Adds JsonWebToken Constructor that accepts Memory<char>

[sruke]

Introduces a JsonWebToken Constructor that accepts ReadOnlyMemory representation of the encodedToken

Description

This PR

• Introduces a JsonWebToken constructor that accepts ReadOnlyMemory representation of the encodedToken
• Updates existing JsonWebToken constructors to use string.AsMemory() to maintain a single code path
• ReadTokens(..) Uses Memory.Span to extract the token segments such as the header, payload, encryptedKeyBytes etc.
• Refactors CreateClaimSet by moving ArrayPool usage into the JsonWebToken that reduces memory allocation by 4.6% during JWS creation.

Performance Assessment on .NET 8:
When using the preview version and opting for the constructor that accepts a string for the encoded token, the generation of a JWS yields a 3.95% decrease in memory usage, a 4.52% reduction in Gen0, and a 4.61% decrease in execution time compared to the latest release. The performance of the constructor that accepts ReadOnlyMemory is comparable to the constructor accepting a string for the encoded token, both for JWS and JWE.

Preview branch:

Dev (main) branch:

Why?
The new constructor allows a customer to leverage spans for slicing the encoded token directly from the Authorization Header eliminating additional String.Split operations on their end and can further improve performance.

[jennyf19]

Length

what's the issue with .Length? 'dotIndex' is the first dot character, so I'd think dotIndex == encodedJsonSpanCopy.Length would work.

[sruke]

My thinking is that the value of EncodedJsonSpanCopy.Length will fall outside the range of the string and EncodedJsonSpanCopy.Length - 1 would be more appropriate.

[Saibamen]

In BenchmarkDotNet.Artifacts\results folder, you have *.md file to copy and paste markdown table, no need to make screenshots

[Saibamen]

ReadJWE_FromString has less allocations in current master branch

You can use profiling to search for more optimization: https://devblogs.microsoft.com/visualstudio/a-look-back-visual-studios-profiling-tool-advancements-in-2023/

[sruke]

ReadJWE_FromString has less allocations in current master branch

You can use profiling to search for more optimization: https://devblogs.microsoft.com/visualstudio/a-look-back-visual-studios-profiling-tool-advancements-in-2023/

Yes I have noticed and working towards finding a way to reduce allocations using the profiler.

[sruke]

ReadJWE_FromString has less allocations in current master branch
You can use profiling to search for more optimization: https://devblogs.microsoft.com/visualstudio/a-look-back-visual-studios-profiling-tool-advancements-in-2023/

Yes I have noticed and working towards finding a way to reduce allocations using the profiler.

ReadJWE_FromString has less allocations in current master branch

You can use profiling to search for more optimization: https://devblogs.microsoft.com/visualstudio/a-look-back-visual-studios-profiling-tool-advancements-in-2023/

ReadJWE_FromString has less allocations in current master branch
You can use profiling to search for more optimization: https://devblogs.microsoft.com/visualstudio/a-look-back-visual-studios-profiling-tool-advancements-in-2023/

Yes I have noticed and working towards finding a way to reduce allocations using the profiler.

@keegan-caruso found that ReadJWETokenTests is not an appropriate benchmark for assessing JWE creation performance. Claims are not populated during JsonWebToken creation since the token is encrypted; they are only populated during validation when the token is decrypted. ValidateJWEAsyncTests is a more suitable benchmarks that creates and validates a JWE. On running this benchmark, there is no performance concern. There is a .5% reduction in memory allocation.

sruthi/JsonWebTokenSpan

BenchmarkDotNet=v0.13.5, OS=Windows 11 (10.0.22631.3007), VM=Hyper-V
AMD EPYC 7763, 1 CPU, 16 logical and 8 physical cores
.NET SDK=8.0.200-preview.23624.5
  [Host]    : .NET 8.0.1 (8.0.123.58001), X64 RyuJIT AVX2
  MediumRun : .NET 8.0.1 (8.0.123.58001), X64 RyuJIT AVX2

Job=MediumRun  MaxAbsoluteError=10.0000 ms  IterationCount=15  
LaunchCount=4  WarmupCount=10  

Method	Mean	Error	StdDev	P90	P95	P100	Gen0	Gen1	Allocated
JsonWebTokenHandler_ValidateJWEAsync	40.20 μs	0.113 μs	0.252 μs	40.56 μs	40.63 μs	40.90 μs	0.9766	0.1221	16.2 KB

dev

BenchmarkDotNet=v0.13.5, OS=Windows 11 (10.0.22631.3007), VM=Hyper-V
AMD EPYC 7763, 1 CPU, 16 logical and 8 physical cores
.NET SDK=8.0.200-preview.23624.5
  [Host]    : .NET 8.0.1 (8.0.123.58001), X64 RyuJIT AVX2
  MediumRun : .NET 8.0.1 (8.0.123.58001), X64 RyuJIT AVX2

Job=MediumRun  MaxAbsoluteError=10.0000 ms  IterationCount=15  
LaunchCount=4  WarmupCount=10  

Method	Mean	Error	StdDev	P90	P95	P100	Gen0	Gen1	Allocated
JsonWebTokenHandler_ValidateJWEAsync	40.33 μs	0.133 μs	0.286 μs	40.61 μs	40.72 μs	41.39 μs	0.9766	0.1221	16.29 KB

[jennyf19]

[Saibamen]

Can you paste final benchmark results here?