Disable frequently failing CriticalFinalizer test

[jkotas]

Contributes to #76041

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[filipnavara]

I thought https://github.com/dotnet/runtime/pull/76089/files was already supposed to do that and got merged...

[jkotas]

I thought https://github.com/dotnet/runtime/pull/76089/files was already supposed to do that and got merged...

It disabled the test for coreclr only. The test is failing everywhere else as well. And the issue was not labelled disabled-test so I did not even know that this was done.

[VSadov]

It looks like #76089 did this incorrectly. The format of this file is confusing. - there should be a name of the test (not the containing folder) followed by a "/".

Also I disabled the test for coreclr only, but it fails on Mono as well, with the same "Finalized 0 Normal and 0 Critical objects." failure.
That likely indicates the issue with the test. Hard to imagine the same bug in two different GCs.

[jkotas]

That likely indicates the issue with the test.

Yes, it looks like a test issue.

Note that this test was recently moved from old Mono test bed that was not running to the actively running tests

[VSadov]

That likely indicates the issue with the test. Hard to imagine the same bug in two different GCs.

There is still a possibility that there are different reasons for objects not collected - Mono does not guarantee anything since the stack scan is conservative, although the stack should be nearly empty when Collect() is called in the test, so chances for that being the reason are very low. The test originates from mono, so it was passing for a very long time (if it was running).

The reason for CoreCLR failing to collect is harder to guess. Collect(); WaitForPendingFinalizers(); should be deterministic.

[VSadov]

Thanks!

[VSadov]

moved from old Mono test bed that was not running

Not running? That would explain failures on Mono.

[AntonLapounov]

For CoreCLR, I have found that either setting arr = null or doing GC.Collect(); GC.WaitForPendingFinalizers(); two times in a row fixes the issue, which may indicate the arr array remains rooted on the first GC.Collect() call. Apparently, a number of existing tests for finalizers also call GC.Collect() twice in a row, e.g.,

runtime/src/tests/GC/Features/Finalizer/finalizeother/finalizearray.cs

Lines 44 to 46 in ab56392

	GC.Collect();
	GC.WaitForPendingFinalizers();
	GC.Collect();

Also, it is much easier to reproduce the issue with tiered compilation turned off.

The original Mono test would not use the array of references and it would pass if at least one of the Normal (P) instances out of 100 was collected, so it used to be much weaker (and not compatible with GC stress).

[AntonLapounov]

Thanks!

[AntonLapounov]

The failure in 'CoreCLR Product Build windows x86 release' is #48070.

[VSadov]

I suppose, since this is blocking clean CI and tests are passing, we can merge.