twee-2.6.1: executable/SequentialMain.hs
{-# LANGUAGE CPP, RecordWildCards, FlexibleInstances, PatternGuards, DeriveAnyClass, RankNTypes, ApplicativeDo #-}
{-# OPTIONS_GHC -flate-specialise #-}
module SequentialMain(main) where
import Control.Monad
import Data.Char
import Data.Either
import Twee hiding (message)
import Twee.Base hiding (char, lookup, vars, ground)
import Twee.Rule(lhs, rhs, unorient)
import Twee.Equation
import qualified Twee.Proof as Proof
import Twee.Proof hiding (Config, defaultConfig)
import qualified Twee.Join as Join
import Twee.Utils
import qualified Twee.CP as CP
import Data.Ord
import qualified Data.Map.Strict as Map
import qualified Twee.KBO as KBO
import Data.List.Split
import Data.List
import Data.Maybe
import Jukebox.Options
import Jukebox.Toolbox
import Jukebox.Name hiding (lhs, rhs, label)
import qualified Jukebox.Form as Jukebox
import Jukebox.Form hiding ((:=:), Var, Symbolic(..), Term, Axiom, size, Subst, subst)
import Jukebox.Tools.EncodeTypes
import Jukebox.TPTP.Print
import Jukebox.Tools.HornToUnit
import qualified Data.IntMap.Strict as IntMap
import System.IO
import System.Exit
import qualified Data.Set as Set
import qualified Data.Intern as Intern
import System.Console.ANSI
import Data.Symbol
import Twee.Profile
data MainFlags =
MainFlags {
flags_proof :: Bool,
flags_proof_on_saturation :: Bool,
flags_trace :: Maybe (String, String),
flags_formal_proof :: Bool,
flags_explain_encoding :: Bool,
flags_flip_ordering :: Bool,
flags_give_up_on_saturation :: Bool,
flags_flatten_goals :: Bool,
flags_flatten_nonground :: Bool,
flags_flatten_goals_lightly :: Bool,
flags_flatten_all :: Bool,
flags_flatten_regeneralise :: Bool,
flags_eliminate :: [String],
flags_backwards_goal :: Int,
flags_flatten_backwards_goal :: Int,
flags_equals_transformation :: Bool,
flags_distributivity_heuristic :: Bool,
flags_kbo_weight0 :: Bool,
flags_kbo_weight0_unary :: Bool,
flags_goal_heuristic :: Bool }
parseMainFlags :: OptionParser MainFlags
parseMainFlags = do
let argModule = arg "<module>" "expected a Prolog module name" Just
flags_proof <-
inGroup "Output options" $
bool "proof" ["Produce proofs (on by default)."]
True
flags_proof_on_saturation <-
expert $
inGroup "Output options" $
bool "proof-on-saturation" ["Produce proofs of all rewrite rules on saturation (off by default)."]
False
flags_trace <-
expert $
inGroup "Output options" $
flag "trace"
["Write a Prolog-format execution trace to this file (off by default)."]
Nothing ((\x y -> Just (x, y)) <$> argFile <*> argModule)
flags_formal_proof <-
expert $
inGroup "Output options" $
bool "formal-proof" ["Print proof as formal TSTP derivation (requires --tstp; off by default)."] False
flags_explain_encoding <-
expert $
inGroup "Output options" $
bool "explain-encoding" ["In CASC mode, explain the conditional encoding (off by default)."] False
flags_flip_ordering <-
expert $
inGroup "Term order options" $
bool "flip-ordering" ["Make more common function symbols smaller (off by default)."] False
flags_kbo_weight0 <-
expert $
inGroup "Term order options" $
bool "kbo-weight0" ["Give functions of arity >= 2 a weight of 0."] False
flags_kbo_weight0_unary <-
expert $
inGroup "Term order options" $
bool "kbo-weight0-unary" ["Give one function of arity 1 a weight of 0."] True
flags_give_up_on_saturation <-
expert $
inGroup "Output options" $
bool "give-up-on-saturation" ["Report SZS status GiveUp rather than Unsatisfiable on saturation (off by default)."] False
flags_flatten_goals <-
expert $
inGroup "Completion heuristics" $
bool "flatten-goal" ["Flatten goal by adding new axioms (on by default)."] True
flags_flatten_nonground <-
expert $
inGroup "Completion heuristics" $
bool "flatten-nonground" ["Flatten even non-ground clauses (off by default)."] False
flags_flatten_goals_lightly <-
expert $
inGroup "Completion heuristics" $
bool "flatten-goal-lightly" ["Flatten goal non-recursively by adding new axioms (off by default)."] False
flags_flatten_all <-
expert $
inGroup "Completion heuristics" $
bool "flatten" ["Flatten all clauses by adding new axioms (off by default)."] False
flags_flatten_regeneralise <-
expert $
inGroup "Completion heuristics" $
bool "flatten-regeneralise" ["Regeneralise rules involving flattened goal terms (off by default)."] False
flags_backwards_goal <-
expert $
inGroup "Completion heuristics" $
flag "backwards-goal" ["Try rewriting backwards from the goal this many times (0 by default)."] 0 argNum
flags_flatten_backwards_goal <-
expert $
inGroup "Completion heuristics" $
flag "flatten-backwards-goal" ["Try rewriting backwards from the goal this many times when flattening (0 by default)."] 0 argNum
flags_equals_transformation <-
expert $
inGroup "Completion heuristics" $
bool "equals-transformation" ["Apply the 'equals transformation' even to ground goals (off by default)."] False
flags_distributivity_heuristic <-
expert $
inGroup "Completion heuristics" $
bool "distributivity-heuristic" ["Treat distributive operators specially (off by default)."] False
flags_goal_heuristic <-
expert $
inGroup "Completion heuristics" $
bool "goal-heuristic" ["Use the CP weighting heuristic from Anantharaman and Andrianarievelo (off by default)."] False
flags_eliminate <-
inGroup "Proof presentation" $
concat <$>
manyFlags "eliminate"
["Treat these axioms as definitions and eliminate them from the proof.",
"The axiom must have the shape f(x1...xn) = t, where x1...xn are",
"distinct variables. The term f must not otherwise appear in the problem!",
"This is not checked."]
