Environment variable should override property for gcServer.

[AaronRobinsonMSFT]

Fixes #61674

Add tests for the scenario.

The added tests can also be used to validate any of the configuration properties and environment variables used for start-up.

/cc @cshung @elinor-fung

@jkoritzinsky and @trylek Can you please confirm the methodology in this new test works with the system being built?

[ghost]

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

Issue Details

---

Add tests for the scenario.

Fixes #61674

/cc @cshung @elinor-fung

Author:	AaronRobinsonMSFT
Assignees:	-
Labels:	area-GC-coreclr
Milestone:	.NET 7.0

[cshung]

LGTM, thanks for adding the test!

[mangod9]

Thanks @AaronRobinsonMSFT for fixing, this feels like something we should queue up for 6.0.x. Thx!

[AaronRobinsonMSFT]

/backport to release/6.0

[github-actions]

Started backporting to release/6.0: https://github.com/dotnet/runtime/actions/runs/1495545849

[am11]

@AaronRobinsonMSFT, FYI, this test failed on GCStandAloneServer legs:

• Windows arm64 checked https://helix.dot.net/api/2019-06-17/jobs/1a1caea9-5980-4759-a7d1-da30e1a225a6/workitems/baseservices.RuntimeConfiguration/console
• Windows x64 checked https://helix.dot.net/api/2019-06-17/jobs/6c1bb956-e09d-4509-bc09-9d28f99834cf/workitems/baseservices.RuntimeConfiguration/console.
• Linux arm64 checked https://helix.dot.net/api/2019-06-17/jobs/de826141-187b-4ee7-ba88-31c2637b46af/workitems/baseservices.RuntimeConfiguration/console

[AaronRobinsonMSFT]

@cshung this seems like something the GC team should investigate. I am unclear why this would be occurring but given the simplicity of the test I'd wager it is unlikely a test issue.

[cshung]

@mangod9 It looks like the GCStandAloneServer set COMPlus_gcServer=1, so even when the code set DOTNET_gcServer=0, we are still unable to disable it.

[mangod9]

ok makes sense. @am11 what CI job was failing due to this issue? Is the failure only after this change or unrelated to this?

[am11]

@mangod9, it's a new test added by this PR, which started to fail in outerloop test when this PR was merged. Perhaps the root-cause was preexisting and the test has surfaced the problem.

It looks like the GCStandAloneServer set COMPlus_gcServer=1, so even when the code set DOTNET_gcServer=0, we are still unable to disable it.

Based on the config parsing implementation:

runtime/src/coreclr/inc/clrconfignocache.h

Lines 75 to 86 in 4da6b9a

	// Priority order is DOTNET_ and then COMPlus_.
	strcpy_s(nameBuffer, ARRAY_SIZE(nameBuffer), DOTNET_PREFIX_A);
	fallbackPrefix = COMPLUS_PREFIX_A;
	}
	strcat_s(nameBuffer, ARRAY_SIZE(nameBuffer), cfg);
	LPCSTR val = getEnvFptr != NULL ? getEnvFptr(nameBuffer) : getenv(nameBuffer);
	if (val == NULL && fallbackPrefix != NULL)
	{
	strcpy_s(nameBuffer, ARRAY_SIZE(nameBuffer), fallbackPrefix);
	strcat_s(nameBuffer, ARRAY_SIZE(nameBuffer), cfg);

first DOTNET_gcServer will be checked and if the value is NULL, it will fallback to legacy, COMPlus_gcServer. It is LPCSTR, so =0 shouldn't be confused by NULL/0, right?