Make the output of ThunkGenerator more diffable

[AustinWise]

Before this change, the addition of a new method on the JIT<->EE interface resulted in large diffs in CorInfoImpl_generated.cs when viewed on GitHub.com. For example the changes from #120411 look like this:

callbacks[154] = (delegate* unmanaged<IntPtr, IntPtr*, CORINFO_RESOLVED_TOKEN*, CORINFO_SIG_INFO*, CORINFO_GET_TAILCALL_HELPERS_FLAGS, CORINFO_TAILCALL_HELPERS*, byte>)&_getTailCallHelpers;
-callbacks[155] = (delegate* unmanaged<IntPtr, IntPtr*, CORINFO_METHOD_STRUCT_*>)&_getAsyncResumptionStub;
-callbacks[156] = (delegate* unmanaged<IntPtr, IntPtr*, CORINFO_RESOLVED_TOKEN*, byte, byte>)&_convertPInvokeCalliToCall;
-callbacks[157] = (delegate* unmanaged<IntPtr, IntPtr*, InstructionSet, byte, byte>)&_notifyInstructionSetUsage;
+callbacks[155] = (delegate* unmanaged<IntPtr, IntPtr*, nuint, bool*, nuint, CORINFO_CLASS_STRUCT_*>)&_getContinuationType;
+callbacks[156] = (delegate* unmanaged<IntPtr, IntPtr*, CORINFO_METHOD_STRUCT_*>)&_getAsyncResumptionStub;
+callbacks[157] = (delegate* unmanaged<IntPtr, IntPtr*, CORINFO_RESOLVED_TOKEN*, byte, byte>)&_convertPInvokeCalliToCall;
+callbacks[158] = (delegate* unmanaged<IntPtr, IntPtr*, InstructionSet, byte, byte>)&_notifyInstructionSetUsage;
// ...

After this change, the diffs are more readable:

 s_vtbl.GetTailCallHelpers = &_getTailCallHelpers;
+s_vtbl.GetContinuationType = &_getContinuationType;
 s_vtbl.GetAsyncResumptionStub = &_getAsyncResumptionStub;
 s_vtbl.NotifyInstructionSetUsage = &_notifyInstructionSetUsage;
// ...

Because the new pattern is pre-initializable and does not allocate native memory for each CorInfoImpl, it should be a little faster and use less a little less memory. I doubt that makes an observable difference though since CorInfoImpl instances are reused.

[EgorBo]

Heh, I actually wanted to do the same once, thanks! 🙂

[jkotas]

Should we rather change this to the canonical pre-initialized vtable pattern

runtime/src/tests/nativeaot/SmokeTests/Preinitialization/Preinitialization.cs

Lines 1761 to 1762 in 86dfeb9

	[FixedAddressValueType]
	public static readonly IInspectableVftbl Vtbl;

? It is nicely diffable as well.

[AustinWise]

Should we rather change this to the canonical pre-initialized vtable pattern

runtime/src/tests/nativeaot/SmokeTests/Preinitialization/Preinitialization.cs

Lines 1761 to 1762 in 86dfeb9

	[FixedAddressValueType]
	public static readonly IInspectableVftbl Vtbl;

? It is nicely diffable as well.

Oh, that looks nicer. I'll mark this as draft while I rework this PR to look more like that.

[jkotas]

runtime/src/coreclr/tools/Common/JitInterface/JitConfigProvider.cs

Lines 168 to 171 in 74cf618

	void** callbacks = (void**)NativeMemory.Alloc((nuint)(sizeof(void*) * numCallbacks));
	callbacks[0] = (delegate* unmanaged<IntPtr, byte*, int, int>)&getIntConfigValue;
	callbacks[1] = (delegate* unmanaged<IntPtr, byte*, byte*, int, int>)&getStringConfigValue;

can use the same treatment (it can be a separate PR).

[AustinWise]

I removed the extra new lines.

[AustinWise]

I applied the same pattern to JitConfigProvider. I think that addresses all feedback so far.

[jkotas]

Suggested change

	private struct ICorJitInfoVtbl
	private struct ICorJitInfoCallbacks

One last nit: This is not actually a ICorJitInfo vtable. The actual ICorJitInfo vtable is implemented in C++ by wrapping the callbacks from this array with exception handling goo. Maybe this should be called ICorJitInfoCallbacks, UnmanagedCallbacks (matches GetUnmanagedCallbacks) to avoid confusion?