Improve inference for interpolated tuple parameters in stagedfunctions

[mbauman]

Would it be possible easy to improve the inference here?

julia> immutable Foo{T} end

julia> stagedfunction bar{T}(x::Foo{T})
         return :(Foo{$T}())
       end
bar (generic function with 1 method)

julia> @code_warntype bar(Foo{(1,2)}())
Variables:
  x::Foo{(1,2)}

Body:
  begin
      return ((top(apply_type))(Foo,(1,2))::Type{_<:Foo{T}})()::Foo{T}
  end::Foo{T}

Cc @Keno. This seems like an extension of #9016 (which makes the inference for the normal function equivalent concrete).

[simonster]

It seems like this works:

julia> stagedfunction bar{T}(x::Foo{T})
                return :($(Foo{T})())
              end
bar (generic function with 1 method)

julia> @code_warntype bar(Foo{(1,2)}())
Variables:
  x::Foo{(1,2)}

Body:
  begin 
      return $(Expr(:new, Foo{(1,2)}))
  end::Foo{(1,2)}

although it would be better if what you wrote worked as well.

[mbauman]

Thanks, Simon, that workaround does the trick perfectly.

[mbauman]

My original example is now fixed.