Migrate JitOptimizationSensitive and GCStressIncompatible to SkipOnCoreClr attributes

[Copilot]

Replaces eligible MSBuild <JitOptimizationSensitive> and <GCStressIncompatible> properties with [SkipOnCoreClr(...)] applied directly to test entry points, moving skip logic from build-time property injection into the XUnit test runner where the test can safely run without process isolation.

Description

Why

MSBuild properties like <GCStressIncompatible> and <JitOptimizationSensitive> require <RequiresProcessIsolation>true</RequiresProcessIsolation> to take effect — process isolation is expensive. Moving eligible skips to XUnit attributes allows tests to be skipped in-process without spawning a new process per test. Tests that still require process isolation for other reasons retain <RequiresProcessIsolation>true</RequiresProcessIsolation>.

Changes

JitOptimizationSensitive — ilproj tests (4 .il files, 4 .ilproj files)

• Added SkipOnCoreClrAttribute with AnyJitOptimizationStress (0x11E) to each IL entry point
• Added .assembly extern Microsoft.DotNet.XUnitExtensions where missing
• Removed <JitOptimizationSensitive> and sole-cause <RequiresProcessIsolation> from project files
• Added <ProjectReference Include="$(TestLibraryProjectPath)" /> to project files that were missing it

GCStressIncompatible — all project types (~310 files total)

C# (150 .cs + 150 .csproj):

• Added [SkipOnCoreClr("This test is not compatible with GC stress.", RuntimeTestModes.AnyGCStress)] before each [Fact]/[Theory]/[ConditionalFact]/[ConditionalTheory]
• Removed <GCStressIncompatible> and sole-cause <RequiresProcessIsolation> from project files
• Added <ProjectReference Include="$(TestLibraryProjectPath)" /> to 65 csproj files that were missing it
• Removed dangling <GCStressIncompatible> from project files where the compiled source already had a corresponding [SkipOnCoreClr(..., RuntimeTestModes.AnyGCStress)] attribute

IL (6 .ilproj, 3 .il):

• 3 GenericContext tests: IL was pre-migrated; removed MSBuild properties from project files
• 3 JIT regression tests: Added SkipOnCoreClrAttribute (0xC0 = AnyGCStress) to IL entry points; updated project files
• Added <ProjectReference Include="$(TestLibraryProjectPath)" /> to 6 ilproj files that were missing it

F#:

• Runtime_72845 was left on the standalone [<EntryPoint>]/exit-code pattern with <GCStressIncompatible>true</GCStressIncompatible> and <RequiresProcessIsolation>true</RequiresProcessIsolation> retained, because the attempted XUnit conversion did not compile in targeted test builds

HeapVerifyIncompatible — migrated tests

• Added <ProjectReference Include="$(TestLibraryProjectPath)" /> to project files that were missing it
• For tests where <HeapVerifyIncompatible> was conditional on architecture, replaced the unconditional [SkipOnCoreClr(..., RuntimeTestModes.HeapVerify)] with [ConditionalFact] backed by a static bool property that combines the architecture check with TestLibrary.CoreClrConfigurationDetection.IsHeapVerify, preserving the original per-architecture skip semantics

Comment cleanup

• Audited edited project files where <GCStressIncompatible> was removed and updated or removed stale RequiresProcessIsolation comments
• Audited orphaned project comments that provided GC stress skip context, removed the orphaned comments, and moved important issue/context details into the corresponding [SkipOnCoreClr(...)] attribute messages

Infrastructure and documentation

• Doubled the browser-wasm (CoreCLR) runtime test work item timeout in helixpublishwitharcade.proj (123 → 246 minutes) to accommodate the larger merged test collections
• Updated requiresprocessisolation.md to document two additional triggers for <RequiresProcessIsolation>: <IsLongRunningGCTest> and <CLRTestExecutionArguments>

CI failure fixes and process-isolation audit

• Restored <RequiresProcessIsolation>true</RequiresProcessIsolation> to tests that still require isolation for GC measurement accuracy, long-running GC pre-commands, custom CLR test execution arguments, GC.WaitForPendingFinalizers, unsupported target handling, unloadability, NativeAOT incompatibility, output-copying project references, or Process.Start
• Added targeted [ActiveIssue] suppressions for CI failures on browser-wasm CoreCLR and Mono interpreter browser-wasm where appropriate
• Resolved merge conflict in GetTotalAllocatedBytes.cs while preserving GC stress and HeapVerify skip behavior

Not migrated (MSBuild property retained)

• Tests with ReferenceXUnitWrapperGenerator=false — no XUnit runner
• Profiler tests — custom host runner with no XUnit
• HW intrinsics wrapper projects — architecture-conditional property, no owned source
• OutputType=Library projects — no entry point
• Tests using top-level statements with no [Fact] method
• Runtime_72845.fsproj — F# XUnit conversion did not compile in targeted test builds, so this test retains <GCStressIncompatible>true</GCStressIncompatible> and standalone process isolation
• ReadyToRun tests with crossgen2 shell pre-commands (multifolder, determinism/crossgen2determinism) — [SkipOnCoreClr] only skips the XUnit test body, not pre-commands that run before the test executable
• Tests that still match requiresprocessisolation.md rules retain <RequiresProcessIsolation>, including projects with CLRTestExecutionArguments, IsLongRunningGCTest, GC.WaitForPendingFinalizers, CLRTestTargetUnsupported, UnloadabilityIncompatible, NativeAotIncompatible, ProjectReference/content copying requirements, or Process.Start

[dotnet-policy-service]

Tagging subscribers to this area: @agocke, @dotnet/gc
See info in area-owners.md if you want to be subscribed.

