] = p;
 849}
 850
 851static int oecmp (const void *a_, const void *b_)
 852{
 853        struct object_entry *a = *((struct object_entry**)a_);
 854        struct object_entry *b = *((struct object_entry**)b_);
 855        return hashcmp(a->sha1, b->sha1);
 856}
 857
 858static char *create_index(void)
 859{
 860        static char tmpfile[PATH_MAX];
 861        git_SHA_CTX ctx;
 862        struct sha1file *f;
 863        struct object_entry **idx, **c, **last, *e;
 864        struct object_entry_pool *o;
 865        uint32_t array[256];
 866        int i, idx_fd;
 867
 868        /* Build the sorted table of object IDs. */
 869        idx = xmalloc(object_count * sizeof(struct object_entry*));
 870        c = idx;
 871        for (o = blocks; o; o = o->next_pool)
 872                for (e = o->next_free; e-- != o->entries;)
 873                        if (pack_id == e->pack_id)
 874                                *c++ = e;
 875        last = idx + object_count;
 876        if (c != last)
 877                die("internal consistency error creating the index");
 878        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 879
 880        /* Generate the fan-out array. */
 881        c = idx;
 882        for (i = 0; i < 256; i++) {
 883                struct object_entry **next = c;
 884                while (next < last) {
 885                        if ((*next)->sha1[0] != i)
 886                                break;
 887                        next++;
 888                }
 889                array[i] = htonl(next - idx);
 890                c = next;
 891        }
 892
 893        idx_fd = odb_mkstemp(tmpfile, sizeof(tmpfile),
 894                             "pack/tmp_idx_XXXXXX");
 895        f = sha1fd(idx_fd, tmpfile);
 896        sha1write(f, array, 256 * sizeof(int));
 897        git_SHA1_Init(&ctx);
 898        for (c = idx; c != last; c++) {
 899                uint32_t offset = htonl((*c)->offset);
 900                sha1write(f, &offset, 4);
 901                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 902                git_SHA1_Update(&ctx, (*c)->sha1, 20);
 903        }
 904        sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
 905        sha1close(f, NULL, CSUM_FSYNC);
 906        free(idx);
 907        git_SHA1_Final(pack_data->sha1, &ctx);
 908        return tmpfile;
 909}
 910
 911static char *keep_pack(char *curr_index_name)
 912{
 913        static char name[PATH_MAX];
 914        static const char *keep_msg = "fast-import";
 915        int keep_fd;
 916
 917        keep_fd = odb_pack_keep(name, sizeof(name), pack_data->sha1);
 918        if (keep_fd < 0)
 919                die_errno("cannot create keep file");
 920        write_or_die(keep_fd, keep_msg, strlen(keep_msg));
 921        if (close(keep_fd))
 922                die_errno("failed to write keep file");
 923
 924        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 925                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 926        if (move_temp_to_file(pack_data->pack_name, name))
 927                die("cannot store pack file");
 928
 929        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 930                 get_object_directory(), sha1_to_hex(pack_data->sha1));
 931        if (move_temp_to_file(curr_index_name, name))
 932                die("cannot store index file");
 933        return name;
 934}
 935
 936static void unkeep_all_packs(void)
 937{
 938        static char name[PATH_MAX];
 939        int k;
 940
 941        for (k = 0; k < pack_id; k++) {
 942                struct packed_git *p = all_packs[k];
 943                snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 944                         get_object_directory(), sha1_to_hex(p->sha1));
 945                unlink_or_warn(name);
 946        }
 947}
 948
 949static void end_packfile(void)
 950{
 951        struct packed_git *old_p = pack_data, *new_p;
 952
 953        clear_delta_base_cache();
 954        if (object_count) {
 955                char *idx_name;
 956                int i;
 957                struct branch *b;
 958                struct tag *t;
 959
 960                close_pack_windows(pack_data);
 961                fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
 962                                    pack_data->pack_name, object_count,
 963                                    NULL, 0);
 964                close(pack_data->pack_fd);
 965                idx_name = keep_pack(create_index());
 966
 967                /* Register the packfile with core git's machinery. */
 968                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 969                if (!new_p)
 970                        die("core git rejected index %s", idx_name);
 971                all_packs[pack_id] = new_p;
 972                install_packed_git(new_p);
 973
 974                /* Print the boundary */
 975                if (pack_edges) {
 976                        fprintf(pack_edges, "%s:", new_p->pack_name);
 977                        for (i = 0; i < branch_table_sz; i++) {
 978                                for (b = branch_table[i]; b; b = b->table_next_branch) {
 979                                        if (b->pack_id == pack_id)
 980                                                fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
 981                                }
 982                        }
 983                        for (t = first_tag; t; t = t->next_tag) {
 984                                if (t->pack_id == pack_id)
 985                                        fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
 986                        }
 987                        fputc('\n', pack_edges);
 988                        fflush(pack_edges);
 989                }
 990
 991                pack_id++;
 992        }
 993        else {
 994                close(old_p->pack_fd);
 995                unlink_or_warn(old_p->pack_name);
 996        }
 997        free(old_p);
 998
 999        /* We can't carry a delta across packfiles. */
1000        strbuf_release(&last_blob.data);
1001        last_blob.offset = 0;
1002        last_blob.depth = 0;
1003}
1004
1005static void cycle_packfile(void)
1006{
1007        end_packfile();
1008        start_packfile();
1009}
1010
1011static size_t encode_header(
1012        enum object_type type,
1013        size_t size,
1014        unsigned char *hdr)
1015{
1016        int n = 1;
1017        unsigned char c;
1018
1019        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
1020                die("bad type %d", type);
1021
1022        c = (type << 4) | (size & 15);
1023        size >>= 4;
1024        while (size) {
1025                *hdr++ = c | 0x80;
1026                c = size & 0x7f;
1027                size >>= 7;
1028                n++;
1029        }
1030        *hdr = c;
1031        return n;
1032}
1033
1034static int store_object(
1035        enum object_type type,
1036        struct strbuf *dat,
1037        struct last_object *last,
1038        unsigned char *sha1out,
1039        uintmax_t mark)
1040{
1041        void *out, *delta;
1042        struct object_entry *e;
1043        unsigned char hdr[96];
1044        unsigned char sha1[20];
1045        unsigned long hdrlen, deltalen;
1046        git_SHA_CTX c;
1047        z_stream s;
1048
1049        hdrlen = sprintf((char *)hdr,"%s %lu", typename(type),
1050                (unsigned long)dat->len) + 1;
1051        git_SHA1_Init(&c);
1052        git_SHA1_Update(&c, hdr, hdrlen);
1053        git_SHA1_Update(&c, dat->buf, dat->len);
1054        git_SHA1_Final(sha1, &c);
1055        if (sha1out)
1056                hashcpy(sha1out, sha1);
1057
1058        e = insert_object(sha1);
1059        if (mark)
1060                insert_mark(mark, e);
1061        if (e->offset) {
1062                duplicate_count_by_type[type]++;
1063                return 1;
1064        } else if (find_sha1_pack(sha1, packed_git)) {
1065                e->type = type;
1066                e->pack_id = MAX_PACK_ID;
1067                e->offset = 1; /* just not zero! */
1068                duplicate_count_by_type[type]++;
1069                return 1;
1070        }
1071
1072        if (last && last->data.buf && last->depth < max_depth) {
1073                delta = diff_delta(last->data.buf, last->data.len,
1074                        dat->buf, dat->len,
1075                        &deltalen, 0);
1076                if (delta && deltalen >= dat->len) {
1077                        free(delta);
1078                        delta = NULL;
1079                }
1080        } else