(splitOn "," <$> arg "<axioms>" "expected a list of axiom names" Just)
return MainFlags{..}
parseConfig :: OptionParser (Config Constant)
parseConfig = do
cfg_accept_term <-
let checkSize n t = KBO.size (t :: Term Constant) <= n in
inGroup "Resource limits" $
flag "max-term-size" ["Discard rewrite rules whose left-hand side is bigger than this limit (unlimited by default)."] Nothing (Just <$> checkSize <$> argNum)
cfg_max_critical_pairs <-
inGroup "Resource limits" $
flag "max-cps" ["Give up after considering this many critical pairs (unlimited by default)."] maxBound argNum
cfg_max_cp_depth <-
inGroup "Resource limits" $
flag "max-cp-depth" ["Only consider critical pairs up to this depth (unlimited by default)."] maxBound argNum
cfg_max_rules <-
inGroup "Resource limits" $
flag "max-rules" ["Give up after generating this many rules (unlimited by default)."] maxBound argNum
cfg_max_time <-
inGroup "Resource limits" $
flag "max-time" ["Give up after running for this long in seconds (unlimited by default)."] Nothing (Just <$> argNum)
cfg_simplify <-
expert $
inGroup "Completion heuristics" $
bool "simplify"
["Simplify rewrite rules with respect to one another (on by default)."]
True
cfg_renormalise_percent <-
expert $
inGroup "Completion heuristics" $
defaultFlag "normalise-queue-percent" "Percent of time spent renormalising queued critical pairs" cfg_renormalise_percent argNum
cfg_cp_sample_size <-
expert $
inGroup "Completion heuristics" $
defaultFlag "cp-sample-size" "Size of random CP sample used to trigger renormalisation" cfg_cp_sample_size argNum
cfg_renormalise_threshold <-
expert $
inGroup "Completion heuristics" $
defaultFlag "cp-renormalise-threshold" "Trigger renormalisation when this percentage of CPs can be simplified" cfg_renormalise_threshold argNum
cfg_set_join_goals <-
expert $
inGroup "Critical pair joining heuristics" $
bool "set-join-goals"
["Compute all normal forms when joining goal terms (on by default)."]
True
cfg_always_simplify <-
expert $
inGroup "Debugging options" $
bool "always-simplify"
["Interreduce rules after every step."]
False
cfg_complete_subsets <-
expert $
inGroup "Critical pair joining heuristics" $
bool "complete-subsets"
["Identify and exploit complete subsets of the axioms in joining (off by default)."]
False
let cfg_score_cp = undefined -- filled in later, in runTwee
cfg_join <- do
cfg_ground_join <-
expert $
inGroup "Critical pair joining heuristics" $
bool "ground-joining"
["Test terms for ground joinability (on by default)."]
True
cfg_use_connectedness_standalone <-
expert $
inGroup "Critical pair joining heuristics" $
bool "connectedness"
["Test terms for subconnectedness, as a separate check (on by default)."]
True
cfg_use_connectedness_in_ground_joining <-
expert $
inGroup "Critical pair joining heuristics" $
bool "ground-connectedness"
["Test terms for subconnectedness, as part of ground joinability testing (off by default)."]
False
cfg_set_join <-
expert $
inGroup "Critical pair joining heuristics" $
bool "set-join"
["Compute all normal forms when joining critical pairs (off by default)."]
False
cfg_ground_join_limit <-
inGroup "Critical pair joining heuristics" $
flag "ground-joining-limit" ["Assume not ground joinable after considering this many orderings (unlimited by default)."] maxBound argNum
cfg_ground_join_incomplete_limit <-
inGroup "Critical pair joining heuristics" $
flag "ground-joining-incomplete-limit" ["Assume ground joinable after considering this many orderings (unlimited by default)."] maxBound argNum
return Join.Config{..}
cfg_proof_presentation <- do
cfg_all_lemmas <-
inGroup "Proof presentation" $
bool "all-lemmas"
["Produce a proof with one lemma for each critical pair (off by default)."]
False
cfg_no_lemmas <-
inGroup "Proof presentation" $
bool "no-lemmas"
["Produce a proof with no lemmas (off by default).",
"May lead to exponentially large proofs."]
False
cfg_ground_proof <-
inGroup "Proof presentation" $
bool "ground-proof"
["Produce a ground proof (off by default).",
"May lead to exponentially large proofs."]
False
cfg_show_instances <-
inGroup "Proof presentation" $
bool "show-instances"
["Show which instance of a lemma or axiom each rewrite step uses (off by default)."]
False
cfg_use_colour <-
let
colourFlag =
inGroup "Proof presentation" $
primFlag "(no-)colour"
["Produce output in colour (on by default if writing output to a terminal)."]
(`elem` map fst colourFlags)
(\_ y -> return y)
Nothing
(pure (`lookup` colourFlags))
colourFlags = [("--colour", True), ("--no-colour", False),
("--color", True), ("--no-color", False)]
colourSupported =
liftM2 (&&) (hSupportsANSIColor stdout)
(return (setSGRCode [] /= "")) -- Check for Windows terminal not supporting ANSI
in fromMaybe <$> io colourSupported <*> colourFlag
cfg_show_uses_of_axioms <-
let interpret xss ax = axiom_name ax `elem` xss || "all" `elem` xss in
inGroup "Proof presentation" $
interpret <$>
concat <$>
manyFlags "show-uses-of"
["Show which instances of the given axioms were needed (none by default).",
"Separate multiple axiom names with commas.",
"Use --show-uses-of all to show uses of all axioms."]
