Optimization: convert ResolveRequest into readonly struct

[SergeiPavlov]

This will save GC work (one allocation per resolving).
Also:

• No need for null-checking
• Remove GetOrCreateInstanceThrowsArgumentNullExceptionWhenResolveRequestIsNull test as not-applicable anymore
• Add in function argument specifier to pass pointer instead of struct contents.

[codecov]

Codecov Report

Attention: 2 lines in your changes are missing coverage. Please review.

Comparison is base (5af8326) 78.47% compared to head (095a0c3) 78.49%.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1397      +/-   ##
===========================================
+ Coverage    78.47%   78.49%   +0.02%     
===========================================
  Files          201      201              
  Lines         5715     5716       +1     
  Branches      1169     1169              
===========================================
+ Hits          4485     4487       +2     
  Misses         716      716              
+ Partials       514      513       -1     

Files	Coverage Δ
src/Autofac/Core/Container.cs	67.74% <ø> (ø)
src/Autofac/Core/ImplicitRegistrationSource.cs	86.84% <ø> (ø)
src/Autofac/Core/Lifetime/LifetimeScope.cs	83.73% <ø> (+0.99%)	⬆️
...Resolving/Pipeline/DefaultResolveRequestContext.cs	87.09% <100.00%> (+3.22%)	⬆️
src/Autofac/Core/Resolving/ResolveOperation.cs	84.09% <ø> (-0.36%)	⬇️
.../Autofac/Diagnostics/DiagnosticSourceExtensions.cs	100.00% <ø> (ø)
...utofac/Diagnostics/OperationStartDiagnosticData.cs	100.00% <100.00%> (ø)
...rc/Autofac/Features/Decorators/DecoratorContext.cs	100.00% <100.00%> (ø)
...eatures/LazyDependencies/LazyRegistrationSource.cs	83.33% <100.00%> (+3.33%)	⬆️
...utofac/Features/Metadata/MetaRegistrationSource.cs	75.00% <ø> (ø)
... and 2 more

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[alistairjevans]

Hi @SergeiPavlov, could you please run a few of our benchmarks (I'm thinking DeepGraphResolve and ChildScopeResolve at least) and show before and after numbers here to prove that the benefit of allocation removal outweighs the one or two places we copy ResolveRequest into a class field in terms of the size of those classes and the copy time?

If you're unsure on how to run our benchmarks (under the 'bench' folder), let me know.

[SergeiPavlov]

Sure. Here are the results:

Branch develop:

ChildScopeResolve

Method	Mean	Error	StdDev	Gen0	Gen1	Allocated
Resolve	26.78 us	0.277 us	0.272 us	3.7231	1.8311	30.51 KB
ResolveNeverRegisteredFromChild	13.15 us	0.067 us	0.053 us	2.0905	1.0376	17.1 KB

DeepGraphResolve

Method	Mean	Error	StdDev	Gen0	Gen1	Allocated
Resolve	9.775 us	0.0253 us	0.0249 us	1.7700	0.0458	14.54 KB

==============

Branch struct_ResolveRequest (THIS)

ChildScopeResolve

Method	Mean	Error	StdDev	Gen0	Gen1	Allocated
Resolve	26.97 us	0.270 us	0.341 us	3.6621	1.8005	30.13 KB
ResolveNeverRegisteredFromChild	12.87 us	0.033 us	0.029 us	2.0905	1.0376	17.12 KB

DeepGraphResolve

Method	Mean	Error	StdDev	Gen0	Gen1	Allocated
Resolve	9.438 us	0.1794 us	0.2395 us	1.6937	0.0305	13.85 KB

[alistairjevans]

I have no issue with the changes here, but adding in is a binary-breaking change, as is changing ResolveRequest to a struct; which really should make this a major version bump.

Sure, most people will use the extension methods, but I almost guarantee someone somewhere has their own extension method that instantiates ResolveRequest and calls ResolveComponent.

We may have to do a release for the .NET 8 release soonish, so I'm tempted to hold this change until we're prepared to do a major version bump.

[tillig]

Related: autofac/Autofac.Extensions.DependencyInjection#112

I'm not sure we can support M.E.DI keyed services without core Autofac changes. If we have to cut a new/breaking version for .NET 8, it might be worth trying to get these keyed service updates in place at the same time or we could end up with two breaking changes in a row.