1081                delta = NULL;
1082
1083        memset(&s, 0, sizeof(s));
1084        deflateInit(&s, pack_compression_level);
1085        if (delta) {
1086                s.next_in = delta;
1087                s.avail_in = deltalen;
1088        } else {
1089                s.next_in = (void *)dat->buf;
1090                s.avail_in = dat->len;
1091        }
1092        s.avail_out = deflateBound(&s, s.avail_in);
1093        s.next_out = out = xmalloc(s.avail_out);
1094        while (deflate(&s, Z_FINISH) == Z_OK)
1095                /* nothing */;
1096        deflateEnd(&s);
1097
1098        /* Determine if we should auto-checkpoint. */
1099        if ((pack_size + 60 + s.total_out) > max_packsize
1100                || (pack_size + 60 + s.total_out) < pack_size) {
1101
1102                /* This new object needs to *not* have the current pack_id. */
1103                e->pack_id = pack_id + 1;
1104                cycle_packfile();
1105
1106                /* We cannot carry a delta into the new pack. */
1107                if (delta) {
1108                        free(delta);
1109                        delta = NULL;
1110
1111                        memset(&s, 0, sizeof(s));
1112                        deflateInit(&s, pack_compression_level);
1113                        s.next_in = (void *)dat->buf;
1114                        s.avail_in = dat->len;
1115                        s.avail_out = deflateBound(&s, s.avail_in);
1116                        s.next_out = out = xrealloc(out, s.avail_out);
1117                        while (deflate(&s, Z_FINISH) == Z_OK)
1118                                /* nothing */;
1119                        deflateEnd(&s);
1120                }
1121        }
1122
1123        e->type = type;
1124        e->pack_id = pack_id;
1125        e->offset = pack_size;
1126        object_count++;
1127        object_count_by_type[type]++;
1128
1129        if (delta) {
1130                unsigned long ofs = e->offset - last->offset;
1131                unsigned pos = sizeof(hdr) - 1;
1132
1133                delta_count_by_type[type]++;
1134                e->depth = last->depth + 1;
1135
1136                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
1137                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1138                pack_size += hdrlen;
1139
1140                hdr[pos] = ofs & 127;
1141                while (ofs >>= 7)
1142                        hdr[--pos] = 128 | (--ofs & 127);
1143                write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
1144                pack_size += sizeof(hdr) - pos;
1145        } else {
1146                e->depth = 0;
1147                hdrlen = encode_header(type, dat->len, hdr);
1148                write_or_die(pack_data->pack_fd, hdr, hdrlen);
1149                pack_size += hdrlen;
1150        }
1151
1152        write_or_die(pack_data->pack_fd, out, s.total_out);
1153        pack_size += s.total_out;
1154
1155        free(out);
1156        free(delta);
1157        if (last) {
1158                if (last->no_swap) {
1159                        last->data = *dat;
1160                } else {
1161                        strbuf_swap(&last->data, dat);
1162                }
1163                last->offset = e->offset;
1164                last->depth = e->depth;
1165        }
1166        return 0;
1167}
1168
1169/* All calls must be guarded by find_object() or find_mark() to
1170 * ensure the 'struct object_entry' passed was written by this
1171 * process instance.  We unpack the entry by the offset, avoiding
1172 * the need for the corresponding .idx file.  This unpacking rule
1173 * works because we only use OBJ_REF_DELTA within the packfiles
1174 * created by fast-import.
1175 *
1176 * oe must not be NULL.  Such an oe usually comes from giving
1177 * an unknown SHA-1 to find_object() or an undefined mark to
1178 * find_mark().  Callers must test for this condition and use
1179 * the standard read_sha1_file() when it happens.
1180 *
1181 * oe->pack_id must not be MAX_PACK_ID.  Such an oe is usually from
1182 * find_mark(), where the mark was reloaded from an existing marks
1183 * file and is referencing an object that this fast-import process
1184 * instance did not write out to a packfile.  Callers must test for
1185 * this condition and use read_sha1_file() instead.
1186 */
1187static void *gfi_unpack_entry(
1188        struct object_entry *oe,
1189        unsigned long *sizep)
1190{
1191        enum object_type type;
1192        struct packed_git *p = all_packs[oe->pack_id];
1193        if (p == pack_data && p->pack_size < (pack_size + 20)) {
1194                /* The object is stored in the packfile we are writing to
1195                 * and we have modified it since the last time we scanned
1196                 * back to read a previously written object.  If an old
1197                 * window covered [p->pack_size, p->pack_size + 20) its
1198                 * data is stale and is not valid.  Closing all windows
1199                 * and updating the packfile length ensures we can read
1200                 * the newly written data.
1201                 */
1202                close_pack_windows(p);
1203
1204                /* We have to offer 20 bytes additional on the end of
1205                 * the packfile as the core unpacker code assumes the
1206                 * footer is present at the file end and must promise
1207                 * at least 20 bytes within any window it maps.  But
1208                 * we don't actually create the footer here.
1209                 */
1210                p->pack_size = pack_size + 20;
1211        }
1212        return unpack_entry(p, oe->offset, &type, sizep);
1213}
1214
1215static const char *get_mode(const char *str, uint16_t *modep)
1216{
1217        unsigned char c;
1218        uint16_t mode = 0;
1219
1220        while ((c = *str++) != ' ') {
1221                if (c < '0' || c > '7')
1222                        return NULL;
1223                mode = (mode << 3) + (c - '0');
1224        }
1225        *modep = mode;
1226        return str;
1227}
1228
1229static void load_tree(struct tree_entry *root)
1230{
1231        unsigned char *sha1 = root->versions[1].sha1;
1232        struct object_entry *myoe;
1233        struct tree_content *t;
1234        unsigned long size;
1235        char *buf;
1236        const char *c;
1237
1238        root->tree = t = new_tree_content(8);
1239        if (is_null_sha1(sha1))
1240                return;
1241
1242        myoe = find_object(sha1);
1243        if (myoe && myoe->pack_id != MAX_PACK_ID) {
1244                if (myoe->type != OBJ_TREE)
1245                        die("Not a tree: %s", sha1_to_hex(sha1));
1246                t->delta_depth = myoe->depth;
1247                buf = gfi_unpack_entry(myoe, &size);
1248                if (!buf)
1249                        die("Can't load tree %s", sha1_to_hex(sha1));
1250        } else {
1251                enum object_type type;
1252                buf = read_sha1_file(sha1, &type, &size);
1253                if (!buf || type != OBJ_TREE)
1254                        die("Can't load tree %s", sha1_to_hex(sha1));
1255        }
1256
1257        c = buf;
1258        while (c != (buf + size)) {
1259                struct tree_entry *e = new_tree_entry();
1260
1261                if (t->entry_count == t->entry_capacity)
1262                        root->tree = t = grow_tree_content(t, t->entry_count);
1263                t->entries[t->entry_count++] = e;
1264
1265                e->tree = NULL;
1266                c = get_mode(c, &e->versions[1].mode);
1267                if (!c)
1268                        die("Corrupt mode in %s", sha1_to_hex(sha1));
1269                e->versions[0].mode = e->versions[1].mode;
1270                e->name = to_atom(c, strlen(c));
1271                c += e->name->str_len + 1;
1272                hashcpy(e->versions[0].sha1, (unsigned char *)c);
1273                hashcpy(e->versions[1].sha1, (unsigned char *)c);
1274                c += 20;
1275        }
1276        free(buf);
1277}
1278
1279static int tecmp0 (const void *_a, const void *_b)
1280{
1281        struct tree_entry *a = *((struct tree_entry**)_a);
1282        struct tree_entry *b = *((struct tree_entry**)_b);
1283        return base_name_compare(
1284                a->name->str_dat, a->name->str_len, a->versions[0].mode,
1285                b->name->str_dat, b->name->str_len, b->versions[0].mode);
1286}
1287
1288static int tecmp1 (const void *_a, const void *_b)
1289{
1290        struct tree_entry *a = *((struct tree_entry**)_a);
