Fix invalid continuation caused by exception after inlining in TryRunInline

[kotlarmilos]

Description

When a task was inlined in TryRunInline, and if an exception was thrown after inlining, it would call Finish in the catch block due to an invalid taskQueued value, which set the sentinel in the continuation object. Later, this could cause ExecuteWithThreadLocal to invoke Finish again and try to run the sentinel as a continuation, throwing InvalidCastException.

Contributes to #83520

[dotnet-policy-service]

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

[tannergooding]

Cc. @stephentoub

[stephentoub]

Thanks! If this is the root cause, it's interesting that folks have only started reporting it recently. As far as I can tell, it's been like this for 15 years. :) You also mentions this "contributes" to the original issue; do you believe this is only part of it?

[kotlarmilos]

Thanks! If this is the root cause, it's interesting that folks have only started reporting it recently. As far as I can tell, it's been like this for 15 years. :) You also mentions this "contributes" to the original issue; do you believe this is only part of it?

I believe this is the only part to the fix, though I can’t say for sure. It’s interesting that reports have only started appearing recently.

[kotlarmilos]

The reproduction test is unreliable. It fails under debug configurations, and I haven’t been able to set conditions for consistently successful runs. Adding more retry attempts could improve reliability, but it would also slow down the CI.

@stephentoub Let me know how you want to proceed:

1. Spend more time to get reliable test
2. Merge this change and provide a private build so customers can test it against .NET 9 before backporting

[stephentoub]

Whether or not it fully solves the issue, the fix is small and correct and shouldn't hurt, so let's go ahead and merge it. Thanks.