[bgavran]

Indeed, the author seems to have a misunderstanding about both undecidability and complexity as they pertain to dependent types.

Dependent types do not add complexity to our system, they reveal it.

Case in point: here is a fully dependently-typed tensor processing framework written in Idris, which I believe matches most of the desiderata of his talk, capturing even a generalisation of arrays via Naperian functors that is mentioned at one point.

https://github.com/bgavran/TypeSafe_Tensors