_build_cmp_and_eq

construct comparison/equality callbacks

my $cmp1 = sub 
{
    my ($k1, $k2) = @_;

    # NOTE: use "spaceship" (-1,0,1) comparison with 
    # short-circuit OR (which returns 0 or VALUE, not 0 or 1) 
    # to perform multi-column key comparison 
    # a la Schwartzian Transform

    return (
            (   ($k1->[0] <=> $k2->[0])
             || ($k1->[1] <=> $k2->[1])) == -1
            );
};

my $eq1 = sub 
{
    my ($k1, $k2) = @_;
    return (($k1->[0] == $k2->[0]) 
            && ($k1->[1] == $k2->[1]) 
            );
};

NAME

Genezzo::Index::bt2 - basic btree

A btree built of row directory blocks.

SYNOPSIS

use Genezzo::Index::bt?;

my $tt = Genezzo::Index::btree->new();

$tt->insert(1, "hi");
$tt->insert(7, "there");

DESCRIPTION

This btree algorithm is a bottom-up implementation based upon ideas from Chapter 16 of "Algorithms in C++ (third edition)", by Robert Sedgewick, 1998 and Chapter 15, "Access Paths", of "Transaction Processing: Concepts and Techniques" by Jim Gray and Andreas Reuter, 1993. The pedagogical examples use a fixed number of entries per node, or fixed-size keys in each block, but this implementation has significant extensions to support variable numbers of variably-sized keys in fixed-size disk blocks, with the associated error handling, plus support for reverse scans.

FUNCTIONS

This package supports a constructor "new", plus standard b-tree methods like insert, delete, search.

"new" constructor

The "new" constructor takes many arguments, but they are all optional. If none are specified, the constructor will allocate 100 blocks of the default size for a b-tree. The default assumption is to support scalar string keys with a scalar string values. The tree will have a maximum of 50 entries per node.

maxsize (default 50)

The maximum number of entries in a node. If set to zero, the insert will pack as many entries as space allows in each node

numblocks (default 100)

The constructor will allocate a private buffer cache for the b-tree of up to the number of blocks specified. If numblocks=0, no cache is created. In this case, the user must create a subclass to overload the make_new_block and getarr methods.

blocksize (default DEFBLOCKSIZE)

The size of each block in the b-tree

key_type (null by default)

The key type is either a single scalar "c" (for char) or "n" (for number), or a ref to an array of "c" and/or "n" values. If key_type is specified, bt2 finds or constructs the appropriate compare/equals and pack/unpack functions, overriding any user-supplied arguments. If key type is not specified, bt2 processes the insert keys as a scalar strings.

compare, equal (default string comparison -- ignored if key_type argument specified)

Supply methods to compare your key. This package contains special comparison methods for numeric and multi-column keys, and their associated packing functions.

pack_fn/unpack_fn (default single scalar key and value -- ignored if key_type specified)

"Packing" functions convert key/value pairs to and from a byte representation which gets stored in the nodes of the b-tree. The b-tree package supports scalar keys and values by default. It also contains methods for multi-column keys with a single value.

use_IOT (default off)

special flag for Index Organized Tables, which means the "value" can be an array, not a scalar. This approach requires a couple extra checks in the branch nodes, since branches contain (key, nodeid) pairs, and leaves contain (key, array of values). Normally, indexes only have a scalar value: a nodeid or a rid.

unique_key (default off)

Enforce uniqueness (no duplicates) at insertion time

use_keycount (default off)

Special case for building non-unique indexes where the "value" is null because it is already part of the key vector. In this usage, we construct a unique index (unique_key=1) where the key vector is the key columns *plus* the table rid, and the value is null. The key columns might be duplicates, but the addition of the rid guarantees uniqueness. The fetch is asymmetric: the table rid is returned as both the last key column and the value.

Q: Why not just have a non-unique index and store the rids as regular values? A: This approach clusters related rids, so index scans are more efficient and deletes are easier. Note that the basic index row physical storage is unaffected. Only the unpack function needs an extra argument to describe the number of key columns. Q: But doesn't the extra comparison for the rid column make inserts more expensive? A: Yes, but we're trading off insert performance against index scan performance. The workload of most database applications is typically dominated by selects, not inserts.

functions

insert
delete
btCLEAR
hash_key/array_offset iterators: FIRSTKEY, NEXTKEY, FETCH, plus reverse iterators LASTKEY, PREVKEY.
DBI-style search interface: SQLPrepare, Execute, Fetch

EXPORT

none

TODO

hkey/offset functions: should be able to convert between different "place" formats (Array and Hash prefixes), like the common fetch routine, or ASSERT that prefix matches.
add reverse scan to search/SQLFetch
support multicol keys, non-unique keys (via combo of key + rid as unique)
support transaction unique constraints -- probably via treat key+rid as unique, then turn on true unique key, and scan for duplicates?
find out why can't do pctfree=0
Work on RDBlk_NN support.
search with startkey/stopkey support, vs supplying compare/equal methods. restricting the search api to straight "=","<" comparisons means can try the estimation function
need to handle partial startkey/stopkey comparison in searchR/SQLFetch for multi-col keys
semantics of nulls in multi-col keys -- sort low?
simplify _pack_row with splice and a supplied split position, something like -1 for normal indexes (n-1 key cols, 1 val col, so pop the val) or "N=?" for index-organized tables (N key cols, M val cols, so splice N)
reorganize along the lines of "GiST" Generalized Search Trees (Paul Aoki, J. Hellerstein, UCB)
ecount support?

AUTHOR

Jeffrey I. Cohen, jcohen@genezzo.com

SEE ALSO

perl(1).

Copyright (c) 2003, 2004 Jeffrey I Cohen. All rights reserved.

This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

Address bug reports and comments to: jcohen@genezzo.com