mirror of
https://github.com/idris-lang/Idris2.git
synced 2024-12-21 18:51:40 +03:00
355 lines
12 KiB
Idris
355 lines
12 KiB
Idris
||| Implementation of ordering relations for the primitive type `Int`
|
|
||| This is full of tricks as `Int` is a primitive type that can only
|
|
||| be interacted with using either literals or built-in functions.
|
|
||| The exported interface should however be safe.
|
|
module Data.Int.Order
|
|
|
|
import Data.Order
|
|
|
|
%default total
|
|
|
|
------------------------------------------------------------------------
|
|
-- Introduction
|
|
|
|
-- The type `Int` is a primitive type in Idris. The only handle we have on
|
|
-- it is provided by built-in functions. This is what we are going to use
|
|
-- to define relations on Int values. For instance, we will declare that
|
|
-- `a` is strictly less than `b` whenever `a < b` is provably equal to `True`.
|
|
|
|
-- These built-in functions only reduce on literals. This means we will not
|
|
-- be able to rely on their computational behaviour to prove statements in
|
|
-- open contexts.
|
|
|
|
-- For instance, no amount of pattern-matching will help us prove:
|
|
-- LT_not_EQ : LT a b -> Not (EQ a b)
|
|
|
|
-- Our solution in this file is to use unsafe primitives to manufacture such
|
|
-- proofs. This means we are going to essentially postulate some *conditional*
|
|
-- results. We do not want such conditional results to reduce to canonical
|
|
-- forms too eagerly.
|
|
|
|
-- Indeed the statement `GT 0 1 -> EQ 0 1` should be provable because 0 is not
|
|
-- greater than 1. But its proof should not reduce to a constant function
|
|
-- returning the value `Refl` because it is not true that `0` and `1` can be
|
|
-- unified. If the proof were to behave this way, we could, in an absurd context,
|
|
-- coerce values from any type to any other and cause segmentation faults.
|
|
|
|
-- Our solution is to be extremely wary of proofs that are passed to us
|
|
-- and to only consider returning a magically-crafted output if we have
|
|
-- managed to observe that the input is itself in canonical form i.e. to
|
|
-- have evaluation stuck on open terms.
|
|
|
|
-- This is the purpose of the `strictX : X -> Lazy c -> c` functions defined in
|
|
-- this file. They all will be waiting until their first argument is in canonical
|
|
-- form before returning their second.
|
|
|
|
|
|
------------------------------------------------------------------------
|
|
-- Utility functions for equality
|
|
|
|
||| Pattern-match on an equality proof so that the second argument to
|
|
||| strictRefl is only returned once the first is canonical.
|
|
export
|
|
strictRefl : a === b -> Lazy c -> c
|
|
strictRefl Refl p = p
|
|
|
|
||| Do NOT re-export
|
|
unsafeRefl : {0 a, b : t} -> a === b
|
|
unsafeRefl = believe_me (the (a === a) Refl)
|
|
|
|
------------------------------------------------------------------------
|
|
-- LT
|
|
|
|
namespace LT
|
|
|
|
||| `LT a b` is a proof that `a` is less than `b` which is to say that
|
|
||| the function call `a < b` returns `True`.
|
|
|||
|
|
||| NB: we do not re-export the constructor so that users cannot force
|
|
||| our magic functions to blow up in absurd contexts by bypassing the
|
|
||| safety measures introduced by the `strictX` functions.
|
|
||| We do provide functions corresponding to the wrapping and unwrapping
|
|
||| of `LT` but, crucially, they do not let people force `LT` proofs to
|
|
||| be in canonical form.
|
|
export
|
|
data LT : Int -> Int -> Type where
|
|
MkLT : (a < b) === True -> LT a b
|
|
|
|
||| We may prove `LT a b` by using a proof that `a < b` returns `True`.
|
|
export
|
|
mkLT : (a < b) === True -> LT a b
|
|
mkLT = MkLT
|
|
|
|
||| From a proof that `LT a b`, we may concluded that `a < b` returns `True`.
|
|
export
|
|
runLT : LT a b -> (a < b) === True
|
|
runLT (MkLT eq) = eq
|
|
|
|
||| Do not trust arbitrary `LT` proofs when manufacturing magic ones:
|
|
||| be strict!