(splitOn "," <$> arg "<axioms>" "expected a list of axiom names" Just)
cfg_show_peaks <-
inGroup "Proof presentation" $
bool "show-peaks"
["Show peak terms in a proof (off by default)."]
False
cfg_eliminate_existentials_coding <-
inGroup "Proof presentation" $
bool "eliminate-existentials-coding"
["Eliminate $equals from proofs (on by default)."]
True
cfg_show_subterms <-
inGroup "Proof presentation" $
bool "show-subterms"
["Show which subterm is rewritten at each step (off by default)."]
False
return Proof.Config{..}
let cfg_eliminate_axioms = [] -- filled in later
cfg_random_mode <-
expert $
inGroup "Completion heuristics" $
bool "random-mode"
["Use random testing to find suitable CPs (doesn't work yet!) (off by default)."]
False
cfg_random_mode_goal_directed <-
expert $
inGroup "Completion heuristics" $
bool "random-mode-goal-directed"
["Use goal-direction in --random-mode (off by default)."]
False
cfg_random_mode_simple <-
expert $
inGroup "Completion heuristics" $
bool "random-mode-simple"
["Use simple version of --random-mode (off by default)."]
False
cfg_random_mode_best_of <-
inGroup "Completion heuristics" $
defaultFlag "random-mode-best-of" "Generate this many critical pairs at a time and pick the best one" cfg_random_mode_best_of argNum
cfg_always_complete <-
inGroup "Input and clausifier options" $
bool "complete"
["Don't stop until the rewrite system is confluent"]
False
cfg_hint_skel_cost <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "hint-skel-cost" "Size of hint skeletons" cfg_hint_skel_cost argNum
cfg_hint_skel_factor <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "hint-skel-factor" "Size factor of hint skeletons" cfg_hint_skel_factor argNum
cfg_print_score <-
expert $
inGroup "Output options" $
bool "print-score" ["Print score of each generated rule (off by default)."] False
return Config{..}
where
defaultFlag :: Show a => String -> String -> (Config Constant -> a) -> ArgParser a -> OptionParser a
defaultFlag name desc field parser =
flag name [desc ++ " (" ++ show def ++ " by default)."] def parser
where
def = field defaultConfig
parseCPConfig :: OptionParser CP.Config
parseCPConfig = do
cfg_lhsweight <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "lhs-weight" "Weight given to LHS of critical pair" CP.cfg_lhsweight argNum
cfg_rhsweight <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "rhs-weight" "Weight given to RHS of critical pair" CP.cfg_rhsweight argNum
cfg_funweight <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "fun-weight" "Weight given to function symbols" CP.cfg_funweight argNum
cfg_varweight <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "var-weight" "Weight given to variable symbols" CP.cfg_varweight argNum
cfg_depthweight <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "depth-weight" "Weight given to critical pair depth" CP.cfg_depthweight argNum
cfg_dupcost <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "dup-cost" "Cost of duplicate subterms" CP.cfg_dupcost argNum
cfg_dupfactor <-
expert $
inGroup "Critical pair weighting heuristics" $
defaultFlag "dup-factor" "Size factor of duplicate subterms" CP.cfg_dupfactor argNum
cfg_resonance <-
expert $
inGroup "Critical pair weighting heuristics" $
bool "resonance" ["Interpret hints as resonators by only allowing substitutions which map variables to variables (off by default)."] False
return CP.Config{..}
where
defaultFlag name desc field parser =
flag name [desc ++ " (" ++ show def ++ " by default)."] def parser
where
def = field CP.defaultConfig
parsePrecedence :: OptionParser [String]
parsePrecedence =
expert $
inGroup "Term order options" $
fmap (splitOn ",")
(flag "precedence" ["List of functions in descending order of precedence."] [] (arg "<function>" "expected a function name" Just))
data Constant =
Minimal |
Skolem Int |
Constant {
con_prec :: {-# UNPACK #-} !Precedence,
con_id :: {-# UNPACK #-} !Jukebox.Function,
con_arity :: {-# UNPACK #-} !Int,
con_size :: !Integer,
con_weight :: !Integer,
con_bonus :: !Bool }
deriving (Eq, Ord)
data Precedence = Precedence !Bool !Bool !Bool !(Maybe Int) !Int
deriving (Eq, Ord)
instance KBO.Sized Constant where
size Minimal = 1
size Skolem{} = 1
size Constant{..} = con_size
instance KBO.Weighted Constant where
argWeight Minimal = 1
argWeight Skolem{} = 1
argWeight Constant{..} = con_weight
instance Pretty Constant where
pPrint Minimal = text "?"
pPrint (Skolem n) = text ("sk" ++ show n)
pPrint Constant{..} = text (removePostfix (base con_id))
where
removePostfix ('_':x:xs) | con_arity == 1 = x:xs
removePostfix xs = xs
instance PrettyTerm Constant where
termStyle Minimal = uncurried
termStyle Skolem{} = uncurried
termStyle Constant{..}
| hasLabel "type_tag" con_id = invisible
| "_" `isPrefixOf` base con_id && con_arity == 1 = postfix
| any isAlphaNum (base con_id) = uncurried
| otherwise =
case con_arity of
1 -> prefix
2 -> infixStyle 5
_ -> uncurried
instance Minimal Constant where
minimal = Sym Minimal
skolem = Sym . Skolem
instance Ordered Constant where
lessEq t u = KBO.lessEq t u
lessIn model t u = KBO.lessIn model t u
lessEqSkolem t u = KBO.lessEqSkolem t u
instance EqualsBonus Constant where
hasEqualsBonus Minimal = False
hasEqualsBonus Skolem{} = False
hasEqualsBonus c = con_bonus c
isEquals Minimal = False
isEquals Skolem{} = False
isEquals c = SequentialMain.isEquals (con_id c)
isTrue Minimal = False
isTrue Skolem{} = False
isTrue c = SequentialMain.isTrue (con_id c)
isFalse Minimal = False
isFalse Skolem{} = False
isFalse c = SequentialMain.isFalse (con_id c)
data TweeContext =
TweeContext {
ctx_var :: Jukebox.Variable,
ctx_minimal :: Jukebox.Function,
ctx_true :: Jukebox.Function,
ctx_false :: Jukebox.Function,
ctx_equals :: Jukebox.Function,
ctx_type :: Type }
-- Convert back and forth between Twee and Jukebox.
