com.bigdata.rdf.internal.package.html Maven / Gradle / Ivy
RDF Value Internals
This package provides an internal representation of RDF Values. The internal
representation of an RDF Value is either a term identifier or an inline value.
Term identifiers are assigned by consistent writes on the TERM2ID index and
the ID2TERM index (which maps the term identifier back onto the RDF Value).
Inline values directly code the RDF Value and are generally used for datatype
literals having short values (xsd:boolean, xsd:char) or numeric values (xsd:short,
xsd:int, xsd:float, xsd:long, xsd:double).
The unsigned byte[] keys for an RDF Value are formed by appending some bit flags
which partition the key space into URIs, blank nodes, literals, and statement
identifiers, a bit flag indicating whether the value is inline, followed by either
pad bits and the term identifier or a code which identifies the data type of the
inline value and minimum length decodable representation of the data type value
formed using {@link com.bigdata.btree.keys.IKeyBuilder}. This design clusters
different kinds of RDF values within different regions of the ID2TERM index.
Inline values have a significant performance advantage since the RDF Value object
can be recovered directly from the inline value without indirection through the
ID2TERM index. This reduces the costs to materialize RDF Values, greatly speeds
up aggregation style queries over numeric data, and reduces both the maintenance
time and the size on the disk for the lexicon indices since inline values are not
entered into the lexicon.
Inline values also make it possible to translate a LT/GT style filter against
an inlined datatype into key-range queries against the OSP(C) index. Since the
inline values for different numeric data types are located in different parts of
the key space, the use of a cast in the query will require either than key-range
queries are issued against all numeric data types (UNION of access paths) or
that the query is evaluated without the use of key-range scans on OSP(C).
RDF Values which are inlined obey the equality and order semantics of their
value space (think xsd:int). One consequence of this is that comparison of
inlined datatype literals MUST occur in the value space. E.g., 005
and 5 represent the same point in the xsd:int value space. Clients
should be aware of this distinction as statements which have lexical distinctions
which are not distinct in the value space will be mapped onto the same statement
internally.