A Case for Tactics with (Limited) Side Effects

With the notable exception of abstract, Coq’s standard collection of tactics are free from side effects. In this talk, I propose to make the case that additional tactics with limited side effects that are “local” to a given proof can have a considerable impact on performance, presenting as evidence a set of ML tactics that have been developed as part of the Fiat deductive synthesis framework. In particular, I’ll show how a version of abstract that adds transparent definitions to the global context and tactics that manipulate hint databases can shrink both the memory usage and processing time of a proof script by an order of magnitude.