|
|
export
|
|
strictLT : LT a b -> Lazy c -> c
|
|
strictLT (MkLT p) c = strictRefl p c
|
|
|
|
||| LT is decidable, by virtue of being the reflection of a boolean function
|
|
export
|
|
decide : (a, b : Int) -> Dec (LT a b)
|
|
decide a b with (the (test : Bool ** (a < b) === test) (a < b ** Refl))
|
|
decide a b | (True ** p) = Yes (MkLT p)
|
|
decide a b | (False ** p) = No (\ (MkLT q) => absurd (trans (sym p) q))
|
|
|
|
||| LT is a transitive relation. This cannot be proven so we use a magic trick.
|
|
export
|
|
trans : LT a b -> LT b c -> LT a (the Int c)
|
|
trans p q = strictLT p $ strictLT q $ MkLT unsafeRefl
|
|
|
|
||| LT is an irreflexive relation.
|
|
||| The crash will never happen because the `LT a a` argument will never reduce
|
|
||| to a canonical form (unless the built-in function (<) is buggy).
|
|
export
|
|
irrefl : Not (LT a a)
|
|
irrefl p = strictLT p
|
|
$ assert_total $ idris_crash "IMPOSSIBLE: LT is irreflexive"
|
|
|
|
------------------------------------------------------------------------
|
|
-- GT
|
|
|
|
||| GT is defined in terms of LT to avoid code duplication
|
|
public export
|
|
GT : Int -> Int -> Type
|
|
GT = flip LT
|
|
|
|
export
|
|
LT_not_GT : LT a b -> Not (GT a b)
|
|
LT_not_GT p q = irrefl (trans p q)
|
|
|
|
export
|
|
GT_not_LT : GT a b -> Not (LT a b)
|
|
GT_not_LT = flip LT_not_GT
|
|
|
|
------------------------------------------------------------------------
|
|
-- EQ
|
|
|
|
namespace EQ
|
|
|
|
||| `EQ a b` is a proof that `a` is equal to `b` which is to say that
|
|
||| the function call `a == b` returns `True`.
|
|
|||
|
|
||| NB: we do not re-export the constructor so that users cannot force
|
|
||| our magic functions to blow up in absurd contexts by bypassing the
|
|
||| safety measures introduced by the `strictX` functions.
|
|
||| We do provide functions corresponding to the wrapping and unwrapping
|
|
||| of `EQ` but, crucially, they do not let people force `EQ` proofs to
|
|
||| be in canonical form.
|
|
export
|
|
data EQ : Int -> Int -> Type where
|
|
MkEQ : (a == b) === True -> EQ a b
|
|
|
|
||| We may prove `EQ a b` by using a proof that `a == b` returns `True`.
|
|
export
|
|
mkEQ : (a == b) === True -> EQ a b
|
|
mkEQ = MkEQ
|
|
|
|
||| From a proof that `EQ a b`, we may concluded that `a == b` returns `True`.
|
|
export
|
|
runEQ : EQ a b -> (a == b) === True
|
|
runEQ (MkEQ eq) = eq
|
|
|
|
||| Do not trust arbitrary `EQ` proofs when manufacturing magic ones:
|
|
||| be strict!
|
|
export
|
|
strictEQ : EQ a b -> Lazy c -> c
|
|
strictEQ (MkEQ p) c = strictRefl p c
|
|
|
|
||| EQ is decidable, by virtue of being the reflection of a boolean function
|
|
export
|
|
decide : (a, b : Int) -> Dec (EQ a b)
|
|
decide a b with (the (test : Bool ** (a == b) === test) (a == b ** Refl))
|
|
decide a b | (True ** p) = Yes (MkEQ p)
|
|
decide a b | (False ** p) = No (\ (MkEQ q) => absurd (trans (sym p) q))
|
|
|
|
||| EQ is a reflexive relation
|
|
export
|
|
refl : EQ a a
|
|
refl = MkEQ unsafeRefl
|
|
|
|
||| EQ is substitutive
|
|
export
|
|
elimEQ : (0 p : Int -> Type) -> EQ a b -> p a -> p b