1291        struct tree_entry *b = *((struct tree_entry**)_b);
1292        return base_name_compare(
1293                a->name->str_dat, a->name->str_len, a->versions[1].mode,
1294                b->name->str_dat, b->name->str_len, b->versions[1].mode);
1295}
1296
1297static void mktree(struct tree_content *t, int v, struct strbuf *b)
1298{
1299        size_t maxlen = 0;
1300        unsigned int i;
1301
1302        if (!v)
1303                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1304        else
1305                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1306
1307        for (i = 0; i < t->entry_count; i++) {
1308                if (t->entries[i]->versions[v].mode)
1309                        maxlen += t->entries[i]->name->str_len + 34;
1310        }
1311
1312        strbuf_reset(b);
1313        strbuf_grow(b, maxlen);
1314        for (i = 0; i < t->entry_count; i++) {
1315                struct tree_entry *e = t->entries[i];
1316                if (!e->versions[v].mode)
1317                        continue;
1318                strbuf_addf(b, "%o %s%c", (unsigned int)e->versions[v].mode,
1319                                        e->name->str_dat, '\0');
1320                strbuf_add(b, e->versions[v].sha1, 20);
1321        }
1322}
1323
1324static void store_tree(struct tree_entry *root)
1325{
1326        struct tree_content *t = root->tree;
1327        unsigned int i, j, del;
1328        struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1329        struct object_entry *le;
1330
1331        if (!is_null_sha1(root->versions[1].sha1))
1332                return;
1333
1334        for (i = 0; i < t->entry_count; i++) {
1335                if (t->entries[i]->tree)
1336                        store_tree(t->entries[i]);
1337        }
1338
1339        le = find_object(root->versions[0].sha1);
1340        if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1341                mktree(t, 0, &old_tree);
1342                lo.data = old_tree;
1343                lo.offset = le->offset;
1344                lo.depth = t->delta_depth;
1345        }
1346
1347        mktree(t, 1, &new_tree);
1348        store_object(OBJ_TREE, &new_tree, &lo, root->versions[1].sha1, 0);
1349
1350        t->delta_depth = lo.depth;
1351        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1352                struct tree_entry *e = t->entries[i];
1353                if (e->versions[1].mode) {
1354                        e->versions[0].mode = e->versions[1].mode;
1355                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1356                        t->entries[j++] = e;
1357                } else {
1358                        release_tree_entry(e);
1359                        del++;
1360                }
1361        }
1362        t->entry_count -= del;
1363}
1364
1365static int tree_content_set(
1366        struct tree_entry *root,
1367        const char *p,
1368        const unsigned char *sha1,
1369        const uint16_t mode,
1370        struct tree_content *subtree)
1371{
1372        struct tree_content *t = root->tree;
1373        const char *slash1;
1374        unsigned int i, n;
1375        struct tree_entry *e;
1376
1377        slash1 = strchr(p, '/');
1378        if (slash1)
1379                n = slash1 - p;
1380        else
1381                n = strlen(p);
1382        if (!n)
1383                die("Empty path component found in input");
1384        if (!slash1 && !S_ISDIR(mode) && subtree)
1385                die("Non-directories cannot have subtrees");
1386
1387        for (i = 0; i < t->entry_count; i++) {
1388                e = t->entries[i];
1389                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1390                        if (!slash1) {
1391                                if (!S_ISDIR(mode)
1392                                                && e->versions[1].mode == mode
1393                                                && !hashcmp(e->versions[1].sha1, sha1))
1394                                        return 0;
1395                                e->versions[1].mode = mode;
1396                                hashcpy(e->versions[1].sha1, sha1);
1397                                if (e->tree)
1398                                        release_tree_content_recursive(e->tree);
1399                                e->tree = subtree;
1400                                hashclr(root->versions[1].sha1);
1401                                return 1;
1402                        }
1403                        if (!S_ISDIR(e->versions[1].mode)) {
1404                                e->tree = new_tree_content(8);
1405                                e->versions[1].mode = S_IFDIR;
1406                        }
1407                        if (!e->tree)
1408                                load_tree(e);
1409                        if (tree_content_set(e, slash1 + 1, sha1, mode, subtree)) {
1410                                hashclr(root->versions[1].sha1);
1411                                return 1;
1412                        }
1413                        return 0;
1414                }
1415        }
1416
1417        if (t->entry_count == t->entry_capacity)
1418                root->tree = t = grow_tree_content(t, t->entry_count);
1419        e = new_tree_entry();
1420        e->name = to_atom(p, n);
1421        e->versions[0].mode = 0;
1422        hashclr(e->versions[0].sha1);
1423        t->entries[t->entry_count++] = e;
1424        if (slash1) {
1425                e->tree = new_tree_content(8);
1426                e->versions[1].mode = S_IFDIR;
1427                tree_content_set(e, slash1 + 1, sha1, mode, subtree);
1428        } else {
1429                e->tree = subtree;
1430                e->versions[1].mode = mode;
1431                hashcpy(e->versions[1].sha1, sha1);
1432        }
1433        hashclr(root->versions[1].sha1);
1434        return 1;
1435}
1436
1437static int tree_content_remove(
1438        struct tree_entry *root,
1439        const char *p,
1440        struct tree_entry *backup_leaf)
1441{
1442        struct tree_content *t = root->tree;
1443        const char *slash1;
1444        unsigned int i, n;
1445        struct tree_entry *e;
1446
1447        slash1 = strchr(p, '/');
1448        if (slash1)
1449                n = slash1 - p;
1450        else
1451                n = strlen(p);
1452
1453        for (i = 0; i < t->entry_count; i++) {
1454                e = t->entries[i];
1455                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1456                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1457                                goto del_entry;
1458                        if (!e->tree)
1459                                load_tree(e);
1460                        if (tree_content_remove(e, slash1 + 1, backup_leaf)) {
1461                                for (n = 0; n < e->tree->entry_count; n++) {
1462                                        if (e->tree->entries[n]->versions[1].mode) {
1463                                                hashclr(root->versions[1].sha1);
1464                                                return 1;
1465                                        }
1466                                }
1467                                backup_leaf = NULL;
1468                                goto del_entry;
1469                        }
1470                        return 0;
1471                }
1472        }
1473        return 0;
1474
1475del_entry:
1476        if (backup_leaf)
1477                memcpy(backup_leaf, e, sizeof(*backup_leaf));
1478        else if (e->tree)
1479                release_tree_content_recursive(e->tree);
1480        e->tree = NULL;
1481        e->versions[1].mode = 0;
1482        hashclr(e->versions[1].sha1);
1483        hashclr(root->versions[1].sha1);
1484        return 1;
1485}
1486
1487static int tree_content_get(
1488        struct tree_entry *root,
1489        const char *p,
1490        struct tree_entry *leaf)
1491{
1492        struct tree_content *t = root->tree;
1493        const char *slash1;
1494        unsigned int i, n;
1495        struct tree_entry *e;
1496
1497        slash1 = strchr(p, '/');
1498        if (slash1)
1499                n = slash1 - p;
1500        else
1501                n = strlen(p);
1502
1503        for (i = 0; i < t->entry_count; i++) {
1504                e = t->entries[i];
1505                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1506                        if (!slash1) {
1507                                memcpy(leaf, e, sizeof(*leaf));
