Date: Feb 14, 2003, 1.30 pm
Place: Columbia
I will present a formulation of binding-time analysis as a type-inference problem for a staged programming language. The type inference emits arithmetic constraints which are then simplified by a simple homegrown algorithm. More sophisticated algorithms surely exist.
The generated constraint sets may admit multiple solutions (annotated expressions), but I claim there is always a "best" solution in the sense that all sub-expressions are executed in the earliest stage possible. I will give evidence supporting this claim.
[ This is a work with Tim Sheard and Peter Thiemann ]