emkornfield commented on code in PR #12158: URL: https://github.com/apache/arrow/pull/12158#discussion_r853669960
########## go/arrow/array/dictionary.go: ########## @@ -0,0 +1,1305 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package array + +import ( + "errors" + "fmt" + "math" + "sync/atomic" + "unsafe" + + "github.com/apache/arrow/go/v7/arrow" + "github.com/apache/arrow/go/v7/arrow/bitutil" + "github.com/apache/arrow/go/v7/arrow/decimal128" + "github.com/apache/arrow/go/v7/arrow/float16" + "github.com/apache/arrow/go/v7/arrow/internal/debug" + "github.com/apache/arrow/go/v7/arrow/memory" + "github.com/apache/arrow/go/v7/internal/hashing" + "github.com/goccy/go-json" +) + +// Dictionary represents the type for dictionary-encoded data with a data +// dependent dictionary. +// +// A dictionary array contains an array of non-negative integers (the "dictionary" +// indices") along with a data type containing a "dictionary" corresponding to +// the distinct values represented in the data. +// +// For example, the array: +// +// ["foo", "bar", "foo", "bar", "foo", "bar"] +// +// with dictionary ["bar", "foo"], would have the representation of: +// +// indices: [1, 0, 1, 0, 1, 0] +// dictionary: ["bar", "foo"] +// +// The indices in principle may be any integer type. +type Dictionary struct { + array + + indices Interface + dict Interface +} + +// NewDictionaryArray constructs a dictionary array with the provided indices +// and dictionary using the given type. +func NewDictionaryArray(typ arrow.DataType, indices, dict Interface) *Dictionary { + a := &Dictionary{} + a.array.refCount = 1 + dictdata := NewData(typ, indices.Len(), indices.Data().Buffers(), indices.Data().Children(), indices.NullN(), indices.Data().Offset()) + dictdata.dictionary = dict.Data().(*Data) + dict.Data().Retain() + + defer dictdata.Release() + a.setData(dictdata) + return a +} + +// checkIndexBounds returns an error if any value in the provided integer +// arraydata is >= the passed upperlimit or < 0. otherwise nil +func checkIndexBounds(indices *Data, upperlimit uint64) error { + if indices.length == 0 { + return nil + } + + var maxval uint64 + switch indices.dtype.ID() { + case arrow.UINT8: + maxval = math.MaxUint8 + case arrow.UINT16: + maxval = math.MaxUint16 + case arrow.UINT32: + maxval = math.MaxUint32 + case arrow.UINT64: + maxval = math.MaxUint64 + } + isSigned := maxval == 0 + if !isSigned && upperlimit > maxval { + return nil + } + + // TODO(mtopol): lift BitSetRunReader from parquet to utils + // and use it here for performance improvement. + var nullbitmap []byte + if indices.buffers[0] != nil { + nullbitmap = indices.buffers[0].Bytes() + } + + var outOfBounds func(i int) error + switch indices.dtype.ID() { + case arrow.INT8: + data := arrow.Int8Traits.CastFromBytes(indices.buffers[1].Bytes()) + outOfBounds = func(i int) error { + if data[i] < 0 || data[i] >= int8(upperlimit) { + return fmt.Errorf("index %d out of bounds", data[i]) + } + return nil + } + case arrow.UINT8: + data := arrow.Uint8Traits.CastFromBytes(indices.buffers[1].Bytes()) + outOfBounds = func(i int) error { + if data[i] >= uint8(upperlimit) { + return fmt.Errorf("index %d out of bounds", data[i]) + } + return nil + } + case arrow.INT16: + data := arrow.Int16Traits.CastFromBytes(indices.buffers[1].Bytes()) + outOfBounds = func(i int) error { + if data[i] < 0 || data[i] >= int16(upperlimit) { + return fmt.Errorf("index %d out of bounds", data[i]) + } + return nil + } + case arrow.UINT16: + data := arrow.Uint16Traits.CastFromBytes(indices.buffers[1].Bytes()) + outOfBounds = func(i int) error { + if data[i] >= uint16(upperlimit) { + return fmt.Errorf("index %d out of bounds", data[i]) + } + return nil + } + case arrow.INT32: + data := arrow.Int32Traits.CastFromBytes(indices.buffers[1].Bytes()) + outOfBounds = func(i int) error { + if data[i] < 0 || data[i] >= int32(upperlimit) { + return fmt.Errorf("index %d out of bounds", data[i]) + } + return nil + } + case arrow.UINT32: + data := arrow.Uint32Traits.CastFromBytes(indices.buffers[1].Bytes()) + outOfBounds = func(i int) error { + if data[i] >= uint32(upperlimit) { + return fmt.Errorf("index %d out of bounds", data[i]) + } + return nil + } + case arrow.INT64: + data := arrow.Int64Traits.CastFromBytes(indices.buffers[1].Bytes()) + outOfBounds = func(i int) error { + if data[i] < 0 || data[i] >= int64(upperlimit) { + return fmt.Errorf("index %d out of bounds", data[i]) + } + return nil + } + case arrow.UINT64: + data := arrow.Uint64Traits.CastFromBytes(indices.buffers[1].Bytes()) + outOfBounds = func(i int) error { + if data[i] >= upperlimit { + return fmt.Errorf("index %d out of bounds", data[i]) + } + return nil + } + default: + return fmt.Errorf("invalid type for bounds checking: %T", indices.dtype) + } + + for i := 0; i < indices.length; i++ { + if len(nullbitmap) > 0 && bitutil.BitIsNotSet(nullbitmap, i+indices.offset) { + continue + } + + if err := outOfBounds(i + indices.offset); err != nil { + return err + } + } + return nil +} + +// NewValidatedDictionaryArray constructs a dictionary array from the provided indices +// and dictionary arrays, while also performing validation checks to ensure correctness +// such as bounds checking at are usually skipped for performance. +func NewValidatedDictionaryArray(typ *arrow.DictionaryType, indices, dict Interface) (*Dictionary, error) { + if indices.DataType().ID() != typ.IndexType.ID() { + return nil, fmt.Errorf("dictionary type index (%T) does not match indices array type (%T)", typ.IndexType, indices.DataType()) + } + + if !arrow.TypeEqual(typ.ValueType, dict.DataType()) { + return nil, fmt.Errorf("dictionary value type (%T) does not match dict array type (%T)", typ.ValueType, dict.DataType()) + } + + if err := checkIndexBounds(indices.Data().(*Data), uint64(dict.Len())); err != nil { + return nil, err + } + + return NewDictionaryArray(typ, indices, dict), nil +} + +// NewDictionaryData creates a strongly typed Dictionary array from +// an ArrayData object with a datatype of arrow.Dictionary and a dictionary +func NewDictionaryData(data arrow.ArrayData) *Dictionary { + a := &Dictionary{} + a.refCount = 1 + a.setData(data.(*Data)) + return a +} + +func (d *Dictionary) Retain() { + atomic.AddInt64(&d.refCount, 1) +} + +func (d *Dictionary) Release() { + debug.Assert(atomic.LoadInt64(&d.refCount) > 0, "too many releases") + + if atomic.AddInt64(&d.refCount, -1) == 0 { + d.data.Release() + d.data, d.nullBitmapBytes = nil, nil + d.indices.Release() + d.indices = nil + if d.dict != nil { + d.dict.Release() + d.dict = nil + } + } +} + +func (d *Dictionary) setData(data *Data) { + d.array.setData(data) + + if data.dictionary == nil { + panic("arrow/array: no dictionary set in Data for Dictionary array") + } + + dictType := data.dtype.(*arrow.DictionaryType) + debug.Assert(arrow.TypeEqual(dictType.ValueType, data.dictionary.DataType()), "mismatched dictionary value types") + + indexData := NewData(dictType.IndexType, data.length, data.buffers, data.childData, data.nulls, data.offset) + defer indexData.Release() + d.indices = MakeFromData(indexData) +} + +// Dictionary returns the values array that makes up the dictionary for this +// array. +func (d *Dictionary) Dictionary() Interface { + if d.dict == nil { + d.dict = MakeFromData(d.data.dictionary) + } + return d.dict +} + +// Indices returns the underlying array of indices as it's own array +func (d *Dictionary) Indices() Interface { + return d.indices +} + +// CanCompareIndices returns true if the dictionary arrays can be compared +// without having to unify the dictionaries themselves first. +func (d *Dictionary) CanCompareIndices(other *Dictionary) bool { + if !arrow.TypeEqual(d.indices.DataType(), other.indices.DataType()) { + return false + } + + minlen := int64(min(d.data.dictionary.length, other.data.dictionary.length)) + return ArraySliceEqual(d.Dictionary(), 0, minlen, other.Dictionary(), 0, minlen) +} + +func (d *Dictionary) String() string { + return fmt.Sprintf("{ dictionary: %v\n indices: %v }", d.Dictionary(), d.Indices()) +} + +// GetValueIndex returns the dictionary index for the value at index i of the array. +// The actual value can be retrieved by using d.Dictionary().(valuetype).Value(d.GetValueIndex(i)) +func (d *Dictionary) GetValueIndex(i int) int { + indiceData := d.data.buffers[1].Bytes() + // we know the value is non-negative per the spec, so + // we can use the unsigned value regardless. + switch d.indices.DataType().ID() { + case arrow.UINT8, arrow.INT8: + return int(uint8(indiceData[d.data.offset+i])) + case arrow.UINT16, arrow.INT16: + return int(arrow.Uint16Traits.CastFromBytes(indiceData)[d.data.offset+i]) + case arrow.UINT32, arrow.INT32: + return int(arrow.Uint32Traits.CastFromBytes(indiceData)[d.data.offset+i]) + case arrow.UINT64, arrow.INT64: + return int(arrow.Uint64Traits.CastFromBytes(indiceData)[d.data.offset+i]) Review Comment: maybe just panic for truncation? like I said i think this is probably unlikely. If you want to go through the effort int64 seems sufficient. most architectures I'm aware of can't even address a full 64 bits of ram. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