1508                                if (e->tree && is_null_sha1(e->versions[1].sha1))
1509                                        leaf->tree = dup_tree_content(e->tree);
1510                                else
1511                                        leaf->tree = NULL;
1512                                return 1;
1513                        }
1514                        if (!S_ISDIR(e->versions[1].mode))
1515                                return 0;
1516                        if (!e->tree)
1517                                load_tree(e);
1518                        return tree_content_get(e, slash1 + 1, leaf);
1519                }
1520        }
1521        return 0;
1522}
1523
1524static int update_branch(struct branch *b)
1525{
1526        static const char *msg = "fast-import";
1527        struct ref_lock *lock;
1528        unsigned char old_sha1[20];
1529
1530        if (is_null_sha1(b->sha1))
1531                return 0;
1532        if (read_ref(b->name, old_sha1))
1533                hashclr(old_sha1);
1534        lock = lock_any_ref_for_update(b->name, old_sha1, 0);
1535        if (!lock)
1536                return error("Unable to lock %s", b->name);
1537        if (!force_update && !is_null_sha1(old_sha1)) {
1538                struct commit *old_cmit, *new_cmit;
1539
1540                old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1541                new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1542                if (!old_cmit || !new_cmit) {
1543                        unlock_ref(lock);
1544                        return error("Branch %s is missing commits.", b->name);
1545                }
1546
1547                if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1548                        unlock_ref(lock);
1549                        warning("Not updating %s"
1550                                " (new tip %s does not contain %s)",
1551                                b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1552                        return -1;
1553                }
1554        }
1555        if (write_ref_sha1(lock, b->sha1, msg) < 0)
1556                return error("Unable to update %s", b->name);
1557        return 0;
1558}
1559
1560static void dump_branches(void)
1561{
1562        unsigned int i;
1563        struct branch *b;
1564
1565        for (i = 0; i < branch_table_sz; i++) {
1566                for (b = branch_table[i]; b; b = b->table_next_branch)
1567                        failure |= update_branch(b);
1568        }
1569}
1570
1571static void dump_tags(void)
1572{
1573        static const char *msg = "fast-import";
1574        struct tag *t;
1575        struct ref_lock *lock;
1576        char ref_name[PATH_MAX];
1577
1578        for (t = first_tag; t; t = t->next_tag) {
1579                sprintf(ref_name, "tags/%s", t->name);
1580                lock = lock_ref_sha1(ref_name, NULL);
1581                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1582                        failure |= error("Unable to update %s", ref_name);
1583        }
1584}
1585
1586static void dump_marks_helper(FILE *f,
1587        uintmax_t base,
1588        struct mark_set *m)
1589{
1590        uintmax_t k;
1591        if (m->shift) {
1592                for (k = 0; k < 1024; k++) {
1593                        if (m->data.sets[k])
1594                                dump_marks_helper(f, (base + k) << m->shift,
1595                                        m->data.sets[k]);
1596                }
1597        } else {
1598                for (k = 0; k < 1024; k++) {
1599                        if (m->data.marked[k])
1600                                fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1601                                        sha1_to_hex(m->data.marked[k]->sha1));
1602                }
1603        }
1604}
1605
1606static void dump_marks(void)
1607{
1608        static struct lock_file mark_lock;
1609        int mark_fd;
1610        FILE *f;
1611
1612        if (!export_marks_file)
1613                return;
1614
1615        mark_fd = hold_lock_file_for_update(&mark_lock, export_marks_file, 0);
1616        if (mark_fd < 0) {
1617                failure |= error("Unable to write marks file %s: %s",
1618                        export_marks_file, strerror(errno));
1619                return;
1620        }
1621
1622        f = fdopen(mark_fd, "w");
1623        if (!f) {
1624                int saved_errno = errno;
1625                rollback_lock_file(&mark_lock);
1626                failure |= error("Unable to write marks file %s: %s",
1627                        export_marks_file, strerror(saved_errno));
1628                return;
1629        }
1630
1631        /*
1632         * Since the lock file was fdopen()'ed, it should not be close()'ed.
1633         * Assign -1 to the lock file descriptor so that commit_lock_file()
1634         * won't try to close() it.
1635         */
1636        mark_lock.fd = -1;
1637
1638        dump_marks_helper(f, 0, marks);
1639        if (ferror(f) || fclose(f)) {
1640                int saved_errno = errno;
1641                rollback_lock_file(&mark_lock);
1642                failure |= error("Unable to write marks file %s: %s",
1643                        export_marks_file, strerror(saved_errno));
1644                return;
1645        }
1646
1647        if (commit_lock_file(&mark_lock)) {
1648                int saved_errno = errno;
1649                rollback_lock_file(&mark_lock);
1650                failure |= error("Unable to commit marks file %s: %s",
1651                        export_marks_file, strerror(saved_errno));
1652                return;
1653        }
1654}
1655
1656static void read_marks(void)
1657{
1658        char line[512];
1659        FILE *f = fopen(import_marks_file, "r");
1660        if (!f)
1661                die_errno("cannot read '%s'", import_marks_file);
1662        while (fgets(line, sizeof(line), f)) {
1663                uintmax_t mark;
1664                char *end;
1665                unsigned char sha1[20];
1666                struct object_entry *e;
1667
1668                end = strchr(line, '\n');
1669                if (line[0] != ':' || !end)
1670                        die("corrupt mark line: %s", line);
1671                *end = 0;
1672                mark = strtoumax(line + 1, &end, 10);
1673                if (!mark || end == line + 1
1674                        || *end != ' ' || get_sha1(end + 1, sha1))
1675                        die("corrupt mark line: %s", line);
1676                e = find_object(sha1);
1677                if (!e) {
1678                        enum object_type type = sha1_object_info(sha1, NULL);
1679                        if (type < 0)
1680                                die("object not found: %s", sha1_to_hex(sha1));
1681                        e = insert_object(sha1);
1682                        e->type = type;
1683                        e->pack_id = MAX_PACK_ID;
1684                        e->offset = 1; /* just not zero! */
1685                }
1686                insert_mark(mark, e);
1687        }
1688        fclose(f);
1689}
1690
1691
1692static int read_next_command(void)
1693{
1694        static int stdin_eof = 0;
1695
1696        if (stdin_eof) {
1697                unread_command_buf = 0;
1698                return EOF;
1699        }
1700
1701        do {
1702                if (unread_command_buf) {
1703                        unread_command_buf = 0;
1704                } else {
1705                        struct recent_command *rc;
1706
1707                        strbuf_detach(&command_buf, NULL);
1708                        stdin_eof = strbuf_getline(&command_buf, stdin, '\n');
1709                        if (stdin_eof)
1710                                return EOF;
1711
1712                        if (!seen_data_command
1713                                && prefixcmp(command_buf.buf, "feature ")
1714                                && prefixcmp(command_buf.buf, "option ")) {
1715                                parse_argv();
1716                        }
1717
1718                        rc = rc_free;
1719                        if (rc)
1720                                rc_free = rc->next;
1721                        else {
1722                                rc = cmd_hist.next;
1723                                cmd_hist.next = rc->next;
1724                                cmd_hist.next->prev = &cmd_hist;
1725                                free(rc->buf);
1726                        }
1727
1728                        rc->buf = command_buf.buf;
1729                        rc->prev = cmd_tail;
