NetFabric.Hyperlinq
contains alternative implementations of many operations found in the System.Linq
namespace:
- Uses value-types to improve performance by making method calls non-virtual and reducing GC collections by not allocating on the heap.
- Adds support for
Span<>
,ReadOnlySpan<>
,Memory<>
andReadOnlyMemory<>
. - Nullable reference type annotations.
- One single NuGet package support for both sync and async enumerables.
This implementation favors performance in detriment of assembly binary size (lots of overloads).
- Fast enumeration
- Reduced heap allocations
- Benchmarks
- Usage
- Documentation
- Supported operations
- References
- Credits
- License
NetFabric.Hyperlinq
can enumerate faster the results of a query than System.Linq
by performing all of the following:
- Merges multiple enumerators into a single one for several more scenarios.
- It does not box value-type enumerators so, calls to the
Current
property and theMoveNext()
method are non-virtual. - All the enumerables returned by operations define a value-type enumerator.
- Whenever possible, the enumerator returned by the public
GetEnumerator()
orGetAsyncEnumerator()
does not implementIDisposable
. This allows theforeach
that enumerates the result to be inlinable. - Operations enumerate the source using the indexer when the source is an array,
ArraySegment<>
,Span<>
,ReadOnlySpan<>
,Memory<>
,ReadOnlyMemory<>
, or implementsIReadOnlyList<>
. The indexer performs fewer operations than the enumerator. Range()
andRepeat()
return enumerables that implementIReadOnlyCollection<>
andICollection<>
.Return()
andSelect()
return enumerables that implementIReadOnlyList<>
andIList<>
.- Use of buffer pools in operations like
Distinct()
,ToArray()
andToList()
. - Elimination of conditional branchs in
Where().Count()
. - Allows the JIT compiler to perform optimizations on array enumeration whenever possible.
- Takes advantage of
EqualityComparer<>.Default
devirtualization whenever possible. ToList()
usesICollection<>.CopyTo()
to add the items to the resultingList<>
. This removes many of the operations performed when adding items one-by-one or using anIEnumerable<>
.
The performance is equivalent when the enumerator is a reference-type. This happens when the enumerable is generated using yield
or when it's cast to one of the BCL enumerable interfaces (IEnumerable
, IEnumerable<>
, IReadOnlyCollection<>
, ICollection<>
, IReadOnlyList<>
, IList<>
, or IAsyncEnumerable<>
). In the case of operation composition, this only affects the first operation. The subsequent operations will have value-type enumerators.
NetFabric.Hyperlinq
allocates as much as possible on the stack. Enumerables and enumerators are defined as value-types. Generics constraints are used for the operation parameters so that the value-types are not boxed.
When the indexers are used, no allocation is performed.
It only allocates on the heap for the following cases:
- Operations that use
ICollection<>.CopyTo()
,ICollection<>.Contains()
, orIList<>.IndexOf()
will box enumerables that are value-types. ToList()
, when applied to collections that implementIReadOnlyCollection<>
but notICollection<>
, allocates an instance of an helper class so thatICollection<>.CopyTo()
can be used.ToArray()
andToList()
allocate their results on the heap. You can use theToArray()
overload that take an buffer pool as parameter so that its result is not managed by the garbage collector.
The repository contains a benchmarks project based on BenchmarkDotNet that compares the performance of the operators for diferent libraries and source types.
The results can be found in the Benchmarks folder.