tweeConstant :: MainFlags -> HornFlags -> TweeContext -> Precedence -> Jukebox.Function -> Constant
tweeConstant MainFlags{..} flags TweeContext{..} prec fun
| fun == ctx_minimal = Minimal
| otherwise =
Constant {
con_prec = prec,
con_id = fun,
con_arity = Jukebox.arity fun,
con_size = if flags_kbo_weight0 && Jukebox.arity fun >= 2 then 0 else if flags_kbo_weight0_unary && isInv then 0 else 1,
con_weight = 1,
con_bonus = bonus fun }
where
bonus fun =
(isIfeq fun && encoding flags /= Asymmetric2) ||
SequentialMain.isEquals fun
isInv =
case prec of
Precedence _ x _ _ _ -> x
isType :: Jukebox.Function -> Bool
isType fun =
hasLabel "type_tag" (name fun) && Jukebox.arity fun == 1
isIfeq :: Jukebox.Function -> Bool
isIfeq fun =
hasLabel "ifeq" (name fun)
isEquals :: Jukebox.Function -> Bool
isEquals fun =
hasLabel "equals" (name fun) && Jukebox.arity fun == 2
isTrue :: Jukebox.Function -> Bool
isTrue fun =
hasLabel "true" (name fun) && Jukebox.arity fun == 0
isFalse :: Jukebox.Function -> Bool
isFalse fun =
hasLabel "false" (name fun) && Jukebox.arity fun == 0
jukeboxFunction :: TweeContext -> Constant -> Jukebox.Function
jukeboxFunction _ Constant{..} = con_id
jukeboxFunction TweeContext{..} Minimal = ctx_minimal
tweeTerm :: MainFlags -> HornFlags -> TweeContext -> (Jukebox.Function -> Precedence) -> Jukebox.Term -> Term Constant
tweeTerm flags horn ctx prec t = build (tm t)
where
tm (Jukebox.Var (x ::: _)) =
var (V (Intern.symId (Intern.intern x)))
tm (f :@: ts) =
app (Sym (tweeConstant flags horn ctx (prec f) f)) (map tm ts)
jukeboxTerm :: TweeContext -> Term Constant -> Jukebox.Term
jukeboxTerm TweeContext{..} (Var (V x)) =
Jukebox.Var (Unique (fromIntegral x) (intern "X") Nothing defaultRenamer ::: ctx_type)
jukeboxTerm ctx@TweeContext{..} (App (Sym f) t) =
jukeboxFunction ctx f :@: map (jukeboxTerm ctx) ts
where
ts = unpack t
makeContext :: [Jukebox.Term] -> Problem Clause -> TweeContext
makeContext hints prob = run (hints, prob) $ \(_, prob) -> do
let
ty =
case types' prob of
[] -> indType
[ty] -> ty
var <- newSymbol "X" ty
minimal <- newFunction (withLabel "minimal" (name "constant")) [] ty
true <- newFunction (withLabel "true" (name "true")) [] ty
false <- newFunction (withLabel "false" (name "false")) [] ty
equals <- newFunction (withLabel "equals" (name "equals")) [ty, ty] ty
return TweeContext {
ctx_var = var,
ctx_minimal = minimal,
ctx_true = true,
ctx_false = false,
ctx_equals = equals,
ctx_type = ty }
flattenGoals :: Int -> Bool -> Bool -> Bool -> [Jukebox.Term] -> Problem Clause -> Problem Clause
flattenGoals backwardsGoal flattenNonGround flattenAll full hints prob =
run (hints, prob) $ \(_, prob) -> do
let ts = usort $ extraTerms prob
cs <- mapM define ts
return (prob ++ cs)
where
extraTerms prob = concatMap (input prob) prob
input prob Input{what = Clause (Bind _ [Neg (x Jukebox.:=: y)])} =
concatMap term (backwards backwardsGoal prob x) ++
concatMap term (backwards backwardsGoal prob y)
input _ Input{what = Clause (Bind _ [Pos (x Jukebox.:=: y)])}
| flattenAll = term x ++ term y
input _ _ = []
term t@(_f :@: ts) =
[ t
| ground t || flattenNonGround,
not (all isVar ts) || usort ts /= sort ts ] ++
if full then concatMap term ts else []
term _ = []
isVar (Jukebox.Var _) = True
isVar _ = False
define (f :@: ts) = do
name <- newName f
let vs = Jukebox.vars ts
g = name ::: FunType (map typ vs) (typ f)
c = clause [Pos (g :@: map Jukebox.Var vs Jukebox.:=: f :@: ts)]
return Input{ident = Nothing, tag = "flattening", kind = Jukebox.Ax Definition,
what = c, source = Unknown }
backwards 0 _ t = [t]
backwards n cs t =
t:
[ v
| Input{what = Clause (Bind _ [Pos (x0 Jukebox.:=: y0)])} <- cs,