1730                        rc->next = cmd_hist.prev;
1731                        rc->prev->next = rc;
1732                        cmd_tail = rc;
1733                }
1734        } while (command_buf.buf[0] == '#');
1735
1736        return 0;
1737}
1738
1739static void skip_optional_lf(void)
1740{
1741        int term_char = fgetc(stdin);
1742        if (term_char != '\n' && term_char != EOF)
1743                ungetc(term_char, stdin);
1744}
1745
1746static void parse_mark(void)
1747{
1748        if (!prefixcmp(command_buf.buf, "mark :")) {
1749                next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1750                read_next_command();
1751        }
1752        else
1753                next_mark = 0;
1754}
1755
1756static void parse_data(struct strbuf *sb)
1757{
1758        strbuf_reset(sb);
1759
1760        if (prefixcmp(command_buf.buf, "data "))
1761                die("Expected 'data n' command, found: %s", command_buf.buf);
1762
1763        if (!prefixcmp(command_buf.buf + 5, "<<")) {
1764                char *term = xstrdup(command_buf.buf + 5 + 2);
1765                size_t term_len = command_buf.len - 5 - 2;
1766
1767                strbuf_detach(&command_buf, NULL);
1768                for (;;) {
1769                        if (strbuf_getline(&command_buf, stdin, '\n') == EOF)
1770                                die("EOF in data (terminator '%s' not found)", term);
1771                        if (term_len == command_buf.len
1772                                && !strcmp(term, command_buf.buf))
1773                                break;
1774                        strbuf_addbuf(sb, &command_buf);
1775                        strbuf_addch(sb, '\n');
1776                }
1777                free(term);
1778        }
1779        else {
1780                size_t n = 0, length;
1781
1782                length = strtoul(command_buf.buf + 5, NULL, 10);
1783
1784                while (n < length) {
1785                        size_t s = strbuf_fread(sb, length - n, stdin);
1786                        if (!s && feof(stdin))
1787                                die("EOF in data (%lu bytes remaining)",
1788                                        (unsigned long)(length - n));
1789                        n += s;
1790                }
1791        }
1792
1793        skip_optional_lf();
1794}
1795
1796static int validate_raw_date(const char *src, char *result, int maxlen)
1797{
1798        const char *orig_src = src;
1799        char *endp;
1800        unsigned long num;
1801
1802        errno = 0;
1803
1804        num = strtoul(src, &endp, 10);
1805        /* NEEDSWORK: perhaps check for reasonable values? */
1806        if (errno || endp == src || *endp != ' ')
1807                return -1;
1808
1809        src = endp + 1;
1810        if (*src != '-' && *src != '+')
1811                return -1;
1812
1813        num = strtoul(src + 1, &endp, 10);
1814        if (errno || endp == src + 1 || *endp || (endp - orig_src) >= maxlen ||
1815            1400 < num)
1816                return -1;
1817
1818        strcpy(result, orig_src);
1819        return 0;
1820}
1821
1822static char *parse_ident(const char *buf)
1823{
1824        const char *gt;
1825        size_t name_len;
1826        char *ident;
1827
1828        gt = strrchr(buf, '>');
1829        if (!gt)
1830                die("Missing > in ident string: %s", buf);
1831        gt++;
1832        if (*gt != ' ')
1833                die("Missing space after > in ident string: %s", buf);
1834        gt++;
1835        name_len = gt - buf;
1836        ident = xmalloc(name_len + 24);
1837        strncpy(ident, buf, name_len);
1838
1839        switch (whenspec) {
1840        case WHENSPEC_RAW:
1841                if (validate_raw_date(gt, ident + name_len, 24) < 0)
1842                        die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1843                break;
1844        case WHENSPEC_RFC2822:
1845                if (parse_date(gt, ident + name_len, 24) < 0)
1846                        die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1847                break;
1848        case WHENSPEC_NOW:
1849                if (strcmp("now", gt))
1850                        die("Date in ident must be 'now': %s", buf);
1851                datestamp(ident + name_len, 24);
1852                break;
1853        }
1854
1855        return ident;
1856}
1857
1858static void parse_new_blob(void)
1859{
1860        static struct strbuf buf = STRBUF_INIT;
1861
1862        read_next_command();
1863        parse_mark();
1864        parse_data(&buf);
1865        store_object(OBJ_BLOB, &buf, &last_blob, NULL, next_mark);
1866}
1867
1868static void unload_one_branch(void)
1869{
1870        while (cur_active_branches
1871                && cur_active_branches >= max_active_branches) {
1872                uintmax_t min_commit = ULONG_MAX;
1873                struct branch *e, *l = NULL, *p = NULL;
1874
1875                for (e = active_branches; e; e = e->active_next_branch) {
1876                        if (e->last_commit < min_commit) {
1877                                p = l;
1878                                min_commit = e->last_commit;
1879                        }
1880                        l = e;
1881                }
1882
1883                if (p) {
1884                        e = p->active_next_branch;
1885                        p->active_next_branch = e->active_next_branch;
1886                } else {
1887                        e = active_branches;
1888                        active_branches = e->active_next_branch;
1889                }
1890                e->active = 0;
1891                e->active_next_branch = NULL;
1892                if (e->branch_tree.tree) {
1893                        release_tree_content_recursive(e->branch_tree.tree);
1894                        e->branch_tree.tree = NULL;
1895                }
1896                cur_active_branches--;
1897        }
1898}
1899
1900static void load_branch(struct branch *b)
1901{
1902        load_tree(&b->branch_tree);
1903        if (!b->active) {
1904                b->active = 1;
1905                b->active_next_branch = active_branches;
1906                active_branches = b;
1907                cur_active_branches++;
1908                branch_load_count++;
1909        }
1910}
1911
1912static void file_change_m(struct branch *b)
1913{
1914        const char *p = command_buf.buf + 2;
1915        static struct strbuf uq = STRBUF_INIT;
1916        const char *endp;
1917        struct object_entry *oe = oe;
1918        unsigned char sha1[20];
1919        uint16_t mode, inline_data = 0;
1920
1921        p = get_mode(p, &mode);
1922        if (!p)
1923                die("Corrupt mode: %s", command_buf.buf);
1924        switch (mode) {
1925        case 0644:
1926        case 0755:
1927                mode |= S_IFREG;
1928        case S_IFREG | 0644:
1929        case S_IFREG | 0755:
1930        case S_IFLNK:
1931        case S_IFGITLINK:
1932                /* ok */
1933                break;
1934        default:
1935                die("Corrupt mode: %s", command_buf.buf);
1936        }
1937
1938        if (*p == ':') {
1939                char *x;
1940                oe = find_mark(strtoumax(p + 1, &x, 10));
1941                hashcpy(sha1, oe->sha1);
1942                p = x;
1943        } else if (!prefixcmp(p, "inline")) {
1944                inline_data = 1;
1945                p += 6;
1946        } else {
1947                if (get_sha1_hex(p, sha1))
1948                        die("Invalid SHA1: %s", command_buf.buf);
1949                oe = find_object(sha1);
1950                p += 40;
1951        }
1952        if (*p++ != ' ')
1953                die("Missing space after SHA1: %s", command_buf.buf);
1954
1955        strbuf_reset(&uq);
1956        if (!unquote_c_style(&uq, p, &endp)) {
1957                if (*endp)
1958                        die("Garbage after path in: %s", command_buf.buf);
1959                p = uq.buf;
1960        }
1961
1962        if (S_ISGITLINK(mode)) {
1963                if (inline_data)
1964                        die("Git links cannot be specified 'inline': %s",
1965                                command_buf.buf);
1966                else if (oe) {
1967                        if (oe->type != OBJ_COMMIT)
1968                                die("Not a commit (actually a %s): %s",
1969                                        typename(oe->type), command_buf.buf);
1970                }
1971                /*
1972                 * Accept the sha1 without checking; it expected to be in
1973                 * another repository.