|
|
elimEQ _ p v = strictEQ p $ believe_me v
|
|
|
|
||| EQ implies propositional equality
|
|
export
|
|
reflect : EQ a b -> a === b
|
|
reflect p = elimEQ (\ b => a === b) p Refl
|
|
|
|
||| EQ is a symmetric relation
|
|
export
|
|
sym : EQ a b -> EQ b a
|
|
sym p = elimEQ (\ b => EQ b a) p refl
|
|
|
|
||| EQ is a transitive relation
|
|
export
|
|
trans : EQ a b -> EQ b c -> EQ a c
|
|
trans p q = elimEQ (\ b => EQ b c) (sym p) q
|
|
|
|
export
|
|
trans_LT_EQ : LT a b -> EQ b c -> LT a c
|
|
trans_LT_EQ p q = elimEQ (LT a) q p
|
|
|
|
export
|
|
trans_EQ_LT : EQ a b -> LT b c -> LT a c
|
|
trans_EQ_LT p q = elimEQ (\ b => LT b c) (sym p) q
|
|
|
|
export
|
|
LT_not_EQ : LT a b -> Not (EQ a b)
|
|
LT_not_EQ p q = irrefl (trans_LT_EQ p (sym q))
|
|
|
|
export
|
|
EQ_not_LT : EQ a b -> Not (LT a b)
|
|
EQ_not_LT = flip LT_not_EQ
|
|
|
|
export
|
|
EQ_not_GT : EQ a b -> Not (GT a b)
|
|
EQ_not_GT = EQ_not_LT . sym
|
|
|
|
export
|
|
GT_not_EQ : GT a b -> Not (EQ a b)
|
|
GT_not_EQ = flip EQ_not_GT
|
|
|
|
------------------------------------------------------------------------
|
|
-- LTE
|
|
|
|
namespace LTE
|
|
|
|
||| `LTE a b` is a proof that `a` is less or equal to `b` which is to say that
|
|
||| the function call `a < b` or `a == b` returns `True`.
|
|
|||
|
|
||| NB: we do not re-export the constructor so that users cannot force
|
|
||| our magic functions to blow up in absurd contexts by bypassing the
|
|
||| safety measures introduced by the `strictX` functions.
|
|
export
|
|
data LTE : Int -> Int -> Type where
|
|
MkLT : (a < b) === True -> LTE a b
|
|
MkEQ : (a == b) === True -> LTE a b
|
|
|
|
||| Unwrap an LTE proof to get either an LT or an EQ one
|
|
export
|
|
runLTE : LTE a b -> Either (LT a b) (EQ a b)
|
|
runLTE (MkLT p) = Left (mkLT p)
|
|
runLTE (MkEQ p) = Right (mkEQ p)
|
|
|
|
||| Do not trust arbitrary `LTE` proofs when manufacturing magic ones:
|
|
||| be strict!
|
|
export
|
|
strictLTE : LTE a b -> Lazy c -> c
|
|
strictLTE (MkLT p) q = strictRefl p q
|
|
strictLTE (MkEQ p) q = strictRefl p q
|
|
|
|
||| LTE is decidable by virture of both LT and EQ being decidable
|
|
export
|
|
decide : (a, b : Int) -> Dec (LTE a b)
|
|
decide a b with (LT.decide a b)
|
|
decide a b | Yes p = Yes (MkLT (runLT p))
|
|
decide a b | No notLT with (EQ.decide a b)
|
|
decide a b | No notLT | Yes p = Yes (MkEQ (runEQ p))
|
|
decide a b | No notLT | No notEQ = No $ \ case
|
|
MkLT p => notLT (mkLT p)
|
|
MkEQ p => notEQ (mkEQ p)
|
|
|
|
||| LTE is a reflexive relation
|
|
export
|
|
refl : LTE a a
|
|
refl = MkEQ unsafeRefl
|
|
|
|
export
|
|
trans_LT_LTE : LT a b -> LTE b c -> LT a c
|
|
trans_LT_LTE p (MkLT q) = trans p (mkLT q)
|
|
trans_LT_LTE p (MkEQ q) = trans_LT_EQ p (mkEQ q)
|
|
|
|
export
|
|
trans_LTE_LT : LTE a b -> LT b c -> LT a c
|
|
trans_LTE_LT (MkLT p) q = trans (mkLT p) q
|
|
trans_LTE_LT (MkEQ p) q = trans_EQ_LT (mkEQ p) q
|
|
|
|
export
|
|
inject_LT_LTE : LT a b -> LTE a b
|
|
inject_LT_LTE = MkLT . runLT
|
|
|
|
export
|
|
inject_EQ_LTE : EQ a b -> LTE a b
|
|
inject_EQ_LTE p = MkEQ (runEQ p)