(x, y) <- [(x0, y0), (y0, x0)],
(s, k) <- contexts t,
sub <- maybeToList (Jukebox.match x s),
let u = k (Jukebox.subst sub y),
ground u,
v <- backwards (n-1) cs u ]
addDistributivityHeuristic :: [Jukebox.Term] -> Problem Clause -> Problem Clause
addDistributivityHeuristic hints prob =
run (hints, prob) $ \(_, prob) -> do
cs <- mapM add prob
return (prob ++ catMaybes cs)
where
add Input{what = Clause (Bind _ [Pos (t Jukebox.:=: u)])} =
case checkDistributivity t u `mplus` checkDistributivity u t of
Just (f, g, ty) -> do
name <- newName (base f ++ "_" ++ base g)
x <- Jukebox.Var <$> newSymbol "X" ty
y <- Jukebox.Var <$> newSymbol "Y" ty
z <- Jukebox.Var <$> newSymbol "Z" ty
Just <$> define name (g :@: [f :@: [x, y], z])
_ -> return Nothing
add _ = return Nothing
checkDistributivity
(f1 :@: [Jukebox.Var x1, g1 :@: [Jukebox.Var y1, Jukebox.Var z1]])
(g2 :@: [f2 :@: [Jukebox.Var x2, Jukebox.Var y2],
f3 :@: [Jukebox.Var x3, Jukebox.Var z2]])
| f1 == f2 && f2 == f3 && g1 == g2 &&
x1 == x2 && x2 == x3 && y1 == y2 && z1 == z2 =
Just (f1, g1, Jukebox.typ x1)
checkDistributivity
(f1 :@: [g1 :@: [Jukebox.Var x1, Jukebox.Var y1], Jukebox.Var z1])
(g2 :@: [f2 :@: [Jukebox.Var x2, Jukebox.Var z2],
f3 :@: [Jukebox.Var y2, Jukebox.Var z3]])
| f1 == f2 && f2 == f3 && g1 == g2 &&
x1 == x2 && y1 == y2 && z1 == z2 && z2 == z3 =
Just (f1, g1, Jukebox.typ x1)
checkDistributivity _ _ = Nothing
define name t = do
let vs = Jukebox.vars t
g = name ::: FunType (map typ vs) (typ t)
c = clause [Pos (g :@: map Jukebox.Var vs Jukebox.:=: t)]
return Input{ident = Nothing, tag = "distributivity_heuristic", kind = Jukebox.Ax Definition,
what = c, source = Unknown }
-- Encode existentials so that all goals are ground.
addNarrowing :: Bool -> TweeContext -> Problem Clause -> Problem Clause
addNarrowing alwaysNarrow TweeContext{..} prob =
unchanged ++ equalityClauses
where
prob' = [inp { ident = Just (variant "addNarrowing" [i :: Int]) } | (i, inp) <- zip [0..] prob]
(unchanged, nonGroundGoals) = partitionEithers (map f prob')
where
f inp@Input{what = Clause (Bind _ [Neg (x Jukebox.:=: y)])}
| not (ground x) || not (ground y) || alwaysNarrow =
Right (inp, (x, y))
f inp = Left inp
equalityClauses
| null nonGroundGoals = []
| otherwise =
-- Turn a != b & c != d & ...
-- into eq(a,b)=false & eq(c,d)=false & eq(X,X)=true & true!=false (esa)
-- and then extract the individual components (thm)
let
equalityLiterals =
-- true != false
("true_equals_false", Neg ((ctx_true :@:) [] Jukebox.:=: (ctx_false :@: []))):
-- eq(X,X)=true
("reflexivity", Pos (ctx_equals :@: [Jukebox.Var ctx_var, Jukebox.Var ctx_var] Jukebox.:=: (ctx_true :@: []))):
-- [eq(a,b)=false, eq(c,d)=false, ...]
[ (tag, Pos (ctx_equals :@: [x, y] Jukebox.:=: (ctx_false :@: [])))
| (Input{tag = tag}, (x, y)) <- nonGroundGoals ]
-- Equisatisfiable to the input clauses
justification =
Input {
ident = Just (name "addNarrowing2"),
tag = "new_negated_conjecture",
kind = Jukebox.Ax NegatedConjecture,
what =
let form = And (map (Literal . snd) equalityLiterals) in
ForAll (Bind (Set.fromList (vars form)) form),
source =
inference "encode_existential" "esa"
(map (fmap toForm . fst) nonGroundGoals) }
input tag form i =
Input {
ident = Just (variant "addNarrowing3" [i :: Int]),
tag = tag,
kind = Jukebox.Ax NegatedConjecture,
what = clause [form],
source =
inference "split_conjunct" "thm" [justification] }
in [input tag form i | ((tag, form), i) <- zip equalityLiterals [0..]]
data PreEquation =
PreEquation {
pre_name :: String,
pre_form :: Input Form,
pre_eqn :: (Jukebox.Term, Jukebox.Term) }
-- Split the problem into axioms and ground goals.
