sarsko / creusat Goto Github PK

I feel like it might be a good idea to experiment with using smallvec in the definition of Clause.

Alternately, I wonder if it is possible define a custom unsized type in Rust so that we can layout the clause literals inline with a header.

Could you add what should be cited in the readme? Here is the bibtex entry I currently use, but it might not be optimal:

@mastersthesis{creusat,
  author       = {Skotåm, Sarek Høverstad},
  title	       = {{CreuSAT}, Using {Rust} and {Creusot} to create the
                  world’s fastest deductively verified {SAT} solver},
  school       = {University of Oslo},
  year	       = {2022},
  url	       =
                  {https://www.duo.uio.no/handle/10852/96757},
}

Thank you very much

Running cargo make p crashes for me since it uses a hardcoded path for the why3 ide script. It would be nice to include the script locally. This may also require copying prelude/ from Creusot (until we can figure out how to distribute it with the binary / crate).

It would be a good idea to set up a simple CI script that ensures the proofs are not obsolete (without even running them). This would avoid much risk of breakage.

Is it expected that the fast and slow heuristic have the same initial value in CreuSAT?

    pub fn new(f: &Formula) -> Solver {
        Solver {
            num_lemmas: 0,
            max_lemmas: 2000,
            num_conflicts: 0,
            initial_len: f.clauses.len(),
            inc_reduce_db: 300,
            fast: 16777216, // 1 << 24
            slow: 16777216, // 1 << 24
            perm_diff: vec::from_elem(0, f.num_vars),
        }
    }

That seems odd (and a good way to never restart actually).

Excellent project and thesis! Although I'm not finished reading it :)

I think there's a minor typo for Lit struct in section 3.3. You define Lit as having a polarity bool. But this is subsequently changed to value in later snippets. Changing it value should fix things.

I wonder how much time could be gained by having all formula data stored using arenas. We should investigate using bumpalo for this.

Rather self-explanatory, but taking the time to eliminate all runtime checks would be very beneficial to the proof itself.

Thank you! I love any progress in the Rust SAT space. varisat has an api witch I have used extensively for testing libraries I work on. Would you be interested in adding an Rust API interface to CreuSAT?