1974                 */
1975        } else if (inline_data) {
1976                static struct strbuf buf = STRBUF_INIT;
1977
1978                if (p != uq.buf) {
1979                        strbuf_addstr(&uq, p);
1980                        p = uq.buf;
1981                }
1982                read_next_command();
1983                parse_data(&buf);
1984                store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
1985        } else if (oe) {
1986                if (oe->type != OBJ_BLOB)
1987                        die("Not a blob (actually a %s): %s",
1988                                typename(oe->type), command_buf.buf);
1989        } else {
1990                enum object_type type = sha1_object_info(sha1, NULL);
1991                if (type < 0)
1992                        die("Blob not found: %s", command_buf.buf);
1993                if (type != OBJ_BLOB)
1994                        die("Not a blob (actually a %s): %s",
1995                            typename(type), command_buf.buf);
1996        }
1997
1998        tree_content_set(&b->branch_tree, p, sha1, mode, NULL);
1999}
2000
2001static void file_change_d(struct branch *b)
2002{
2003        const char *p = command_buf.buf + 2;
2004        static struct strbuf uq = STRBUF_INIT;
2005        const char *endp;
2006
2007        strbuf_reset(&uq);
2008        if (!unquote_c_style(&uq, p, &endp)) {
2009                if (*endp)
2010                        die("Garbage after path in: %s", command_buf.buf);
2011                p = uq.buf;
2012        }
2013        tree_content_remove(&b->branch_tree, p, NULL);
2014}
2015
2016static void file_change_cr(struct branch *b, int rename)
2017{
2018        const char *s, *d;
2019        static struct strbuf s_uq = STRBUF_INIT;
2020        static struct strbuf d_uq = STRBUF_INIT;
2021        const char *endp;
2022        struct tree_entry leaf;
2023
2024        s = command_buf.buf + 2;
2025        strbuf_reset(&s_uq);
2026        if (!unquote_c_style(&s_uq, s, &endp)) {
2027                if (*endp != ' ')
2028                        die("Missing space after source: %s", command_buf.buf);
2029        } else {
2030                endp = strchr(s, ' ');
2031                if (!endp)
2032                        die("Missing space after source: %s", command_buf.buf);
2033                strbuf_add(&s_uq, s, endp - s);
2034        }
2035        s = s_uq.buf;
2036
2037        endp++;
2038        if (!*endp)
2039                die("Missing dest: %s", command_buf.buf);
2040
2041        d = endp;
2042        strbuf_reset(&d_uq);
2043        if (!unquote_c_style(&d_uq, d, &endp)) {
2044                if (*endp)
2045                        die("Garbage after dest in: %s", command_buf.buf);
2046                d = d_uq.buf;
2047        }
2048
2049        memset(&leaf, 0, sizeof(leaf));
2050        if (rename)
2051                tree_content_remove(&b->branch_tree, s, &leaf);
2052        else
2053                tree_content_get(&b->branch_tree, s, &leaf);
2054        if (!leaf.versions[1].mode)
2055                die("Path %s not in branch", s);
2056        tree_content_set(&b->branch_tree, d,
2057                leaf.versions[1].sha1,
2058                leaf.versions[1].mode,
2059                leaf.tree);
2060}
2061
2062static void note_change_n(struct branch *b)
2063{
2064        const char *p = command_buf.buf + 2;
2065        static struct strbuf uq = STRBUF_INIT;
2066        struct object_entry *oe = oe;
2067        struct branch *s;
2068        unsigned char sha1[20], commit_sha1[20];
2069        uint16_t inline_data = 0;
2070
2071        /* <dataref> or 'inline' */
2072        if (*p == ':') {
2073                char *x;
2074                oe = find_mark(strtoumax(p + 1, &x, 10));
2075                hashcpy(sha1, oe->sha1);
2076                p = x;
2077        } else if (!prefixcmp(p, "inline")) {
2078                inline_data = 1;
2079                p += 6;
2080        } else {
2081                if (get_sha1_hex(p, sha1))
2082                        die("Invalid SHA1: %s", command_buf.buf);
2083                oe = find_object(sha1);
2084                p += 40;
2085        }
2086        if (*p++ != ' ')
2087                die("Missing space after SHA1: %s", command_buf.buf);
2088
2089        /* <committish> */
2090        s = lookup_branch(p);
2091        if (s) {
2092                hashcpy(commit_sha1, s->sha1);
2093        } else if (*p == ':') {
2094                uintmax_t commit_mark = strtoumax(p + 1, NULL, 10);
2095                struct object_entry *commit_oe = find_mark(commit_mark);
2096                if (commit_oe->type != OBJ_COMMIT)
2097                        die("Mark :%" PRIuMAX " not a commit", commit_mark);
2098                hashcpy(commit_sha1, commit_oe->sha1);
2099        } else if (!get_sha1(p, commit_sha1)) {
2100                unsigned long size;
2101                char *buf = read_object_with_reference(commit_sha1,
2102                        commit_type, &size, commit_sha1);
2103                if (!buf || size < 46)
2104                        die("Not a valid commit: %s", p);
2105                free(buf);
2106        } else
2107                die("Invalid ref name or SHA1 expression: %s", p);
2108
2109        if (inline_data) {
2110                static struct strbuf buf = STRBUF_INIT;
2111
2112                if (p != uq.buf) {
2113                        strbuf_addstr(&uq, p);
2114                        p = uq.buf;
2115                }
2116                read_next_command();
2117                parse_data(&buf);
2118                store_object(OBJ_BLOB, &buf, &last_blob, sha1, 0);
2119        } else if (oe) {
2120                if (oe->type != OBJ_BLOB)
2121                        die("Not a blob (actually a %s): %s",
2122                                typename(oe->type), command_buf.buf);
2123        } else {
2124                enum object_type type = sha1_object_info(sha1, NULL);
2125                if (type < 0)
2126                        die("Blob not found: %s", command_buf.buf);
2127                if (type != OBJ_BLOB)
2128                        die("Not a blob (actually a %s): %s",
2129                            typename(type), command_buf.buf);
2130        }
2131
2132        tree_content_set(&b->branch_tree, sha1_to_hex(commit_sha1), sha1,
2133                S_IFREG | 0644, NULL);
2134}
2135
2136static void file_change_deleteall(struct branch *b)
2137{
2138        release_tree_content_recursive(b->branch_tree.tree);
2139        hashclr(b->branch_tree.versions[0].sha1);
2140        hashclr(b->branch_tree.versions[1].sha1);
2141        load_tree(&b->branch_tree);
2142}
2143
2144static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2145{
2146        if (!buf || size < 46)
2147                die("Not a valid commit: %s", sha1_to_hex(b->sha1));
2148        if (memcmp("tree ", buf, 5)
2149                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
2150                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
2151        hashcpy(b->branch_tree.versions[0].sha1,
2152                b->branch_tree.versions[1].sha1);
2153}
2154
2155static void parse_from_existing(struct branch *b)
2156{
2157        if (is_null_sha1(b->sha1)) {
2158                hashclr(b->branch_tree.versions[0].sha1);
2159                hashclr(b->branch_tree.versions[1].sha1);
2160        } else {
2161                unsigned long size;
2162                char *buf;
2163
2164                buf = read_object_with_reference(b->sha1,
2165                        commit_type, &size, b->sha1);
2166                parse_from_commit(b, buf, size);
2167                free(buf);
2168        }
2169}
2170
2171static int parse_from(struct branch *b)
2172{
2173        const char *from;
2174        struct branch *s;
2175
2176        if (prefixcmp(command_buf.buf, "from "))
2177                return 0;
2178
2179        if (b->branch_tree.tree) {
2180                release_tree_content_recursive(b->branch_tree.tree);
2181                b->branch_tree.tree = NULL;
2182        }
2183
2184        from = strchr(command_buf.buf, ' ') + 1;
2185        s = lookup_branch(from);
2186        if (b == s)
2187                die("Can't create a branch from itself: %s", b->name);
2188        else if (s) {
2189                unsigned char *t = s->branch_tree.versions[1].sha1;
2190                hashcpy(b->sha1, s->sha1);
2191                hashcpy(b->branch_tree.versions[0].sha1, t);
2192                hashcpy(b->branch_tree.versions[1].sha1, t);
2193        } else if (*from == ':') {
2194                uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2195                struct object_entry *oe = find_mark(idnum);
2196                if (oe->type != OBJ_COMMIT)
2197                        die("Mark :%" PRIuMAX " not a commit", idnum);
2198                hashcpy(b->sha1, oe->sha1);
2199                if (oe->pack_id != MAX_PACK_ID) {
2200                        unsigned long size;
2201                        char *buf = gfi_unpack_entry(oe, &size);
2202                        parse_from_commit(b, buf, size);
2203                        free(buf);
2204                } else
2205                        parse_from_existing(b);
2206        } else if (!get_sha1(from, b->sha1))
2207                parse_from_existing(b);
2208        else
2209                die("Invalid ref name or SHA1 expression: %s", from);
2210
2211        read_next_command();
2212        return 1;
2213}
2214
2215static struct hash_list *parse_merge(unsigned int *count)
2216{
2217        struct hash_list *list = NULL, *n, *e = e;
2218        const char *from;
2219        struct branch *s;
2220
