}
  db_finalize(&q);
  db_finalize(&q2);
  db_finalize(&q3);

  /* Output the commit records.
  */
  topological_sort_checkins(0);
  db_prepare(&q,
    "SELECT strftime('%%s',mtime), objid, coalesce(ecomment,comment),"
    "       coalesce(euser,user),"
    "       (SELECT value FROM tagxref WHERE rid=objid AND tagid=%d)"
    "  FROM toponode, event"
    " WHERE toponode.tid=event.objid"
    "   AND event.type='ci'"
    "   AND NOT EXISTS (SELECT 1 FROM oldcommit WHERE toponode.tid=rid)"
    " ORDER BY toponode.tseq ASC",
    TAG_BRANCH
  );
  db_prepare(&q2, "INSERT INTO oldcommit VALUES (:rid)");
  while( db_step(&q)==SQLITE_ROW ){
    Stmt q4;
    const char *zSecondsSince1970 = db_column_text(&q, 0);
    int ckinId = db_column_int(&q, 1);
    const char *zComment = db_column_text(&q, 2);
    const char *zUser = db_column_text(&q, 3);
    const char *zBranch = db_column_text(&q, 4);
    char *zMark;

    bag_insert(&vers, ckinId);
    db_bind_int(&q2, ":rid", ckinId);
    db_step(&q2);
    db_reset(&q2);
    if( zBranch==0 || fossil_strcmp(zBranch, "trunk")==0 ){
      zBranch = gexport.zTrunkName;
    }
    zMark = mark_name_from_rid(ckinId, &unused_mark);
    printf("commit refs/heads/");
    print_ref(zBranch);
    printf("\nmark %s\n", zMark);
    free(zMark);
    printf("committer");
    print_person(zUser);

    if( ferror(f)!=0 || fclose(f)!=0 ){
      fossil_fatal("error while writing %s", markfile_out);
    }
  }
  bag_clear(&blobs);
  bag_clear(&vers);
}

/*
** Construct the temporary table toposort as follows:
**
**     CREATE TEMP TABLE toponode(
**        tid INTEGER PRIMARY KEY,   -- Check-in id
**        tseq INT                   -- integer total order on check-ins.
**     );
**
** This table contains all check-ins of the repository in topological
** order.  "Topological order" means that every parent check-in comes
** before all of its children.  Topological order is *almost* the same
** thing as "ORDER BY event.mtime".  Differences only arrise when there
** are timewarps.  In as much as Git hates timewarps, we have to compute
** a correct topological order when doing an export.
**
** Since mtime is a usually already nearly in topological order, the
** algorithm is to start with mtime, then make adjustments as necessary
** for timewarps.  This is not a great algorithm for the general case,
** but it is very fast for the overwhelmingly common case where there
** are few timewarps.
*/
int topological_sort_checkins(int bVerbose){
  int nChange = 0;
  Stmt q1;
  Stmt chng;
  db_multi_exec(
    "CREATE TEMP TABLE toponode(\n"
    "  tid INTEGER PRIMARY KEY,\n"
    "  tseq INT\n"
    ");\n"
    "INSERT INTO toponode(tid,tseq) "
    " SELECT objid, CAST(mtime*8640000 AS int) FROM event WHERE type='ci';\n"
    "CREATE TEMP TABLE topolink(\n"
    "  tparent INT,\n"
    "  tchild INT,\n"
    "  PRIMARY KEY(tparent,tchild)\n"
    ") WITHOUT ROWID;"
    "INSERT INTO topolink(tparent,tchild)"
    "  SELECT pid, cid FROM plink;\n"
    "CREATE INDEX topolink_child ON topolink(tchild);\n"
  );

  /* Find a timewarp instance */
  db_prepare(&q1,
    "SELECT P.tid, P.tseq, C.tid, C.tseq\n"
    "  FROM toponode P, toponode C, topolink X\n"
    " WHERE X.tparent=P.tid\n"
    "   AND X.tchild=C.tid\n"
    "   AND P.tseq>=C.tseq;"
  );

  /* Update the timestamp on :tid to have value :tseq */
  db_prepare(&chng,
    "UPDATE toponode SET tseq=:tseq WHERE tid=:tid"
  );

  while( db_step(&q1)==SQLITE_ROW ){
    int iParent = db_column_int(&q1, 0);
    i64 iParentTime = db_column_int64(&q1, 1);
    int iChild = db_column_int(&q1, 2);
    i64 iChildTime = db_column_int64(&q1, 3);
    nChange++;
    if( nChange>10000 ){
      fossil_fatal("failed to fix all timewarps after 100000 attempts");
    }
    db_reset(&q1);
    db_bind_int64(&chng, ":tid", iChild);
    db_bind_int64(&chng, ":tseq", iParentTime+1);
    db_step(&chng);
    db_reset(&chng);
    if( bVerbose ){
      fossil_print("moving %d from %lld to %lld\n",
                   iChild, iChildTime, iParentTime+1);
    }
  }

  db_finalize(&q1);
  db_finalize(&chng);
  return nChange;
}

/*
** COMMAND: test-topological-sort
**
** Invoke the topological_sort_checkins() interface for testing
** purposes.
*/
void test_topological_sort(void){
  int n;
  db_find_and_open_repository(0, 0);
  n = topological_sort_checkins(1);
  fossil_print("%d reorderings required\n", n);
}