identifyProblem ::
TweeContext -> Problem Clause -> Either (Input Clause) ([PreEquation], [PreEquation])
identifyProblem TweeContext{..} prob =
fmap partitionEithers (mapM identify prob)
where
pre inp x =
PreEquation {
pre_name = tag inp,
pre_form = fmap toForm inp,
pre_eqn = x }
identify inp@Input{what = Clause (Bind _ [Pos (t Jukebox.:=: u)])} =
return $ Left (pre inp (t, u))
identify inp@Input{what = Clause (Bind _ [Neg (t Jukebox.:=: u)])}
| ground t && ground u =
return $ Right (pre inp (t, u))
identify inp@Input{what = Clause (Bind _ [])} =
-- The empty clause can appear after clausification if
-- the conjecture was trivial
return $ Left (pre inp (Jukebox.Var ctx_var, ctx_minimal :@: []))
identify inp = Left inp
runTwee :: GlobalFlags -> TSTPFlags -> HornFlags -> [String] -> Config Constant -> CP.Config -> MainFlags -> (IO () -> IO ()) -> [Jukebox.Term] -> Problem Clause -> IO Answer
runTwee globals (TSTPFlags tstp) horn precedence config0 cpConfig flags@MainFlags{..} later hints obligs = {-# SCC runTwee #-} do
let
-- Encode whatever needs encoding in the problem
obligs1
| flags_flatten_goals_lightly = flattenGoals flags_flatten_backwards_goal flags_flatten_nonground False False hints obligs
| flags_flatten_all = flattenGoals flags_flatten_backwards_goal flags_flatten_nonground True True hints obligs
| flags_flatten_goals = flattenGoals flags_flatten_backwards_goal flags_flatten_nonground False True hints obligs
| otherwise = obligs
obligs2
| flags_distributivity_heuristic = addDistributivityHeuristic hints obligs1
| otherwise = obligs1
ctx = makeContext hints obligs2
lowercaseSkolem x
| hasLabel "skolem" x =
withRenamer x $ \s i ->
case defaultRenamer s i of
Renaming xss xs ->
Renaming (map (map toLower) xss) (map toLower xs)
| otherwise = x
(hints', prob) = prettyNames (mapName lowercaseSkolem (hints, addNarrowing flags_equals_transformation ctx obligs2))
(unsortedAxioms0, goals0) <-
case identifyProblem ctx prob of
Left inp -> do
mapM_ (hPutStrLn stderr) [
"The problem contains the following clause, which is not a unit equality:",
indent (show (pPrintClauses [inp])),
"Twee only handles unit equality problems."]
exitWith (ExitFailure 1)
Right x -> return x
let
-- Work out a precedence for function symbols
prec c =
Precedence
(isType c)
(Just c == maxUnary)
(isNothing (elemIndex (base c) precedence))
(fmap negate (elemIndex (base c) precedence))
(maybeNegate (Map.findWithDefault 0 c funOccs))
maybeNegate = if flags_flip_ordering then negate else id
funOccs = funsOcc prob
maxUnary =
case filter (\(f, _) -> arity f == 1 && not (isType f)) (Map.toList funOccs) of
[] -> Nothing
xs -> Just (fst (maximumBy (comparing snd) xs))
-- Translate everything to Twee.
toTerm t = tweeTerm flags horn ctx prec t
toEquation (t, u) = canonicalise (toTerm t :=: toTerm u)
axiomCompare ax1 ax2
| isEquality ax1' && not (isEquality ax2') = GT
| isEquality ax2' && not (isEquality ax1') = LT
| ax1' `simplerThan` ax2' = LT
| ax2' `simplerThan` ax1' = GT
| otherwise = EQ
where
ax1' = toEquation (pre_eqn ax1)
ax2' = toEquation (pre_eqn ax2)
isEquality ax = isJust (decodeEquality (eqn_lhs ax)) || isJust (decodeEquality (eqn_rhs ax))
axioms0 = sortBy axiomCompare unsortedAxioms0
goals =
[ goal n pre_name (toEquation pre_eqn)
| (n, PreEquation{..}) <- zip [1..] goals0 ]
axioms =
[ Axiom n pre_name (toEquation pre_eqn)
| (n, PreEquation{..}) <- zip [1..] axioms0 ]
defs =
[ axiom
| (axiom, PreEquation{..}) <- zip axioms axioms0,
isDefinition pre_form ]
isDefinition Input{source = Unknown} = True
isDefinition inp = tag inp `elem` flags_eliminate
-- Compute CP scoring heuristic
let
goalNests = nests (map goal_eqn goals)
goalOccs = occs (map goal_eqn goals)
score depth hints eqn
| flags_goal_heuristic =
CP.score cpConfig depth hints eqn *
product
[ pos (IntMap.findWithDefault 0 f eqnNests - IntMap.findWithDefault 0 f goalNests) *
pos (IntMap.findWithDefault 0 f eqnOccs - IntMap.findWithDefault 0 f goalOccs)
| f <- IntMap.keys eqnNests ] -- skip constants
| otherwise =
CP.score cpConfig depth hints eqn
where
eqnNests = nests eqn
eqnOccs = occs eqn
pos :: Int -> Float
pos n = if n <= 0 then 1 else fromIntegral n+1
config = config0 { cfg_score_cp = score, cfg_eliminate_axioms = if flags_flatten_regeneralise then defs else [] }
let
withHints = foldl' (addHint config) (initialState config) (map toTerm hints')
withGoals = foldl' (addGoal config) withHints goals
withAxioms = foldl' (addAxiom config) withGoals axioms
withBackwardsGoal = foldn rewriteGoalsBackwards withAxioms flags_backwards_goal
-- Set up tracing.
sayTrace <-
case flags_trace of
Nothing -> return $ \_ -> return ()
Just (file, mod) -> do
h <- openFile file WriteMode
hSetBuffering h LineBuffering
let put msg = hPutStrLn h msg
put $ ":- module(" ++ mod ++ ", [step/1, lemma/1, axiom/1, goal/1])."
put ":- discontiguous(step/1)."
put ":- discontiguous(lemma/1)."
put ":- discontiguous(axiom/1)."
put ":- discontiguous(goal/1)."
put ":- style_check(-singleton)."
