[POPL'26] Local Contextual Type Inference

The podcast delves into local contextual type inference (LCTI), a refinement of local type inference using contextual typing ideas, offering a modular design and closer alignment with canonical systems like System F. Xu Xue explains that LCTI removes practical restrictions of local type inference and enhances locality, emphasizing its mechanization in Agda and Rockprover. The discussion covers how LCTI uses collected contextual information to ensure safe guesses for a decidable system, and introduces matching subtyping with matching variables to track solved instantiations. Instantiability is presented as occurrence checking guided by masks in a left-to-right order, interleaving inference and checking of arguments. The podcast also touches on future extensions to FSUB and connections to mode logic programming.