[PLDI'25] Practical Type Inference with Levels

The podcast delves into the formalization of levels, an implementation technique used in type inference, to bridge the gap between theory and practice. It addresses let generalization, local data types, and higher-rank polymorphism within a level-based type system framework. The discussion covers the challenges and solutions in ensuring the soundness and completeness of level-based systems, including the use of mechanized proofs in the ROC theorem prover. Examples, such as the binary tree data structure and lambda expressions with polymorphic types, illustrate the concepts. The presenter also shares preliminary evaluation results from implementing level-based type inference for the COCA programming language, demonstrating performance improvements in generalization. Questions from the audience cover mechanization plans and benchmark choices.