With the ground-up changes in ASP.NET with ASP.NET Core which is still going on, it appears that now it is turn of C# language and the run-time. I am sure the ground work must have been going on from last couple of years as these changes required massive ground up changes but I am hoping there are lot more changes will take place in future. After using the power of Span<T> and Memory<T> in ASP.NET Core 2.1, which is claimed to be 70-80% faster than ASP.NET Core 1.X, it is planned to release for all of us. In this post, we will be discussing the new type Span<T>. Before jumping directly into Span<T>, let’s have a quick overview current types.
Every object in C# can be categorized in two type: ValueType and Reference Type. We know Value Type instances gets created on stack and only available in the current scope while reference gets created on managed heap which is taken care by Garbage Collector (GC). Garbage collection is expensive process so if we can minimize the reference type object creation and use more Value Type objects then it could be a huge performance benefit. I am not saying that GC is not properly optimized or similar but no GC is always better that any performance optimized GC. Let’s understand how Span can help in few scenarios.
Note – Currently this feature is available in prerelease state so some changes in final release are expected. To use this this feature, we need we need Visual Studio 2017 15.5+ version and need to install nuget package – System.Memory (Check the include prerelease checkbox while searching it). You can have a look here if getting an error.
What is Span<T>
Span<T> is a ValueType which allows us write low level code in safe and efficient way without any GC overhead. It’s as efficient as working with unmanaged pointers with the benefits of C#. Span<T> represents a contiguous region of arbitrary memory which could be unmanaged memory, memory allocated on stack and arrays or strings.
Let’s understand it with the help of a string.
Here we can see that when we use Substring method on a string, a new copy gets created on the heap. But, we may not want a new copy as we just read that particular part and even sometimes we may need to update the original string itself. Let’s see how does Span<T> work here
So here we can see that we got a ReadOnlySpan from the string and then used Slice (similar to Substring) method of span to get a specific part of the string but here both the variables refer to same memory location. As span is a struct, lastName doesn’t cost any GC overhead. The two parameters of the Slice method are, starting index and length of the sliced part. Let’s see the complete example
Span<Char> name = "Brij Mishra".ToCharArray().AsSpan(); Span<char> lastName = name.Slice(5,4); lastName = 'k'; Console.WriteLine(lastName.ToString()); // Prints -> Miskra Console.WriteLine(name.ToString()); // Prints -> Brij Miskra lastName = 'k'; // Throws Exception -> System.IndexOutOfRangeException'
Here we can see that change in one character using lastName reflects in both the variables. In the last line, when we try to access the fifth character via lastName, it throws IndexOutofRangeException even we know that we have character at this location. But since lastName was created with only four characters, we cannot access it.
Span vs ReadOnlySpan
As the name suggests Span<T> allows us update the values but many a times, we use the sub-string only for readonly purposes. Say we have a method where we need to pass the part of the string which is just needed for reading purposes, it is safer to pass ReadOnlySpan to save from any accidental update because it will affect all the places inside the program. The current version of the package, returns the ReadOnlySpan while getting the span out of string. So if you are sure that we need the span for read only purposes then we should use ReadOnlySpan. Trying to make any changes in ReadOnlySpan will be an compile time error.
If you are working with payloads where you create and manipulate strings a lot, it can be quite useful. Similarly, parsing needs usage Substring where it can be quite useful. Other usages are not so common like working with buffers and doing multiple transformations like encoding/decoding, parsing, escaping etc. Working with network buffers can be other quite useful scenario.
Span<T> is designed to stay in the stack onlu which is a limiting factor as well. Few key limitations are
- We can’t box/unbox because it can’t stay on the heap
- We can’t have it as fields in classes or even in non-ref structs
- Cannot be used in lambda method because it might turn into field.
- Due to stack only field, cannot be use in async methods or iterators
Even having quite many limitations, it can be really helpful where you have big structs, file I/O , Network I/O, string manipulations, working with buffers, synchronous methods.
Hope you have enjoyed the post.