return $ \msg -> hPutStrLn h msg
let
say msg = unless (quiet globals) (putStrLn msg)
line = say ""
output = Output {
output_message = \msg -> do
say (prettyShow msg)
sayTrace (show (traceMsg msg)) }
traceMsg (NewActive _ active) =
step "add" [traceActive active]
traceMsg (NewEquation eqn) =
step "hard" [traceEqn eqn]
traceMsg (DeleteActive active) =
step "delete" [traceActive active]
traceMsg SimplifyQueue =
step "simplify_queue" []
traceMsg Interreduce =
step "interreduce" []
traceMsg (Status n) =
step "status" [pPrint n]
traceActive Active{active_top = Nothing, ..} =
traceApp "rule" [pPrint active_id, traceEqn (unorient active_rule)]
traceActive Active{active_top = Just top, ..} =
traceApp "rule" [pPrint active_id, traceEqn (unorient active_rule), traceEqn lemma1, traceEqn lemma2]
where
(lemma1, lemma2) =
find (steps (derivation active_proof))
find (s1:s2:_)
| eqn_rhs (equation (certify s1)) == top && eqn_lhs (equation (certify s2)) == top =
(lemmaOf s1, lemmaOf s2)
find (_:xs) = find xs
lemmaOf s =
case (usedLemmas s, usedAxioms s) of
([p], []) -> equation p
([], [ax]) -> axiom_eqn ax
traceEqn (t :=: u) =
pPrintPrec prettyNormal 6 t <+> text "=" <+> pPrintPrec prettyNormal 6 u
traceApp f xs =
pPrintTerm uncurried prettyNormal 0 (text f) xs
step :: String -> [Doc] -> Doc
step f xs = traceApp "step" [traceApp f xs] <#> text "."
say "Here is the input problem:"
forM_ axioms $ \Axiom{..} ->
say $ show $ nest 2 $
describeEquation "Axiom"
(show axiom_number) (Just axiom_name) axiom_eqn
forM_ goals $ \Goal{..} ->
say $ show $ nest 2 $
describeEquation "Goal"
(show goal_number) (Just goal_name) goal_eqn
line
state <- complete output config withBackwardsGoal
line
when (solved state && flags_proof) $ later $ do
let
cfg_present
| tstp && flags_formal_proof =
(cfg_proof_presentation config){cfg_all_lemmas = True}
| otherwise =
cfg_proof_presentation config
pres = present cfg_present [] $ map (eliminateDefinitionsFromGoal defs) $ solutions state
sayTrace ""
forM_ (pres_axioms pres) $ \p ->
sayTrace $ show $
traceApp "axiom" [traceEqn (axiom_eqn p)] <#> text "."
forM_ (pres_lemmas pres) $ \p ->
sayTrace $ show $
traceApp "lemma" [traceEqn (equation p)] <#> text "."
forM_ (pres_goals pres) $ \p ->
sayTrace $ show $
traceApp "goal" [traceEqn (pg_goal_hint p)] <#> text "."
when (tstp && not flags_formal_proof) $ do
putStrLn "% SZS output start Proof"
let
axiomForms =
Map.fromList
(zip (map axiom_number axioms) (map pre_form axioms0))
goalForms =
Map.fromList
(zip (map goal_number goals) (map pre_form goals0))
findSource forms n =
case Map.lookup n forms of
Nothing -> []
Just inp -> go inp
where
go Input{source = Unknown} = []
go Input{source = Inference _ _ inps} = concatMap (go . inputValue) inps
go inp@Input{source = FromFile _ _} = [inp]
when flags_explain_encoding $ do
putStrLn "Take the following subset of the input axioms:"
mapM_ putStrLn $ map (" " ++) $ lines $ showProblem $
usortBy (comparing show) $
(pres_axioms pres >>= findSource axiomForms . axiom_number) ++
(pres_goals pres >>= findSource goalForms . pg_number)
putStrLn ""
putStrLn "Now clausify the problem and encode Horn clauses using encoding 3 of"
putStrLn "http://www.cse.chalmers.se/~nicsma/papers/horn.pdf."
putStrLn "We repeatedly replace C & s=t => u=v by the two clauses:"
putStrLn " fresh(y, y, x1...xn) = u"
putStrLn " C => fresh(s, t, x1...xn) = v"
putStrLn "where fresh is a fresh function symbol and x1..xn are the free"
putStrLn "variables of u and v."
putStrLn "A predicate p(X) is encoded as p(X)=true (this is sound, because the"
putStrLn "input problem has no model of domain size 1)."
putStrLn ""
putStrLn "The encoding turns the above axioms into the following unit equations and goals:"
putStrLn ""
print $ pPrintPresentation (cfg_proof_presentation config) pres
putStrLn "% SZS output end Proof"
putStrLn ""
when (tstp && flags_formal_proof) $ do
putStrLn "% SZS output start CNFRefutation"
print $ pPrintProof $
presentToJukebox ctx (curry toEquation)
(zip (map axiom_number axioms) (map pre_form axioms0))
(zip (map goal_number goals) (map pre_form goals0))
pres
putStrLn "% SZS output end CNFRefutation"
putStrLn ""
unless tstp $ do
putStrLn "The conjecture is true! Here is a proof."
putStrLn ""
print $ pPrintPresentation (cfg_proof_presentation config) pres
putStrLn ""
when (not (quiet globals) && not (solved state)) $ later $ do
let
state' = interreduce config state
score rule =
(KBO.size (lhs rule), lhs rule,
KBO.size (rhs rule), rhs rule)
actives =
sortBy (comparing (score . active_rule)) $
IntMap.elems (st_active_set state')
when (tstp && configIsComplete config) $ do
putStrLn "% SZS output start Saturation"
print $ pPrintProof $
map pre_form axioms0 ++
map pre_form goals0 ++
[ Input Nothing "rule" (Jukebox.Ax Jukebox.Axiom) Unknown $
toForm $ clause
[Pos (jukeboxTerm ctx (lhs rule) Jukebox.:=: jukeboxTerm ctx (rhs rule))]
| rule <- rules state ]
putStrLn "% SZS output end Saturation"
putStrLn ""
if configIsComplete config then do
putStrLn "Ran out of critical pairs. This means the conjecture is not true."
else do
putStrLn "Gave up on reaching the given resource limit."
putStrLn "Here is the final rewrite system:"
forM_ actives $ \active ->
putStrLn (" " ++ prettyShow (canonicalise (active_rule active)))
putStrLn ""
when flags_proof_on_saturation $ do
let pres = present (cfg_proof_presentation config) (map active_proof actives) []
print $ pPrintPresentation (cfg_proof_presentation config) pres
return $
if solved state then Unsat Unsatisfiable Nothing
else if configIsComplete config && not (dropNonHorn horn) && not flags_give_up_on_saturation then Sat Satisfiable Nothing
else NoAnswer GaveUp
-- Transform a proof presentation into a Jukebox proof.
presentToJukebox ::
TweeContext ->
(Jukebox.Term -> Jukebox.Term -> Equation Constant) ->
-- Axioms, indexed by axiom number.