|
|
|
|
||| LTE is a transitive relation
|
|
export
|
|
trans : LTE a b -> LTE b c -> LTE a c
|
|
trans (MkLT p) q = inject_LT_LTE (trans_LT_LTE (mkLT p) q)
|
|
trans p (MkLT q) = inject_LT_LTE (trans_LTE_LT p (mkLT q))
|
|
trans (MkEQ p) (MkEQ q) = inject_EQ_LTE (trans (mkEQ p) (mkEQ q))
|
|
|
|
||| LTE is an antisymmetric relation
|
|
export
|
|
antisym : LTE a b -> LTE b a -> EQ a b
|
|
antisym (MkEQ p) _ = mkEQ p
|
|
antisym _ (MkEQ p) = EQ.sym (mkEQ p)
|
|
antisym (MkLT p) (MkLT q) = absurd $ LT.irrefl (LT.trans (mkLT p) (mkLT q))
|
|
|
|
------------------------------------------------------------------------
|
|
-- GTE
|
|
|
|
||| GTE is defined in terms of LTE to avoid code duplication
|
|
public export
|
|
GTE : Int -> Int -> Type
|
|
GTE = flip LTE
|
|
|
|
------------------------------------------------------------------------
|
|
-- Trichotomy and other decidability results
|
|
|
|
||| Any pair of Ints is related either via LT, EQ, or GT
|
|
export
|
|
trichotomous : (a, b : Int) -> Trichotomous LT EQ GT a b
|
|
trichotomous a b with (LTE.decide a b)
|
|
trichotomous a b | Yes lte = case runLTE lte of
|
|
Left lt => MkLT lt (LT_not_EQ lt) (LT_not_GT lt)
|
|
Right eq => MkEQ (EQ_not_LT eq) eq (EQ_not_GT eq)
|
|
trichotomous a b | No notLTE = let gt = mkLT unsafeRefl in MkGT (GT_not_LT gt) (GT_not_EQ gt) gt
|
|
|
|
||| Any pair of Ints is related either via LT or GTE
|
|
export
|
|
decide_LT_GTE : (a, b : Int) -> Either (LT a b) (GTE a b)
|
|
decide_LT_GTE a b with (trichotomous a b)
|
|
decide_LT_GTE a b | MkLT lt _ _ = Left lt
|
|
decide_LT_GTE a b | MkEQ _ eq _ = Right (inject_EQ_LTE (sym eq))
|
|
decide_LT_GTE a b | MkGT _ _ gt = Right (inject_LT_LTE gt)
|
|
|
|
||| Any pair of Ints is related either via LTE or GT
|
|
export
|
|
decide_LTE_GT : (a, b : Int) -> Either (LTE a b) (GT a b)
|
|
decide_LTE_GT a b with (trichotomous a b)
|
|
decide_LTE_GT a b | MkLT lt _ _ = Left (inject_LT_LTE lt)
|
|
decide_LTE_GT a b | MkEQ _ eq _ = Left (inject_EQ_LTE eq)
|
|
decide_LTE_GT a b | MkGT _ _ gt = Right gt
|
|
|
|
------------------------------------------------------------------------
|
|
-- Some properties
|
|
|
|
||| Adding one to a strictly smaller Int, yields a smaller Int
|
|
export
|
|
suc_LT_LTE : {a, b : Int} -> LT a b -> LTE (a + 1) b
|
|
suc_LT_LTE p with (the (test : Bool ** (a + 1 == b) === test) (a + 1 == b ** Refl))
|
|
suc_LT_LTE p | (True ** q) = inject_EQ_LTE $ mkEQ q
|
|
suc_LT_LTE p | (False ** q) = strictLT p $ inject_LT_LTE $ mkLT unsafeRefl
|
|
|
|
||| Subtracting one to a strictly larger Int, yields a larger Int
|
|
export
|
|
pred_LT_LTE : {a, b : Int} -> LT a b -> LTE a (b - 1)
|
|
pred_LT_LTE p with (the (test : Bool ** (a == b - 1) === test) (a == b - 1 ** Refl))
|
|
pred_LT_LTE p | (True ** q) = inject_EQ_LTE $ mkEQ q
|
|
pred_LT_LTE p | (False ** q) = strictLT p $ inject_LT_LTE $ mkLT unsafeRefl
|
|
|
|
||| Adding one to an Int yields a strictly larger one,
|
|
||| provided there is no overflow
|
|
export
|
|
sucBounded : LT a b -> LT a (a + 1)
|
|
sucBounded p = strictLT p $ mkLT unsafeRefl
|