Avoid passing stack crawl mark unnecessarily deep in the call stack.

[filipnavara]

TypeName::GetTypeWorker was passed both a stack mark and requesting assembly. Only the requesting assembly is actually needed and the stack mark can be resolved earlier if necessary.

[jkotas]

Can this introduce (expensive) stackcrawl for caller assembly in some cases where it does not happen today? If yes, what are those cases and how important are they?

[filipnavara]

Can this introduce (expensive) stackcrawl for caller assembly in some cases where it does not happen today?

As far as I can see it should not add any new stackcrawl that was not there before.

The code paths in TypeName::GetTypeWorker that didn't resolve the stack mark in the parameter can only occur if pRequestingAssembly != NULL || pAssemblyGetType != NULL. In every other case the stack mark was eventually resolved.

I moved up the stack mark resolution to TypeName::GetTypeManaged. GetTypeManaged always called GetTypeWorker with pRequestingAssembly: NULL. Thus we only need to consider the pAssemblyGetType != NULL code path.

There are three calls to TypeName::GetTypeManaged. Two of them don't pass in the stack mark, so there's nothing to resolve. The third one passes a stack mark, but it sets pAssemblyGetType: NULL. Thus even for the last one the stack mark resolution would always happen even before the change.

[filipnavara]

@dotnet-bot test OSX10.12 x64 Checked Innerloop Build and Test

[filipnavara]

On second thought I can move the resolution from TypeName::GetTypeManaged to RuntimeTypeHandle::GetTypeByName since it's the only call site that actually uses it.

[filipnavara]

Additionally, one stack crawl is saved if GetTypeManaged gets to the fallback case with period prefixes where GetTypeWorker is called twice. Admittedly that's not a common case, but it's an improvement nonetheless.

[filipnavara]

@dotnet-bot test Windows_NT x64 full_opt ryujit CoreCLR Perf Tests Correctness
@dotnet-bot test Windows_NT x64 min_opt ryujit CoreCLR Perf Tests Correctness
@dotnet-bot test Windows_NT x86 full_opt ryujit CoreCLR Perf Tests Correctness
@dotnet-bot test Windows_NT x86 min_opt ryujit CoreCLR Perf Tests Correctness

[jkotas]

@dotnet-bot test Windows_NT x64 full_opt ryujit CoreCLR Perf Tests Correctness
@dotnet-bot test Windows_NT x64 min_opt ryujit CoreCLR Perf Tests Correctness
@dotnet-bot test Windows_NT x86 full_opt ryujit CoreCLR Perf Tests Correctness
@dotnet-bot test Windows_NT x86 min_opt ryujit CoreCLR Perf Tests Correctness

[jkotas]

Thanks