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

[MichalStrehovsky]

Before we make more changes to the test tree, it would be nice to make the test tree buildable again. https://github.com/dotnet/runtime/runs/68503855882

I assume this is from #125460, I've seen similar failures when I worked on the Loader tree. It would be nice to re-validate we're still running the same number of tests after #125460, the validation at #125460 (comment) must have missed this.

[jkoritzinsky]

I must have fat fingered the build command and missed the pri1 tests somehow. I'll fix that up before coming back to this.

[jkoritzinsky]

/azp run runtime-coreclr outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jkoritzinsky]

@copilot some of the projects you've edited still mention that they have RequiresProcessIsolation due to GCStressIncompatible. Please audit the projects you've edited to make sure that the reasons for a project being marked RPI are accurate.

[Copilot]

@copilot some of the projects you've edited still mention that they have RequiresProcessIsolation due to GCStressIncompatible. Please audit the projects you've edited to make sure that the reasons for...

Done in 48bc184. Audited all 78 project files where GCStressIncompatible was removed but the <!-- Needed for GCStressIncompatible, ... --> comment on RequiresProcessIsolation still referenced it. Updated each comment to reflect only the remaining reasons (e.g. UnloadabilityIncompatible, GC.WaitForPendingFinalizers, CMakeProjectReference, etc.). When GCStressIncompatible was the sole reason, the entire comment line was removed.

[jkoritzinsky]

/azp run runtime-coreclr outerloop

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jkoritzinsky]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Done in 1d065ad. Merged origin/main; the only conflict was in GetTotalAllocatedBytes.cs — I kept the [SkipOnCoreClr] (GC stress) and [ConditionalFact] (HeapVerify on ARM) attributes from our PR, and dropped the [ActiveIssue(..., IsArm)] that main's PR #127583 made obsolete by reducing the thread count on 32-bit.

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

[MichalStrehovsky]

/azp run runtime-coreclr outerloop, runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 2 pipeline(s).

[MichalStrehovsky]

@copilot the test src/tests/GC/API/GCHandle/Weak.csproj calls GC.WaitForPendingFinalizers() and is therefore not eligible to drop RequiresProcessIsolation (it also leaks a GCHandle, making it a poor fit for the unisolated runner). Audit all the tests that remove RequiresProcessIsolation=true in this branch against the rules in file docs/workflow/testing/coreclr/requiresprocessisolation.md and make sure we're not dropping RequiresProcessIsolation where we should be keeping it.

[Copilot]

@copilot the test src/tests/GC/API/GCHandle/Weak.csproj calls GC.WaitForPendingFinalizers() and is therefore not eligible to drop RequiresProcessIsolation (it also leaks a GCHandle, making it ...

Done in c307c3f. I audited all projects where this branch removed <RequiresProcessIsolation>true</RequiresProcessIsolation> against requiresprocessisolation.md and restored RPI for the remaining matches, including Weak.csproj for GC.WaitForPendingFinalizers. I also restored RPI for largearraytest, b143840, ParallelCrashTester, and R2RDumpTests, and reverted Runtime_72845 to the standalone GCStressIncompatible pattern after the F# XUnit conversion failed targeted builds. Baseline clr+libs and targeted builds passed; representative tests Weak, Runtime_72845, and b143840 passed.

[MichalStrehovsky]

@copilot changes in this branch leave things in non-sensical state.

For example:

src/tests/baseservices/exceptions/regressions/V1/SEH/VJ/MultipleException.csproj‎

Deleted the <GCStressIncompatible>true</GCStressIncompatible> line, left the <!-- Issue #22339 --> comment over it that provides important context for GCStressIncompatible as an orphaned line and failed to add the issue number to the newly added attribute in ‎src/tests/baseservices/exceptions/regressions/V1/SEH/VJ/MultipleException.cs.

Audit all the changes in this branch for this pattern and fix it by moving the important detail from the orphaned comment to the new attribute.

[Copilot]

@copilot changes in this branch leave things in non-sensical state.

For example:
...

Addressed in f4ac6ac. I audited the migrated GCStressIncompatible removals, removed the orphaned project comments, and moved the issue links/rationale into the corresponding SkipOnCoreClr(...AnyGCStress) attribute messages.