Package org.apache.drill.exec.vector.accessor.reader

Provides the reader hierarchy as explained in the API package.

See: Description

Interface Summary

Interface	Description
BaseScalarReader.BufferAccessor	Provide access to the DrillBuf for the data vector.
ElementReaderIndex
NullStateReader	Internal mechanism to detect if a value is null.
ReaderEvents	Internal operations to wire up a set of readers.
VectorAccessor

Class Summary

Class	Description
AbstractObjectReader
AbstractScalarReader
AbstractScalarReader.NullReader
AbstractScalarReader.ScalarObjectReader
AbstractTupleReader	Reader for a tuple (a row or a map.) Provides access to each column using either a name or a numeric index.
AbstractTupleReader.TupleObjectReader
ArrayReaderImpl	Reader for an array-valued column.
ArrayReaderImpl.ArrayObjectReader	Object representation of an array reader.
ArrayReaderImpl.ElementReaderIndex	Index into the vector of elements for a repeated vector.
BaseScalarReader	Column reader implementation that acts as the basis for the generated, vector-specific implementations.
BaseScalarReader.BaseFixedWidthReader
BaseScalarReader.BaseVarWidthReader
BitColumnReader	Specialized reader for bit columns.
ColumnReaderFactory	Gather generated reader classes into a set of class tables to allow rapid run-time creation of readers.
DictEntryReader	Reader for a Dict entry.
DictReaderImpl
DictReaderImpl.DictObjectReader
MapReader	Reader for a Drill Map type.
NullStateReaders
NullStateReaders.AlwaysNullStateReader
NullStateReaders.ComplexMemberStateReader	Handle the awkward situation with complex types.
NullStateReaders.ListIsSetVectorStateReader	Holder for the NullableVector wrapper around a bits vector and a data vector.
NullStateReaders.MemberNullStateReader	Null state that handles the strange union semantics that both the union and the values can be null.
NullStateReaders.NullableIsSetVectorStateReader	Holder for the NullableVector wrapper around a bits vector and a data vector.
NullStateReaders.RequiredStateReader	Dummy implementation of a null state reader for cases in which the value is never null.
NullStateReaders.TypeVectorStateReader	Extract null state from the union vector's type vector.
OffsetVectorReader	Reader for an offset vector.
UnionReaderImpl	Reader for a union vector.
UnionReaderImpl.UnionObjectReader
VectorAccessors	Collection of vector accessors.
VectorAccessors.ArrayDataHyperVectorAccessor	Vector accessor for RepeatedVector → data vector
VectorAccessors.ArrayOffsetHyperVectorAccessor	Vector accessor for RepeatedVector → offsets vector
VectorAccessors.BaseHyperVectorAccessor	Vector accessor used by the column accessors to obtain the vector for each column value.
VectorAccessors.ListBitsHyperVectorStateReader	Vector accessor for ListVector → bits vector
VectorAccessors.ListMemberHyperVectorAccessor
VectorAccessors.MapMemberHyperVectorAccessor	Vector accessor for AbstractMapVector → member vector
VectorAccessors.NullableBitsHyperVectorStateReader	Vector accessor for NullableVector → bits vector
VectorAccessors.NullableValuesHyperVectorAccessor	Vector accessor for NullableVector → values vector
VectorAccessors.NullVectorAccesor
VectorAccessors.SingleVectorAccessor
VectorAccessors.UnionMemberHyperVectorAccessor	Vector accessor for UnionVector → data vector
VectorAccessors.UnionTypeHyperVectorAccessor	Vector accessor for UnionVector → type vector
VectorAccessors.VarWidthOffsetHyperVectorAccessor	Vector accessor for VariableWidthVector → offsets vector

Package org.apache.drill.exec.vector.accessor.reader Description

Provides the reader hierarchy as explained in the API package.

Structure

The reader implementation divides into four parts:

1. The readers themselves which start with scalar readers to decode data from vectors, then build up to nullable, array, union and list readers. Readers are built up via composition, often using the (internal) offset vector reader.
2. The column index abstraction that steps through items in a collection. At the top level, the index points to the current row. The top level may include an indirection (an SV2 or SV4) which is handled by the column index. Within arrays, the column index points to each element of the array.
3. The vector accessor which provides a unified interface for both the single-batch and hyper-batch cases. The single-batch versions simply hold onto the vector itself. The hyper-batch versions either provide access to a specific vector within a hyper-vector (for top-level vectors), or navigate from a top-level vector down to an inner vector (for nested vectors.)
4. The null state abstraction which provides a uniform way to detect nullability. For example, within the reader system, the reader for nullable and required vectors differ only in the associated null state reader. Unions and lists have complex null state logic: the nullability of a value depends on the nullability of the list, the union, and the value itself. The null state class implements this logic independent of the reader structure.

Composition

The result is that reader structure makes heavy use of composition: readers are built up from each of the above components. The number of actual reader classes is small, but the methods to build the readers are complex. Most structure is built at build time. Indexes, however are provided at a later "bind" time at which a bind call traverses the reader tree to associate an index with each reader and vector accessor. When a reader is for an array, the bind step creates the index for the array elements.

Construction

Construction of readers is a multi-part process.
• Start with a single or hyper-vector batch.
• The reader builders in another package parse the batch structure, create the required metadata, wrap the (single or hyper) vectors in a vector accessor, and call methods in this package.
• Methods here perform the final construction based on the specific type of the reader.

The work divides into two main categories:

• The work which is based on the vector structure and single/hyper-vector structure, which is done elsewhere.
• The work which is based on the structure of the readers (with vector cardinality factored out), which is done here.