Slightly reduce the Depth3Test that often times out on Linux

[trylek]

This is one of the regression tests I introduced as part of implementing protection against infinite generic recursion in Crossgen2. Turns out that on the relatively slow Helix Linux VMs the test often times out due to compilation time exceeding 10 minutes. I propose slightly trimming down the test so that it uses less time in Helix.

Thanks

Tomas

/cc @dotnet/crossgen-contrib, @dotnet/runtime-infrastructure

[ivdiazsa]

Do we know if this is a natural problem with not-too-fast machines, or whether we have an unnatural slowdown we ought to investigate and fix?

[trylek]

Well, in #87532 I proposed a minor optimization in the algorithm but the PR was abandoned as Michal pointed out that this is more or less a pathological case we don't need to optimize for. In a similar vein, on my quite fast devbox the test in checked mode takes about 1 minute, I can easily imagine it can be 10 times slower on a shared Azure VM.

[ivdiazsa]

Well, in #87532 I proposed a minor optimization in the algorithm but the PR was abandoned as Michal pointed out that this is more or less a pathological case we don't need to optimize for. In a similar vein, on my quite fast devbox the test in checked mode takes about 1 minute, I can easily imagine it can be 10 times slower on a shared Azure VM.

I see. Well, in that case it LGTM! I've restarted the failing pipeline because it seems entirely unrelated to this change. So unless @MichalStrehovsky has further feedback, I think we can merge this, assuming that failure is indeed unrelated like I suspect it to be.

[trylek]

Thanks Andy and Ivan for reviewing. I have rebased the change against latest main and I have added the removal of the clause for Depth3Test from issues.targets where Bruce added it last week to baseline the relatively frequent lab failure. I don't think it's necessary to port this basically negative test to .NET 8 RC1 - let me know if you think otherwise.

[ivdiazsa]

Wasm failure with Mono is a known infrastructure issue unrelated to this change, so I'll be merging this right now.