2221        *count = 0;
2222        while (!prefixcmp(command_buf.buf, "merge ")) {
2223                from = strchr(command_buf.buf, ' ') + 1;
2224                n = xmalloc(sizeof(*n));
2225                s = lookup_branch(from);
2226                if (s)
2227                        hashcpy(n->sha1, s->sha1);
2228                else if (*from == ':') {
2229                        uintmax_t idnum = strtoumax(from + 1, NULL, 10);
2230                        struct object_entry *oe = find_mark(idnum);
2231                        if (oe->type != OBJ_COMMIT)
2232                                die("Mark :%" PRIuMAX " not a commit", idnum);
2233                        hashcpy(n->sha1, oe->sha1);
2234                } else if (!get_sha1(from, n->sha1)) {
2235                        unsigned long size;
2236                        char *buf = read_object_with_reference(n->sha1,
2237                                commit_type, &size, n->sha1);
2238                        if (!buf || size < 46)
2239                                die("Not a valid commit: %s", from);
2240                        free(buf);
2241                } else
2242                        die("Invalid ref name or SHA1 expression: %s", from);
2243
2244                n->next = NULL;
2245                if (list)
2246                        e->next = n;
2247                else
2248                        list = n;
2249                e = n;
2250                (*count)++;
2251                read_next_command();
2252        }
2253        return list;
2254}
2255
2256static void parse_new_commit(void)
2257{
2258        static struct strbuf msg = STRBUF_INIT;
2259        struct branch *b;
2260        char *sp;
2261        char *author = NULL;
2262        char *committer = NULL;
2263        struct hash_list *merge_list = NULL;
2264        unsigned int merge_count;
2265
2266        /* Obtain the branch name from the rest of our command */
2267        sp = strchr(command_buf.buf, ' ') + 1;
2268        b = lookup_branch(sp);
2269        if (!b)
2270                b = new_branch(sp);
2271
2272        read_next_command();
2273        parse_mark();
2274        if (!prefixcmp(command_buf.buf, "author ")) {
2275                author = parse_ident(command_buf.buf + 7);
2276                read_next_command();
2277        }
2278        if (!prefixcmp(command_buf.buf, "committer ")) {
2279                committer = parse_ident(command_buf.buf + 10);
2280                read_next_command();
2281        }
2282        if (!committer)
2283                die("Expected committer but didn't get one");
2284        parse_data(&msg);
2285        read_next_command();
2286        parse_from(b);
2287        merge_list = parse_merge(&merge_count);
2288
2289        /* ensure the branch is active/loaded */
2290        if (!b->branch_tree.tree || !max_active_branches) {
2291                unload_one_branch();
2292                load_branch(b);
2293        }
2294
2295        /* file_change* */
2296        while (command_buf.len > 0) {
2297                if (!prefixcmp(command_buf.buf, "M "))
2298                        file_change_m(b);
2299                else if (!prefixcmp(command_buf.buf, "D "))
2300                        file_change_d(b);
2301                else if (!prefixcmp(command_buf.buf, "R "))
2302                        file_change_cr(b, 1);
2303                else if (!prefixcmp(command_buf.buf, "C "))
2304                        file_change_cr(b, 0);
2305                else if (!prefixcmp(command_buf.buf, "N "))
2306                        note_change_n(b);
2307                else if (!strcmp("deleteall", command_buf.buf))
2308                        file_change_deleteall(b);
2309                else {
2310                        unread_command_buf = 1;
2311                        break;
2312                }
2313                if (read_next_command() == EOF)
2314                        break;
2315        }
2316
2317        /* build the tree and the commit */
2318        store_tree(&b->branch_tree);
2319        hashcpy(b->branch_tree.versions[0].sha1,
2320                b->branch_tree.versions[1].sha1);
2321
2322        strbuf_reset(&new_data);
2323        strbuf_addf(&new_data, "tree %s\n",
2324                sha1_to_hex(b->branch_tree.versions[1].sha1));
2325        if (!is_null_sha1(b->sha1))
2326                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(b->sha1));
2327        while (merge_list) {
2328                struct hash_list *next = merge_list->next;
2329                strbuf_addf(&new_data, "parent %s\n", sha1_to_hex(merge_list->sha1));
2330                free(merge_list);
2331                merge_list = next;
2332        }
2333        strbuf_addf(&new_data,
2334                "author %s\n"
2335                "committer %s\n"
2336                "\n",
2337                author ? author : committer, committer);
2338        strbuf_addbuf(&new_data, &msg);
2339        free(author);
2340        free(committer);
2341
2342        if (!store_object(OBJ_COMMIT, &new_data, NULL, b->sha1, next_mark))
2343                b->pack_id = pack_id;
2344        b->last_commit = object_count_by_type[OBJ_COMMIT];
2345}
2346
2347static void parse_new_tag(void)
2348{
2349        static struct strbuf msg = STRBUF_INIT;
2350        char *sp;
2351        const char *from;
2352        char *tagger;
2353        struct branch *s;
2354        struct tag *t;
2355        uintmax_t from_mark = 0;
2356        unsigned char sha1[20];
2357
2358        /* Obtain the new tag name from the rest of our command */
2359        sp = strchr(command_buf.buf, ' ') + 1;
2360        t = pool_alloc(sizeof(struct tag));
2361        t->next_tag = NULL;
2362        t->name = pool_strdup(sp);
2363        if (last_tag)
2364                last_tag->next_tag = t;
2365        else
2366                first_tag = t;
2367        last_tag = t;
2368        read_next_command();
2369
2370        /* from ... */
2371        if (prefixcmp(command_buf.buf, "from "))
2372                die("Expected from command, got %s", command_buf.buf);
2373        from = strchr(command_buf.buf, ' ') + 1;
2374        s = lookup_branch(from);
2375        if (s) {
2376                hashcpy(sha1, s->sha1);
2377        } else if (*from == ':') {
2378                struct object_entry *oe;
2379                from_mark = strtoumax(from + 1, NULL, 10);
2380                oe = find_mark(from_mark);
2381                if (oe->type != OBJ_COMMIT)
2382                        die("Mark :%" PRIuMAX " not a commit", from_mark);
2383                hashcpy(sha1, oe->sha1);
2384        } else if (!get_sha1(from, sha1)) {
2385                unsigned long size;
2386                char *buf;
2387
2388                buf = read_object_with_reference(sha1,
2389                        commit_type, &size, sha1);
2390                if (!buf || size < 46)
2391                        die("Not a valid commit: %s", from);
2392                free(buf);
2393        } else
2394                die("Invalid ref name or SHA1 expression: %s", from);
2395        read_next_command();
2396
2397        /* tagger ... */
2398        if (!prefixcmp(command_buf.buf, "tagger ")) {
2399                tagger = parse_ident(command_buf.buf + 7);
2400                read_next_command();
2401        } else
2402                tagger = NULL;
2403
2404        /* tag payload/message */
2405        parse_data(&msg);
2406
2407        /* build the tag object */
2408        strbuf_reset(&new_data);
2409
2410        strbuf_addf(&new_data,
2411                    "object %s\n"
2412                    "type %s\n"
2413                    "tag %s\n",
2414                    sha1_to_hex(sha1), commit_type, t->name);
2415        if (tagger)
2416                strbuf_addf(&new_data,
2417                            "tagger %s\n", tagger);
2418        strbuf_addch(&new_data, '\n');
2419        strbuf_addbuf(&new_data, &msg);
2420        free(tagger);
2421
2422        if (store_object(OBJ_TAG, &new_data, NULL, t->sha1, 0))
2423                t->pack_id = MAX_PACK_ID;
2424        else
2425                t->pack_id = pack_id;
2426}
2427
2428static void parse_reset_branch(void)
2429{
2430        struct branch *b;
2431        char *sp;
2432
2433        /* Obtain the branch name from the rest of our command */
2434        sp = strchr(command_buf.buf, ' ') + 1;
2435        b = lookup_branch(sp);
2436        if (b) {
2437                hashclr(b->sha1);
2438                hashclr(b->branch_tree.versions[0].sha1);
2439                hashclr(b->branch_tree.versions[1].sha1);
2440                if (b->branch_tree.tree) {
2441                        release_tree_content_recursive(b->branch_tree.tree);
2442                        b->branch_tree.tree = NULL;
2443                }
2444        }
2445        else
2446                b = new_branch(sp);
2447        read_next_command();
2448        parse_from(b);
2449        if (command_buf.len > 0)
2450                unread_command_buf = 1;
2451}
2452
2453static void parse_checkpoint(void)
2454{
2455        if (object_count) {
2456                cycle_packfile();
2457                dump_branches();
2458                dump_tags();
2459                dump_marks();
2460        }
2461        skip_optional_lf();
2462}
2463
2464static void parse_progress(void)
2465{
2466        fwrite(command_buf.buf, 1, command_buf.len, stdout);
2467        fputc('\n', stdout);
2468        fflush(stdout);
2469        skip_optional_lf();
2470}
2471
2472static void option_import_marks(const char *marks)
2473{
2474        import_marks_file = xstrdup(marks);
2475}
2476
2477static void option_date_format(const char *fmt)
2478{
2479        if (!strcmp(fmt, "raw"))