The names of the benchmarks are structured as follow:
- The library used:
- Linq - the
System.Linq
namespace (includes System.Linq.Async, System.Interactive, and System.Interactive.Async) - StructLinq - StructLinq
- Hyperlinq - NetFabric.Hyperlinq
- Linq - the
- The type of collection used as source:
- Array - an array
- Span - a
Span<>
- Memory - a
Memory<>
- Enumerable - implements
IEnumerable<>
- Collection - implements
IReadOnlyCollection<>
andICollection<>
- List - implements
IReadOnlyList<>
andIList<>
but not an array - AsyncEnumerable - implements
IAsyncEnumerable<>
- The type of enumerator provided by the source:
- Value - the enumerator is a value type
- Reference - the enumerator is a reference type
- The contents of the source:
- Sequential - contains numbers in sequential order
- Random - contains numbers in random order
- How the result of the operation is iterated:
- For - a
for
loop is used to call the indexer - Foreach - a
foreach
loop is used to call the enumerator
- For - a
- Add the
NetFabric.Hyperlinq
NuGet package to your project. - Optionally, also add the
NetFabric.Hyperlinq.Analyzer
NuGet package to your project. It's a Roslyn analyzer that suggests performance improvements on your enumeration source code. No dependencies are added to your assemblies. - Add an
using NetFabric.Hyperlinq
directive to all source code files where you want to useNetFabric.Hyperlinq
. It can coexist withSystem.Linq
andSystem.Linq.Async
directives:
using System;
using System.Linq;
using NetFabric.Hyperlinq; // add this directive
NetFabric.Hyperlinq
includes bindings for collections available in the namespaces:
System
- arrays,ArraySegment<>
,Span<>
,ReadOnlySpan<>
,Memory<>
andReadOnlyMemory<>
System.Collections.Generic
-List<>
,Dictionary<>
,Stack<>
, ...System.Collections.Immutable
-ImmutableArray<>
,ImmutableList<>
,ImmutableStack<>
, ...
For all these collections, once the directive is added, NetFabric.Hyperlinq
will be used automatically:
public static void Example(ReadOnlySpan<int> span)
{
var result = span
.Where(item => item > 2)
.Select(item => item * 2);
foreach(var value in result)
Console.WriteLine(value);
}
NetFabric.Hyperlinq
implements operations (extension methods) for the interfaces:
IValueEnumerable<,>
IValueReadOnlyCollection<,>
IValueReadOnlyList<,>
IAsyncValueEnumerable<,>
These are extensions to the BCL enumerable interfaces and contain a second generic argument for the enumerator type so that it won't be boxed.
If the collection does not implement any of these, to use NetFabric.Hyperlinq
operations, you have to use the conversion methods AsValueEnumerable()
or AsAsyncValueEnumerable()
. (Except for the BCL collections where specific bindings are provided.)
In the following example, AsValueEnumerable()
converts IReadOnlyList<>
to IValueReadOnlyList<>
. The subsequent operations used are the ones implemented in NetFabric.Hyperlinq
:
public static void Example(IReadOnlyList<int> list)
{
var result = list
.AsValueEnumerable()
.Where(item => item > 2)
.Select(item => item * 2);
foreach(var value in result)
Console.WriteLine(value);
}
All value enumeration interfaces derive from IEnumerable<>
or IAsyncEnumerable
so, if an operation is not available in NetFabric.Hyperlinq
, it will automatically drop to the System.Linq
implementation.
OrderByDescending()
is not yet available in Netfabric.Hyperlinq
but can still be used without requiring any conversion:
public static void Example(IReadOnlyList<int> list)
{
var result = list
.AsValueEnumerable()
.Where(item => item > 2)
.OrderByDescending(item => item) // is not yet available in Netfabric.Hyperlinq
.AsValueEnumerable()
.Select(item => item * 2);
foreach(var value in result)
Console.WriteLine(value);
}
To add NetFabric.Hyperlinq
operations after a System.Linq
operation, simply add one more AsValueEnumerable()
or AsAsyncValueEnumerable()
.
In NetFabric.Hyperlinq
, the generation operations like Empty()
, Range()
, Repeat()
and Return()
are static methods implemented in the static class ValueEnumerable
. To use the equivalent operations from NetFabric.Hyperlinq
, simply replace Enumerable
for ValueEnumerable
.
public static void Example(int count)
{
var source = ValueEnumerable
.Range(0, count)
.Select(item => item * 2);
foreach(var value in source)
Console.WriteLine(value);
}
Usually, when returning a query, it's used IEnumerable<>
as the method return type:
public static IEnumerable<int> Example(int count)
=> ValueEnumerable
.Range(0, count)
.Select(value => value * 2);
This allows the caller to use System.Linq
or a foreach
loop to pull and process the result. NetFabric.Hyperlinq
can also be used but the AsValueEnumerable()
conversion method has to be used.
