Proposal: Allow tuples as attribute parameters

[lukepaw]

Currently attributes may only use a small set of types as their parameters.
As tuples are deterministic in their layout, as long as they consist of types that can be used as attribute parameters (primitives, strings, arrays of the former), they themselves should also be usable as such (including arrays of tuples).

[HaloFour]

This limitation is set by the CLR which would need support for reading the attribute parameters at runtime in order to instantiate the attribute.

[alrz]

I'd like this to be relaxed for any type constructor,

[SomeAttribute(Foo(5))]

We can already serialize types (for typeof), so I think this shouldn't be impossible to do without clr changes?

[Joe4evr]

We can already serialize types (for typeof [...]

That's done by serializing the type's FQDN in its place and looking it up/injecting it at runtime.

[ufcpp]

This could be essentially the same issue as const ValueTuple and Unmanaged Constants.

[alrz]

That's done by serializing the type's FQDN in its place and looking it up/injecting it at runtime.

That's why I think it's not a clr concern. Only the reflection API needs to be updated to recognize the new format?

[HaloFour]

@alrz

That's why I think it's not a clr concern. Only the reflection API needs to be updated to recognize the new format?

That's possible. To me the line between CLR and BCL has always been a little fuzzy when it comes to mscorlib. Either way it requires a modification outside of the C# language to be supported.