2480                whenspec = WHENSPEC_RAW;
2481        else if (!strcmp(fmt, "rfc2822"))
2482                whenspec = WHENSPEC_RFC2822;
2483        else if (!strcmp(fmt, "now"))
2484                whenspec = WHENSPEC_NOW;
2485        else
2486                die("unknown --date-format argument %s", fmt);
2487}
2488
2489static void option_max_pack_size(const char *packsize)
2490{
2491        max_packsize = strtoumax(packsize, NULL, 0) * 1024 * 1024;
2492}
2493
2494static void option_depth(const char *depth)
2495{
2496        max_depth = strtoul(depth, NULL, 0);
2497        if (max_depth > MAX_DEPTH)
2498                die("--depth cannot exceed %u", MAX_DEPTH);
2499}
2500
2501static void option_active_branches(const char *branches)
2502{
2503        max_active_branches = strtoul(branches, NULL, 0);
2504}
2505
2506static void option_export_marks(const char *marks)
2507{
2508        export_marks_file = xstrdup(marks);
2509}
2510
2511static void option_export_pack_edges(const char *edges)
2512{
2513        if (pack_edges)
2514                fclose(pack_edges);
2515        pack_edges = fopen(edges, "a");
2516        if (!pack_edges)
2517                die_errno("Cannot open '%s'", edges);
2518}
2519
2520static int parse_one_option(const char *option)
2521{
2522        if (!prefixcmp(option, "max-pack-size=")) {
2523                option_max_pack_size(option + 14);
2524        } else if (!prefixcmp(option, "depth=")) {
2525                option_depth(option + 6);
2526        } else if (!prefixcmp(option, "active-branches=")) {
2527                option_active_branches(option + 16);
2528        } else if (!prefixcmp(option, "export-pack-edges=")) {
2529                option_export_pack_edges(option + 18);
2530        } else if (!prefixcmp(option, "quiet")) {
2531                show_stats = 0;
2532        } else if (!prefixcmp(option, "stats")) {
2533                show_stats = 1;
2534        } else {
2535                return 0;
2536        }
2537
2538        return 1;
2539}
2540
2541static int parse_one_feature(const char *feature)
2542{
2543        if (!prefixcmp(feature, "date-format=")) {
2544                option_date_format(feature + 12);
2545        } else if (!prefixcmp(feature, "import-marks=")) {
2546                option_import_marks(feature + 13);
2547        } else if (!prefixcmp(feature, "export-marks=")) {
2548                option_export_marks(feature + 13);
2549        } else if (!prefixcmp(feature, "force")) {
2550                force_update = 1;
2551        } else {
2552                return 0;
2553        }
2554
2555        return 1;
2556}
2557
2558static void parse_feature(void)
2559{
2560        char *feature = command_buf.buf + 8;
2561
2562        if (seen_data_command)
2563                die("Got feature command '%s' after data command", feature);
2564
2565        if (parse_one_feature(feature))
2566                return;
2567
2568        die("This version of fast-import does not support feature %s.", feature);
2569}
2570
2571static void parse_option(void)
2572{
2573        char *option = command_buf.buf + 11;
2574
2575        if (seen_data_command)
2576                die("Got option command '%s' after data command", option);
2577
2578        if (parse_one_option(option))
2579                return;
2580
2581        die("This version of fast-import does not support option: %s", option);
2582}
2583
2584static int git_pack_config(const char *k, const char *v, void *cb)
2585{
2586        if (!strcmp(k, "pack.depth")) {
2587                max_depth = git_config_int(k, v);
2588                if (max_depth > MAX_DEPTH)
2589                        max_depth = MAX_DEPTH;
2590                return 0;
2591        }
2592        if (!strcmp(k, "pack.compression")) {
2593                int level = git_config_int(k, v);
2594                if (level == -1)
2595                        level = Z_DEFAULT_COMPRESSION;
2596                else if (level < 0 || level > Z_BEST_COMPRESSION)
2597                        die("bad pack compression level %d", level);
2598                pack_compression_level = level;
2599                pack_compression_seen = 1;
2600                return 0;
2601        }
2602        return git_default_config(k, v, cb);
2603}
2604
2605static const char fast_import_usage[] =
2606"git fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2607
2608static void parse_argv(void)
2609{
2610        unsigned int i;
2611
2612        for (i = 1; i < global_argc; i++) {
2613                const char *a = global_argv[i];
2614
2615                if (*a != '-' || !strcmp(a, "--"))
2616                        break;
2617
2618                if (parse_one_option(a + 2))
2619                        continue;
2620
2621                if (parse_one_feature(a + 2))
2622                        continue;
2623
2624                die("unknown option %s", a);
2625        }
2626        if (i != global_argc)
2627                usage(fast_import_usage);
2628
2629        seen_data_command = 1;
2630        if (import_marks_file)
2631                read_marks();
2632}
2633
2634int main(int argc, const char **argv)
2635{
2636        unsigned int i;
2637
2638        git_extract_argv0_path(argv[0]);
2639
2640        if (argc == 2 && !strcmp(argv[1], "-h"))
2641                usage(fast_import_usage);
2642
2643        setup_git_directory();
2644        git_config(git_pack_config, NULL);
2645        if (!pack_compression_seen && core_compression_seen)
2646                pack_compression_level = core_compression_level;
2647
2648        alloc_objects(object_entry_alloc);
2649        strbuf_init(&command_buf, 0);
2650        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2651        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2652        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2653        marks = pool_calloc(1, sizeof(struct mark_set));
2654
2655        global_argc = argc;
2656        global_argv = argv;
2657
2658        rc_free = pool_alloc(cmd_save * sizeof(*rc_free));
2659        for (i = 0; i < (cmd_save - 1); i++)
2660                rc_free[i].next = &rc_free[i + 1];
2661        rc_free[cmd_save - 1].next = NULL;
2662
2663        prepare_packed_git();
2664        start_packfile();
2665        set_die_routine(die_nicely);
2666        while (read_next_command() != EOF) {
2667                if (!strcmp("blob", command_buf.buf))
2668                        parse_new_blob();
2669                else if (!prefixcmp(command_buf.buf, "commit "))
2670                        parse_new_commit();
2671                else if (!prefixcmp(command_buf.buf, "tag "))
2672                        parse_new_tag();
2673                else if (!prefixcmp(command_buf.buf, "reset "))
2674                        parse_reset_branch();
2675                else if (!strcmp("checkpoint", command_buf.buf))
2676                        parse_checkpoint();
2677                else if (!prefixcmp(command_buf.buf, "progress "))
2678                        parse_progress();
2679                else if (!prefixcmp(command_buf.buf, "feature "))
2680                        parse_feature();
2681                else if (!prefixcmp(command_buf.buf, "option git "))
2682                        parse_option();
2683                else if (!prefixcmp(command_buf.buf, "option "))
2684                        /* ignore non-git options*/;
2685                else
2686                        die("Unsupported command: %s", command_buf.buf);
2687        }
2688
2689        /* argv hasn't been parsed yet, do so */
2690        if (!seen_data_command)
2691                parse_argv();
2692
2693        end_packfile();
2694
2695        dump_branches();
2696        dump_tags();
2697        unkeep_all_packs();
2698        dump_marks();
2699
2700        if (pack_edges)
2701                fclose(pack_edges);
2702
2703        if (show_stats) {
2704                uintmax_t total_count = 0, duplicate_count = 0;
2705                for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2706                        total_count += object_count_by_type[i];
2707                for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2708                        duplicate_count += duplicate_count_by_type[i];
2709
2710                fprintf(stderr, "%s statistics:\n", argv[0]);
2711                fprintf(stderr, "---------------------------------------------------------------------\n");
2712                fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2713                fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2714                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]);
2715                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]);
2716                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]);
2717                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]);
2718                fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2719                fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2720                fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2721                fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2722                fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2723                fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2724                fprintf(stderr, "---------------------------------------------------------------------\n");
2725                pack_report();
2726                fprintf(stderr, "---------------------------------------------------------------------\n");
2727                fprintf(stderr, "\n");
2728        }
2729
2730        return failure ? 1 : 0;
2731}