The operation in NetFabric.Hyperlinq
are implemented so that they return the highest level enumeration interface possible. For example, the Range()
operation return implements IValueReadOnlyList<,>
and the subsequent Select()
return does the same thing. IValueReadOnlyList<,>
derives from IReadOnlyList<>
so, we can change the method to return this interface:
public static IReadOnlyList<int> Example(int count)
=> ValueEnumerable
.Range(0, count)
.Select(value => value * 2);
Now, the caller is free to use the enumerator or the indexer, which are provided by this interface. This means that, all previous methods of iteration can be used plus the for
loops, which is much more efficient. The indexer performs fewer internal operations and doesn't need an enumerator instance. NetFabric.Hyperlinq
can also be used and the AsValueEnumerable()
conversion method still has to be used but, NetFabric.Hyperlinq
will now use the indexer.
Otherwise, NetFabric.Hyperlinq
will use enumerators. It implements all enumerators as value types. This allows the enumerators not to be allocated on the heap and calls to its methods to be non-virtual. The IEnumerable<>
interface, and all other derived BCL interfaces, convert the enumerator to the IEnumerator<>
interface. This results in the enumerator to be boxed, undoing the mentioned benefits.
NetFabric.Hyperlinq
defines enumerable interfaces that contain the enumerator type as a generic parameter. This allows the caller to use it without boxing. Consider changing the method to return one of these interfaces:
public static IValueReadOnlyList<int, ReadOnlyList.SelectEnumerable<ValueEnumerable.RangeEnumerable, int, int>.DisposableEnumerator> Example(int count)
=> ValueEnumerable
.Range(0, count)
.Select(value => value * 2);
The caller is now able to use NetFabric.Hyperlinq
without requiring the AsValueEnumerable()
conversion method. In this case, it will use the indexer in all subsequent operations but the enumerator is still available. In other cases (IValueEnumerable<,>
, IValueReadOnlyCollection<,>
and IAsyncValueEnumerable<,>
) it will use the enumerator.
NetFabric.Hyperlinq
also implements the enumerables as value types. Returning an interface means that the enumerable will still be boxed. If you want to also avoid this, consider changing the return type to be the actual enumerable type. In this case:
public static ReadOnlyList.SelectEnumerable<ValueEnumerable.RangeEnumerable, int, int> Example(int count)
=> ValueEnumerable
.Range(0, count)
.Select(value => value * 2);
NOTE: Returning the enumerable type or a value enumeration interface, allows major performance improvements but creates a library design issue. Changes in the method implementation may result in changes to the retun type. This is a breaking change. This is an issue on the public API but not so much for the private and internal methods. Take this into consideration when deciding on the return type.
NetFabric.Hyperlinq
operations can be composed just like with System.Linq
. The difference is on how each one optimizes the internals to reduce the number of enumerators required to iterate the values.
Both System.Linq
and NetFabric.Hyperlinq
optimize the code in the following example so that only one enumerator is used to perform both the Where()
and the Select()
:
var result = source
.Where(item => item > 2)
.Select(item => item * 2);
But, System.Linq
does not do the same for this other example:
var result = source
.Where(item => item > 2)
.First();
System.Linq
has a second overload for methods, like First()
and Single()
, that take the predicate as a parameter and allow the Where()
to be removed. In NetFabric.Hyperlinq
this is not required. With the intention of reducing the code to be maintained and tested, these other overloads actually are not available in NetFabric.Hyperlinq
.
NetFabric.Hyperlinq
includes many more composition optimizations. In the following code, only one enumerator is used, and only because of the Where()
operation. Otherwise, the indexer would have been used instead. Also, the Select()
is applied after First()
, so that it's applied to only to the resulting item:
var result = array
.Skip(1)
.Take(10)
.Where(item => item > 2)
.Select(item => item * 2)
.First();
In System.Linq
, the aggregation operations like First()
, Single()
and ElementAt()
, throw an exception when the source has no items. Often, empty collections are a valid scenario and exception handling is very slow. System.Linq
has alternative methods like FirstOrDefault()
, SingleOrDefault()
and ElementAtOrDefault()
, that return the default
value instead of throwing. This is still an issue when the items are value-typed, where there's no way to distinguish between an empty collection and a valid item.
In NetFabric.Hyperlinq
, aggregation operations return an Option<>
type. This is similar in behavior to the Nullable<>
but it can contain reference types.
