(* Title: Tools/Argo/argo_core.ML Author: Sascha Boehme Core of the Argo theorem prover implementing the DPLL(T) loop. The implementation is based on: Harald Ganzinger, George Hagen, Robert Nieuwenhuis, Albert Oliveras, Cesare Tinelli. DPLL(T): Fast decision procedures. In Lecture Notes in Computer Science, volume 3114, pages 175-188. Springer, 2004. Robert Nieuwenhuis, Albert Oliveras, Cesare Tinelli. Solving SAT and SAT modulo theories: From an abstract Davis-Putnam-Logemann-Loveland procedure to DPLL(T). In Journal of the ACM, volume 53(6), pages 937-977. ACM, 2006. *) signature ARGO_CORE = sig (* context *) type context val context: context (* enriching the context *) val identify: Argo_Term.item -> context -> Argo_Term.identified * context val add_atom: Argo_Term.item -> context -> Argo_Term.identified * context val add_axiom: Argo_Cls.clause -> context -> context (* DPLL(T) loop *) val run: context -> context (* raises Argo_Proof.UNSAT *) (* model *) val model_of: context -> string * Argo_Expr.typ -> bool option end structure Argo_Core: ARGO_CORE = struct (* context *) type context = { terms: Argo_Term.context, (* the term context to identify equal expressions *) iter: int, (* the current iteration of the search *) cdcl: Argo_Cdcl.context, (* the context of the propositional solver *) thy: Argo_Thy.context} (* the context of the theory solver *) fun mk_context terms iter cdcl thy: context = {terms=terms, iter=iter, cdcl=cdcl, thy=thy} val context = mk_context Argo_Term.context 1 Argo_Cdcl.context Argo_Thy.context fun backjump levels = funpow levels Argo_Thy.backtrack (* enriching the context *) fun identify i ({terms, iter, cdcl, thy}: context) = let val (identified, terms) = Argo_Term.identify_item i terms in (identified, mk_context terms iter cdcl thy) end fun add_atom i cx = (case identify i cx of known as (Argo_Term.Known _, _) => known | (atom as Argo_Term.New t, {terms, iter, cdcl, thy}: context) => (case (Argo_Cdcl.add_atom t cdcl, Argo_Thy.add_atom t thy) of (cdcl, (NONE, thy)) => (atom, mk_context terms iter cdcl thy) | (cdcl, (SOME lit, thy)) => (case Argo_Cdcl.assume Argo_Thy.explain lit cdcl thy of (NONE, cdcl, thy) => (atom, mk_context terms iter cdcl thy) | (SOME _, _, _) => raise Fail "bad conflict with new atom"))) fun add_axiom cls ({terms, iter, cdcl, thy}: context) = let val (levels, cdcl) = Argo_Cdcl.add_axiom cls cdcl in mk_context terms iter cdcl (backjump levels thy) end (* DPLL(T) loop: CDCL with theories *) datatype implications = None | Implications | Conflict of Argo_Cls.clause fun cdcl_assume [] cdcl thy = (NONE, cdcl, thy) | cdcl_assume (lit :: lits) cdcl thy = (* assume an assignment deduced by the theory solver *) (case Argo_Cdcl.assume Argo_Thy.explain lit cdcl thy of (NONE, cdcl, thy) => cdcl_assume lits cdcl thy | (SOME cls, cdcl, thy) => (SOME cls, cdcl, thy)) fun theory_deduce _ (conflict as (Conflict _, _, _)) = conflict | theory_deduce f (result, cdcl, thy) = (case f thy of (Argo_Common.Implied [], thy) => (result, cdcl, thy) | (Argo_Common.Implied lits, thy) => (* turn all implications of the theory solver into propositional assignments *) (case cdcl_assume lits cdcl thy of (NONE, cdcl, thy) => (Implications, cdcl, thy) | (SOME cls, cdcl, thy) => (Conflict cls, cdcl, thy)) | (Argo_Common.Conflict cls, thy) => (Conflict cls, cdcl, thy)) fun theory_assume [] cdcl thy = (None, cdcl, thy) | theory_assume lps cdcl thy = (None, cdcl, thy) (* propagate all propositional implications to the theory solver *) |> fold (theory_deduce o Argo_Thy.assume) lps (* check the consistency of the theory model *) |> theory_deduce Argo_Thy.check fun search limit cdcl thy = (* collect all propositional implications of the last assignments *) (case Argo_Cdcl.propagate cdcl of (Argo_Common.Implied lps, cdcl) => (* propagate all propositional implications to the theory solver *) (case theory_assume lps cdcl thy of (None, cdcl, thy) => (* stop searching if the conflict limit has been exceeded *) if limit <= 0 then (false, cdcl, thy) else (* no further propositional assignments, choose a value for the next unassigned atom *) (case Argo_Cdcl.decide cdcl of NONE => (true, cdcl, thy) (* the context is satisfiable *) | SOME cdcl => search limit cdcl (Argo_Thy.add_level thy)) | (Implications, cdcl, thy) => search limit cdcl thy | (Conflict ([], p), _, _) => Argo_Proof.unsat p | (Conflict cls, cdcl, thy) => analyze cls limit cdcl thy) | (Argo_Common.Conflict cls, cdcl) => analyze cls limit cdcl thy) and analyze cls limit cdcl thy = (* analyze the conflict, probably using lazy explanations from the theory solver *) let val (levels, cdcl, thy) = Argo_Cdcl.analyze Argo_Thy.explain cls cdcl thy in search (limit - 1) cdcl (backjump levels thy) end fun luby_number i = let fun mult p = if p < i + 1 then mult (2 * p) else p val p = mult 2 in if i = p - 1 then p div 2 else luby_number (i - (p div 2) + 1) end fun next_restart_limit iter = 100 * luby_number iter fun loop iter cdcl thy = (* perform a limited search that is stopped after a certain number of conflicts *) (case search (next_restart_limit iter) cdcl thy of (true, cdcl, thy) => (iter + 1, cdcl, thy) | (false, cdcl, thy) => (* restart the solvers to avoid that they get stuck in a fruitless search *) let val (levels, cdcl) = Argo_Cdcl.restart cdcl in loop (iter + 1) cdcl (backjump levels thy) end) fun run ({terms, iter, cdcl, thy}: context) = let val (iter, cdcl, thy) = loop iter cdcl (Argo_Thy.prepare thy) in mk_context terms iter cdcl thy end (* model *) fun model_of ({terms, cdcl, ...}: context) c = (case Argo_Term.identify_item (Argo_Term.Expr (Argo_Expr.E (Argo_Expr.Con c, []))) terms of (Argo_Term.Known t, _) => Argo_Cdcl.assignment_of cdcl (Argo_Lit.Pos t) | (Argo_Term.New _, _) => NONE) end