The current Starling type system, which insists on expressions being partitioned into strongly typed classes of sub-expression, may become very unwieldy as we add n-ary arrays. We need to consider whether we need to make Starling's type system less internally strong (so it allows badly typed expressions in the middle of the Starling proof process, and type-checks them on-demand when it needs to know the type of them).
The current Starling type system, which insists on expressions being partitioned into strongly typed classes of sub-expression, may become very unwieldy as we add n-ary arrays. We need to consider whether we need to make Starling's type system less internally strong (so it allows badly typed expressions in the middle of the Starling proof process, and type-checks them on-demand when it needs to know the type of them).