Closed c-cube closed 5 years ago
Note: proofs are not stored properly in some cases (hole8.cnf
for instance). I suspect that 0-levels resolutions are wrongly skipped when the unsat conflict comes from propagation (rather than learning the empty clause). Please don't merge yet, but discussions welcome ready to merge!
I changed the implementation to not used sorted lists and
merge
, but mutable atoms.A stronger improvement could be gained by staying at the hyper-resolution level even during checking (possibly exposing a per-step expansion into pure resolution for backends that need it). See for example mc2's proof which computes pivot for each step of hyper-res, but doesn't store the (possibly big) intermediate clauses.