[(Int, Input Form)] ->
-- N.B. the formula here proves the negated goal.
[(Int, Input Form)] ->
Presentation Constant ->
Problem Form
presentToJukebox ctx toEquation axioms goals Presentation{..} =
[ Input {
ident = Nothing,
tag = pg_name,
kind = Jukebox.Ax Jukebox.Axiom,
what = false,
source =
inference "resolution" "thm"
[-- A proof of t != u
existentialHack pg_goal_hint (fromJust (lookup pg_number goals)),
-- A proof of t = u
fromJust (Map.lookup pg_number goal_proofs)] }
| ProvedGoal{..} <- pres_goals ]
where
axiom_proofs =
Map.fromList
[ (axiom_number, (fromJust (lookup axiom_number axioms)) { ident = Just (ident axiom_number) })
| Axiom{..} <- pres_axioms ]
where
ident i = variant "axiom" [i]
lemma_proofs =
Map.fromList [(p, (tstp p) { ident = Just (ident i) }) | (i, p) <- zip [0..] pres_lemmas]
where
ident i = variant "lemma" [i :: Int]
goal_proofs =
Map.fromList [(pg_number, tstp pg_proof) | ProvedGoal{..} <- pres_goals]
tstp :: Proof Constant -> Input Form
tstp = deriv . derivation
deriv :: Derivation Constant -> Input Form
deriv p =
Input {
ident = Nothing,
tag = "step",
kind = Jukebox.Ax Jukebox.Axiom,
what = jukeboxEquation (equation (certify p)),
source =
inference name "thm" sources }
where
(name, sources) = unpack p
unpack :: Derivation Constant -> (String, [Input Form])
unpack (Refl _) = ("reflexivity", [])
unpack (Symm p) = ("symmetry", [deriv p])
unpack (Trans p q) = ("transitivity", [deriv p, deriv q])
unpack (Cong _ ps) = ("congruence", [deriv p | p <- ps, let t :=: u = equation (certify p), t /= u])
unpack (UseAxiom Axiom{..} _) =
("substitution", [fromJust (Map.lookup axiom_number axiom_proofs)])
unpack (UseLemma lemma _) =
("substitution", [fromJust (Map.lookup lemma lemma_proofs)])
jukeboxEquation :: Equation Constant -> Form
jukeboxEquation (t :=: u) =
toForm $ clause [Pos (jukeboxTerm ctx t Jukebox.:=: jukeboxTerm ctx u)]
-- An ugly hack: since Twee.Proof decodes $true = $false into a
-- proof of the existentially quantified goal, we need to do the
-- same decoding at the Jukebox level.
existentialHack eqn input =
case find input of
[] -> error $ "bug in TSTP output: can't fix up decoded existential"
(inp:_) -> inp
where
-- Check if this looks like the correct clause;
-- if not, try its ancestors.
find inp | ok inp = [inp]
find Input{source = Inference _ _ inps} =
concatMap (find . inputValue) inps
find _ = []
ok inp =
case toClause (what inp) of
Nothing -> False
Just (Clause (Bind _ [Neg (t' Jukebox.:=: u')])) ->
let
eqn' = toEquation t' u'
ts = buildList [eqn_lhs eqn, eqn_rhs eqn]
us = buildList [eqn_lhs eqn', eqn_rhs eqn']
in
isJust (matchList ts us) && isJust (matchList us ts)
main = do
hSetBuffering stdout LineBuffering
stampM (intern "twee") . join . parseCommandLineWithExtraArgs
["--no-conjunctive-conjectures", "--no-split"]
#ifdef VERSION_twee
"Twee, an equational theorem prover" . version ("twee version " ++ VERSION_twee) $
#else
"Twee, an equational theorem prover" . version "twee development version" $
#endif
globalFlags *> parseMainFlags *>
-- hack: get --quiet and --no-proof options to appear before --tstp
forAllFilesBox <*>
(readProblemBox =>>=
expert clausifyBox =>>=
forAllConjecturesBox <*>
(combine <$>
expert hornToUnitBox <*>
parseConfig <*>
parseCPConfig <*>
parseMainFlags <*>
(toFormulasBox =>>=
expert (toFof <$> clausifyBox <*> pure (tags True)) =>>=
expert clausifyBox =>>= expert oneConjectureBox) <*>
(runTwee <$> globalFlags <*> tstpFlags <*> expert hornFlags <*> parsePrecedence)))
profile
where
getHint Input{what = Clause (Bind _ [Pos (Tru (hint :@: [t]))])}
| base (name hint) == "$hint" = Left t
getHint c = Right c
combine horn config cpConfig main encode prove later prob0 = do
let (hints, nonHints) = partitionEithers (map getHint prob0)
res <- horn nonHints
case res of
Left ans -> return ans
Right prob -> do
let
isUnitEquality [Pos (_ Jukebox.:=: _)] = True
isUnitEquality [Neg (_ Jukebox.:=: _)] = True
isUnitEquality _ = False
isUnit = all isUnitEquality (map (toLiterals . what) prob0)
main' = if isUnit then main{flags_explain_encoding = False} else main{flags_formal_proof = False}
encode prob >>= prove config cpConfig main' later hints