Skip to content

Latest commit

 

History

History

Interfaces

README.md

Interfaces for the EvmYul and KEVM models

This folder contains interfaces to reason with the EvmYul and KEVM models.

File description

  • EvmYulInterface: Interface for the EvmYul model
  • FuncInterface: Interface for dealing with generated KEVM functions
  • GasInterface: Interface for the «_<_>_SCHEDULE_Int_ScheduleConst_Schedule» and «_<<_>>_SCHEDULE_Bool_ScheduleFlag_Schedule» functions
  • Tactics: Tactics to help with the GasInterface.lean proofs
  • Axioms: Axiomatization of the state mapping

Axiomatic mapping

The file Axioms.lean declares the exsistence of functions that convert KEVM maps into EvmYul ones.

The reason for this axiomatization is twofold.

  1. Unimplemented maps in the K Lean 4 backend: K maps are currently undefined in the Lean 4 generated code. We still have to come up with a suitable-enough implementation of K maps that preserve the original semantics.

    Note that this is tricky given that K maps have to be defined inside of a mutual inductive block, which limits the options available to prevent crashing with Lean's positivity checker. A clear disadvantage of this is that we cannot (apparently) use FinMap as our map implementation.

  2. Non reasoning-friendly maps in the EvmYul model: Currently, the EvmYul model is more geared towards execution than reasoning. This means that the map representation is not the most optimal for reasoning, and a more reasoning-friendly interface could be provided in the future. This is an argument in favor of not commiting (yet) to a K map representation tailored to play well with the current EvmYul maps, but this argument could change in the future.

Axioms as pre-theorems

Given that we're axiomatizing part of the state translation, properties on the axiomatic parts need to be provided. How the axiomatic parts of the state mapping are needed to behave is laid out in the Axioms namespace at the end of the StateMap.lean file.

These axioms establish the correspondence between the mapped data and the functions in the EvmYul map implementation. Once we have a clear implementation of K maps in Lean, all these axioms should become theorems.