Simple persistent storage for stone tag/value objects
Boulder:Store;
my $store=new Boulder::Store('test.db',1);
my $s = new Stone (Name=>'george',
Age=>23,
Sex=>M,
Address=>{
Street=>'29 Rockland drive',
Town=>'Fort Washington',
ZIP=>'77777'
}
);
$store->put($s);
$store->put(new Stone(Name=>'fred',
Age=>30,
Sex=>M,
Address=>{
Street=>'19 Gravel Path',
Town=>'Bedrock',
ZIP=>'12345'},
Phone=>{
Day=>'111-1111',
Eve=>'222-2222'
}
));
$store->put(new Stone(Name=>'andrew',
Age=>18,
Sex=>M));
$store->add_index('Name');
my $stone = $store->get(0);
print "name = ",$stone->Name;
Boulder::Store provides persistent storage for Boulder objects using a simple DB_File implementation. To use it, you need to have Berkeley db installed (also known as libdb), and the Perl DB_File module. See the DB_File package for more details on obtaining Berkeley db if you do not already have it.
Boulder::Store provides an unsophisticated query mechanism which takes advantage of indexes that you specify. Despite its lack of sophistication, the query system is often very helpful.
The new() method creates a new Boulder::Store object and associates it with the database file provided in the first parameter (undef is a valid pathname, in which case all methods work but the data isn't stored). The second parameter should be a true value if you want to open the database for writing. Otherwise it's opened read only. Because the underlying storage implementation is not multi-user, only one process can have the database for writing at a time. A fcntl()-based locking mechanism is used to give a process that has the database opened for writing exclusive access to the database. This also prevents the database from being opened for reading while another process is writing to it (this is a good thing). Multiple simultaneous processes can open the database read only. Physically the data is stored in a human-readable file with the extension \*(L".data\*(R".
The semantics of this call are exactly the same as in Boulder::Stream. Stones are returned in sequential order, starting with the first record. In addition to their built-in tags, each stone returned from this call has an additional tag called \*(L"record_no\*(R". This is the zero-based record number of the stone in the database. Use the reset() method to begin iterating from the beginning of the database. If called in an array context, read_record() returns a list of all stones in the database that contains one or more of the provided tags. This has the same semantics as Boulder::Stream. A stone is appended to the end of the database. If successful, this call returns the record number of the new entry. By providing an optional second parameter, you can control where the stone is entered. A positive numeric index will write the stone into the database at that position. A value of -1 will use the Stone's internal record number (if present) to determine where to place it. This is random access to the database. Provide a record number and this call will return the stone stored at that position. This is a random write to the database. Provide a record number and this call stores the stone at the indicated position, replacing whatever was there before. If no record number is provided, this call will look for the presence of a 'record_no' tag in the stone itself and put it back in that position. This allows you to pull a stone out of the database, modify it, and then put it back in without worrying about its record number. If no record is found in the stone, then the effect is identical to write_record(). The record number of the inserted stone is returned from this call, or -1 if an error occurred. These method calls delete a stone from the database. You can provide either the record number or a stone containing the 'record_no' tag. Warning: if the database is heavily indexed deletes can be time-consuming as it requires the index to be brought back into synch. This returns the length of the database, in records. This resets the database, nullifying any queries in effect, and causing read_record() to begin fetching stones from the first record. This creates a query on the database used for selecting stones in read_record(). The query is an associative array. Three types of keys/value pairs are allowed:
This instructs Boulder::Store to look for stones containing the specified tags in which the tag's value (determined by the Stone index() method) exactly matches the provided value. Example: $db->query('STS.left_primer.length'=>30); Only the non-bracketed forms of the index string are allowed (this is probably a bug...) If the tag path was declared to be an index, then this search will be fast. Otherwise Boulder::Store must iterate over every record in the database.
This instructs Boulder::Store to look for stones in which the provided expression evaluates to true. When the expression is evaluated, the variable $s will be set to the current record's stone. As a shortcut, you can use \*(L"<index.string>\*(R" as shorthand for \*(L"$s->index('index.string')\*(R".
This lets you provide a whole bunch of expressions, and is exactly equivalent to EVAL=>'(expression1) && (expression2) && (expression3)'.
You can mix query types in the parameter provided to query(). For example, here's how to look up all stones in which the sex is male and the age is greater than 30: $db->query('sex'=>'M',EVAL=>'<age> > 30'); When a query is in effect, read_record() returns only Stones that satisfy the query. In an array context, read_record() returns a list of all Stones that satisfy the query. When no more satisfactory Stones are found, read_record() returns undef until a new query is entered or reset() is called.
Declare one or more tag paths to be a part of a fast index. read_record() will take advantage of this record when processing queries. For example: $db->add_index('age','sex','person.pets'); You can add indexes any time you like, when the database is first created or later. There is a trade off: write_record(), put(), and other data-modifying calls will become slower as more indexes are added. The index is stored in an external file with the extension \*(L".index\*(R". An index file is created even if you haven't indexed any tags. Call this if the index gets screwed up (or lost). It rebuilds it from scratch.
Boulder::Store makes heavy use of the flock() call in order to avoid corruption of DB_File databases when multiple processes try to write simultaneously. flock() may not work correctly across \s-1NFS\s0 mounts, particularly on Linux machines that are not running the rpc.lockd daemon. Please confirm that your flock() works across \s-1NFS\s0 before attempting to use Boulder::Store. If the store.t test hangs during testing, this is the likely culprit.
Lincoln D. Stein <[email protected]>, Cold Spring Harbor Laboratory, Cold Spring Harbor, \s-1NY\s0. This module can be used and distributed on the same terms as Perl itself.
Boulder, Boulder::Stream, Stone