On: A DoS Attack Against the C# Compiler

[gpvos]

The Don Syme article (first link) mentions "the critical nature of the interaction that app domains pose to generics". What interaction is that, and what is critical about its nature?

[matthewwarren → gpvos]

Hmm, that's a really good question.

I did a quick search in AppDomain.cpp [0] for 'generic' and there's quite a few comments/variables that mention it. Maybe this [1] is one of the tricky interactions?

[0]: https://github.com/dotnet/coreclr/blob/master/src/vm/appdoma... [1]: https://github.com/dotnet/coreclr/blob/master/src/vm/appdoma...

[Aloraman → matthewwarren]

This is CoreCLR, much younger version of CLR that does not support multiple AppDomains In different open-sourced CLR implementation aka SSCLI20 (aka Rotor) there are similar interactions for type's domain discovery and DynamicClassTable, but there are also some interactions about GCStatics. I'm not sure if fullblown .Net CLR implementation was released yet

[matthewwarren → Aloraman]

Yeah that's a good point, CoreCLR only has the single/main AppDomain code and Rotor was only 'sort of' .NET 2.0, but with a different GC and JIT.

So without asking Don Syme himself we can only guess.

> I'm not sure if fullblown .Net CLR implementation was released yet

No and AFAIK they don't plan to. There's 'Reference Source' [0], but that's the Base class libraries only, not the CLR itself.

[0]: referencesource.microsoft.com

[Aloraman → matthewwarren]

I kinda guess this has something to do with System.__Canon type, which is a substitute type for open generic parameter type - there is no benefit in having different canon types per appdomain