Here's a small example using First()
:
var result = source.First();
if (result.IsSome)
Console.WriteLine(result.Value);
It also provides a deconstructor so, you can convert it to a tuple:
var (isSome, value) = source.First();
if (isSome)
Console.WriteLine(value);
If you prefer a more functional approach, you can use Match()
to specify the value returned when the collection has values and when it's empty. Here's how to use it to define the previous behavior of First()
and FirstOrDefault()
:
var first = source.First().Match(
item => item,
() => throw new InvalidOperationException("Sequence contains no elements"));
var firstOrDefault = source.First().Match(
item => item,
() => default);
Console.WriteLine(first);
Console.WriteLine(firstOrDefault);
Match()
can also be used to define actions:
source.First().Match(
item => Console.WriteLine(item),
() => { });
The NetFabric.Hyperlinq
operations can be applied to Option<>
, including Where()
, Select()
and SelectMany()
. These return another Option<>
with the predicate/selector applied to the value, if it exists.
source.First().Where(item => item > 2).Match(
item => Console.WriteLine(item),
() => { });
There are operations that do have to allocate on the heap. Some need to use a collection internally (like Distinct()
) and others need to return a collection (like ToList()
and ToArray()
).
Buffer pools allow the use of heap memory without adding pressure to the garbage collector. It preallocates a chunk of memory and "rents" it as required. The garbage collector will add this memory to the Large Object Heap (LOH).
Netfabric.Hyperlinq
uses the buffer pool for internal collections.
ToArray()
is usually used to cache values for a brief period. Netfabric.Hyperlinq
adds an oveload that takes a MemoryPool<>
as a parameter:
using(var buffer = source.ToArray(MemoryPool<int>.Shared))
{
var memory = buffer.Memory;
// use memory here
}
It returns an IMemoryOwner<>
. The using
statement guarantees that it is disposed and the buffer automatically returned to the pool.
Buffer pools may return buffers larger than requested. ToArray()
always returns buffers with the exact same size as the result. The Length
property can be used to check for the end of an iteration.
Articles explaining implementation:
- Aggregation
Count()
- Conversion
AsEnumerable()
AsValueEnumerable()
ToArray()
ToList()
ToDictionary(Selector<TSource, TKey>)
ToDictionary(Selector<TSource, TKey>, IEqualityComparer<TKey>)
ToDictionary(Selector<TSource, TKey>, Selector<TSource, TElement>)
ToDictionary(Selector<TSource, TKey>, Selector<TSource, TElement>, IEqualityComparer<TKey>)
- Element
ElementAt()
First()
Single()
- Filtering
Where(Predicate<TSource>)
Where(PredicateAt<TSource>)
- Generation
Create(Func<TEnumerator>)
Empty()
Range(int, int)
Repeat(TSource, int)
Return(TSource)
- Projection
Select(Selector<TSource, TResult>)
Select(SelectorAt<TSource, TResult>)
SelectMany(IValueEnumerable<TSource>)
- Partitioning
Take(int)
Skip(int)
- Quantifier
All(Predicate<TSource>)
All(PredicateAt<TSource>)
Any()
Any(Predicate<TSource>)
Any(PredicateAt<TSource>)
Contains(TSource)
Contains(TSource, IEqualityComparer<TSource>)
- Set
Distinct(TSource)
Distinct(TSource, IEqualityComparer<TSource>)
- Enumeration in .NET by Antão Almada
- Performance of value-type vs reference-type enumerators by Antão Almada
- Performance Tuning for .NET Core by Reuben Bond
- ValueLinqBenchmarks by Ben Adams
- C# - How method calling works by Levi Botelho
- Improving .NET Disruptor performance — Part 2 by Olivier Coanet
- Optimizing string.Count all the way from LINQ to hardware accelerated vectorized instructions by Sergio Pedri
- Simulating Return Type Inference in C# by Alexey Golub
- Pooling large arrays with ArrayPool by Adam Sitnik
The following open-source projects are used to build and test this project:
- .NET
- BenchmarkDotNet
- coveralls
- coverlet
- ILRepack
- ILRepack.MSBuild.Task
- NetFabric.Assertive
- NetFabric.Hyperlinq.Analyzer
- Nullable
- Source Link
- xUnit.net
This project is licensed under the MIT license. See the LICENSE file for more info.