NSAutoreleasePool instance should be added to all threads.

[AaronRobinsonMSFT]

This expands the current support where support was only added to ThreadPool threads. Now all managed threads or new native threads entering the runtime will get an NSAutoreleasePool allocated.

New feature switch, "System.Threading.Thread.EnableAutoreleasePool", was created and the "System.Threading.ThreadPool.EnableDispatchAutoreleasePool" was removed.

Updated AutoReleaseTest for the new scenarios. Also added support for tests to pass Runtime Configuration Options to corerun - uses the syntax found here - @LakshanF, @BruceForstall.

This is a follow up PR started with #47592

Contributes to xamarin/xamarin-macios#11256

/cc @rolfbjarne @jkotas @jkoritzinsky @elinor-fung

[AaronRobinsonMSFT]

@lambdageek I took a quick look at how to call these P/Invokes in mono. It isn't obvious to me how to try adding support. My best guess is around

runtime/src/mono/mono/metadata/threads.c

Lines 1160 to 1173 in 01b7e73

	/* On 2.0 profile (and higher), set explicitly since state might have been
	Unknown */
	if (internal->apartment_state == ThreadApartmentState_Unknown)
	internal->apartment_state = ThreadApartmentState_MTA;
	mono_thread_init_apartment_state ();
	/* Let the thread that called Start() know we're ready */
	mono_coop_sem_post (&start_info->registered);
	if (mono_atomic_dec_i32 (&start_info->ref) == 0) {
	mono_coop_sem_destroy (&start_info->registered);
	g_free (start_info);
	}

If you think this support should be easy to add, I'm willing to take a few hours to try it out.

[lambdageek]

@lambdageek I took a quick look at how to call these P/Invokes in mono. It isn't obvious to me how to try adding support. My best guess is around

https://github.com/dotnet/runtime/blob/01b7e73cd378145264a7cb7a09365b41ed42b240/src/mono/mono/metadata/threads.c#L1160-L1173

If you think this support should be easy to add, I'm willing to take a few hours to try it out.

What needs to happen? a call to Thread.AllocateThreadlocalAutoreleasePool ?

yea I think the right place is right before both callers of fire_attach_profiler_events. the spot you found is one, the other is

runtime/src/mono/mono/metadata/threads.c

Lines 1600 to 1601 in 01b7e73

	fire_attach_profiler_events (tid);

That one would set up an autorelease pool when a foreign thread attaches to the runtime via a pinvoke callback. (eg: managed code calls a pinvoke and passes it a delegate (or an UnmanagedCallersOnly function pointer), the native code creates a new thread and calls the unmanaged function pointer from the new thread).

In general the pattern to follow is like this:

runtime/src/mono/mono/metadata/threads.c

Lines 878 to 898 in 01b7e73

	MONO_STATIC_POINTER_INIT (MonoMethod, thread_exiting)
	thread_exiting = mono_class_get_method_from_name_checked (mono_defaults.thread_class, "OnThreadExiting", -1, 0, error);
	mono_error_assert_ok (error);
	MONO_STATIC_POINTER_INIT_END (MonoMethod, thread_exiting)
	g_assert (thread_exiting);
	if (mono_runtime_get_no_exec ())
	return;
	HANDLE_FUNCTION_ENTER ();
	ERROR_DECL (error);
	gpointer args [1];
	args [0] = internal;
	mono_runtime_try_invoke_handle (thread_exiting, NULL_HANDLE, args, error);
	mono_error_cleanup (error);

(some things to note: 1. if the method may not be present (removed by the IL Linker, or not defined on every plaform), don't assert; 2. mono_runtime_get_no_exec (check if we're the AOT compiler) is probably needed - a little bit of EE initialization happens even just during AOT compilation, and some of this thread attach code probably runs. 3. i'm assuming we want to swallow any managed exceptions, so the final mono_error_cleanup will swallow them. 4. since the method takes no arguments, the HANDLE_FUNCTION_ENTER/EXIT business is not needed and the actual call is more like mono_runtime_invoke_checked (method, NULL, NULL, error) )

[jkotas]

LGTM otherwise. Thank you!

[AaronRobinsonMSFT]

@lambdageek I've applied the feedback to Mono. Let me know what you think.

[lambdageek]

LGTM. Thanks!

[rolfbjarne]

@akoeplinger yes, I believe so (yay!)