Properties of programs can be formulated using various techniques: dataflow analysis, abstract interpretation and type-like inference systems. This paper reconstructs strictness analysis (establishing when function parameters are evaluated in a lazy language) as a dataflow analysis, initially at first order, then at higher order by expressing the dataflow properties as an effect system. Strictness properties so expressed give a clearer operational understanding and enable a range of additional optimisations including implicational strictness. At first order strictness effects have the expected principality properties (best-property inference) and can be computed simply; without polymorphic effects principality is lost at higher order. However, adding both polymorphic effects and polymorphic type instantiation to restore principality exposes novel issues.

This is joint work with Tom Schrijvers.