[penteract]

Your other points are more relevant to the content of the article, but point 2. relates the practical consequences of undecidable type-checking, so I'll reply to that.

I don't have a problem with compile time code execution potentially not terminating, since it's clear to the programmer why that may happen. However, conventional type checking/inference is more like solving a system of constraints, and the programmer should understand what the constraints mean, but not need to know how the constraint solver (type checker) operates. If it's undecidable, that means there is a program that a programmer knows should type check, but the implementation won't be happy with; ruining the programmer's blissful ignorance of the internals.