Mark GCStressIncompatible and similar tests as out-of-proc

[trylek]

According to current architecture we must mark conditionally
enabled tests as out of process because the special conditions
are tested in their individual execution scripts that are skipped
for ordinary merged tests.

Thanks

Tomas

[ghost]

Tagging subscribers to this area: @hoyosjs
See info in area-owners.md if you want to be subscribed.

Issue Details

---

According to current architecture we must mark conditionally
enabled tests as out of process because the special conditions
are tested in their individual execution scripts that are skipped
for ordinary merged tests.

Thanks

Tomas

Author:	trylek
Assignees:	-
Labels:	area-Infrastructure-coreclr
Milestone:	-

[trylek]

/azp run runtime-extra-platforms

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[BruceForstall]

This seems fine, although it seems like a better solution might be to have GCStressIncompatible automatically imply RequiresProcessIsolation. Otherwise, if we mark or unmark GCStressIncompatible we'll have to remember to do the correct thing for RequiresProcessIsolation. E.g., it's not necessarily true that removing GCStressIncompatible to fix a bug where a test was temporarily disabled on GCStress means you can also remove RequiresProcessIsolation.

[trylek]

Thanks Bruce, I'll make sure to address the various cleanup feedback we continue receiving as part of these changes and our various discoveries. It's actually not just about GCStressIncompatible, the same holds for UnloadabilityCompatible, JitOptimizationSensitive and other special conditions with targeted support in the generated execution scripts. Ultimately I'd love to minimize the cases forcing out-of-process execution because it incurs extra perf cost and is currently incompatible with Mono, we just need to identify a reasonable middle ground I guess.

[jkoritzinsky]

After looking over things again, I remembered that we can actually use the SkipOnCoreCLR attribute as there's a setting to detect GCStress and skip a test for GCStressIncompatible (same goes for JitOptimizationSensitive).

UnloadabilityIncompatible would have to be handled differently, but it's not impossible to handle.

[trylek]

I'm merging this in for now to expedite the switchover of JIT/Methodical tests as soon as the GC bug I believe to be causing hangs and stack overflows on Linux has been fixed. For the next phase of test merging we should collect all this feedback and prepare a robust design plan ending up with something cleaner than we have now. We can discuss that on the primary issue #54512.