A Coercion is concrete evidence of the equality/convertibility of two types.

A semantically more meaningful type to represent what may or may not be a useful Coercion.

For simplicity, we have just one UnivCo that represents a coercion from some type to some other type, with (in general) no restrictions on the type. The UnivCoProvenance specifies more exactly what the coercion really is and why a program should (or shouldn't!) trust the coercion. It is reasonable to consider each constructor of UnivCoProvenance as a totally independent coercion form; their only commonality is that they don't tell you what types they coercion between. (That info is in the UnivCo constructor of Coercion.

A coercion to be filled in by the type-checker. See Note [Coercion holes]

Variable

Essentially a typed Name, that may also contain some additional information about the Var and its use sites.

Coercion Variable

Type or Coercion Variable

See Note [Roles] in GHC.Core.Coercion

Order of constructors matters: the Ord instance coincides with the *super*typing relation on roles.

Makes a coercion type from two types: the types whose equality is proven by the relevant Coercion

If it is the case that

c :: (t1 ~ t2)

i.e. the kind of c relates t1 and t2, then coercionKind c = Pair t1 t2.

Apply coercionKind to multiple Coercions

Retrieve the role from a coercion.

Get a coercion's kind and role.

Make a generalized reflexive coercion

Make a reflexive coercion

Make a representational reflexive coercion

Make a nominal reflexive coercion

Return the left-hand type of the axiom, when the axiom is instantiated at the types given.

Instantiate the left-hand side of an unbranched axiom

Make a forall Coercion, where both types related by the coercion are quantified over the same variable.

Create a symmetric version of the given Coercion that asserts equality between the same types but in the other "direction", so a kind of t1 ~ t2 becomes the kind t2 ~ t1.

mkTransCo creates a new Coercion by composing the two given Coercions transitively: (co1 ; co2)

Extract the nth field of a FunCo

Instantiates a Coercion. Works for both tyvar and covar

Apply a Coercion to another Coercion. The second coercion must be Nominal, unless the first is Phantom. If the first is Phantom, then the second can be either Phantom or Nominal.

Applies multiple Coercions to another Coercion, from left to right. See also mkAppCo.

Apply a type constructor to a list of coercions. It is the caller's responsibility to get the roles correct on argument coercions.

Build a function Coercion from two other Coercions. That is, given co1 :: a ~ b and co2 :: x ~ y produce co :: (a -> x) ~ (b -> y) or (a => x) ~ (b => y), depending on the kind of a/b. This (most common) version takes a single FunTyFlag, which is used for both fco_afl and ftf_afr of the FunCo

Make a Coercion from a tycovar, a kind coercion, and a body coercion.

Make a Coercion quantified over a type/coercion variable; the variable has the same kind and visibility in both sides of the coercion

Make a phantom coercion between two types. The coercion passed in must be a nominal coercion between the kinds of the types.

Make a coercion from a coercion hole

Make a universal coercion between two arbitrary types.

Make a "coercion between coercions".

Like downgradeRole_maybe, but panics if the change isn't a downgrade. See Note [Role twiddling functions]

Given ty :: k1, co :: k1 ~ k2, produces co' :: ty ~r (ty |> co)

Given r, ty :: k1, and co :: k1 ~N k2, produces co' :: (ty |> co) ~r ty

Given ty :: k1, co :: k1 ~ k2, co2:: ty ~r ty', produces @co' :: (ty |> co) ~r ty' It is not only a utility function, but it saves allocation when co is a GRefl coercion.

Given ty :: k1, co :: k1 ~ k2, co2:: ty' ~r ty, produces @co' :: ty' ~r (ty |> co) It is not only a utility function, but it saves allocation when co is a GRefl coercion.

Given co :: (a :: k) ~ (b :: k') produce co' :: k ~ k'.

Creates a new coercion with both of its types casted by different casts castCoercionKind g h1 h2, where g :: t1 ~r t2, has type (t1 |> h1) ~r (t2 |> h2). h1 and h2 must be nominal. It calls coercionKindRole, so it's quite inefficient (which I stands for) Use castCoercionKind2 instead if t1, t2, and r are known beforehand.

castCoercionKind1 g r t1 t2 h = coercionKind g r t1 t2 h h That is, it's a specialised form of castCoercionKind, where the two kind coercions are identical castCoercionKind1 g r t1 t2 h, where g :: t1 ~r t2, has type (t1 |> h) ~r (t2 |> h). h must be nominal. See Note [castCoercionKind1]

Creates a new coercion with both of its types casted by different casts castCoercionKind2 g r t1 t2 h1 h2, where g :: t1 ~r t2, has type (t1 |> h1) ~r (t2 |> h2). h1 and h2 must be nominal.

Creates a primitive nominal type equality predicate. t1 ~# t2 Invariant: the types are not Coercions

Creates a primitive representational type equality predicate. t1 ~R# t2 Invariant: the types are not Coercions

Makes a lifted equality predicate at the given role

Creates a primitive nominal type equality predicate with an explicit (but homogeneous) kind: (~#) k k ty1 ty2

If `instNewTyCon_maybe T ts = Just (rep_ty, co)` then `co :: T ts ~R# rep_ty`

Checks for a newtype, and for being saturated

A function to check if we can reduce a type by one step. Used with topNormaliseTypeX.

The result of stepping in a normalisation function. See topNormaliseTypeX.

Try one stepper and then try the next, if the first doesn't make progress. So if it returns NS_Done, it means that both steppers are satisfied

A NormaliseStepper that unwraps newtypes, careful not to fall into a loop. If it would fall into a loop, it produces NS_Abort.

Sometimes we want to look through a newtype and get its associated coercion. This function strips off newtype layers enough to reveal something that isn't a newtype. Specifically, here's the invariant:

topNormaliseNewType_maybe rec_nts ty = Just (co, ty')

then (a) co : ty ~R ty'. (b) ty' is not a newtype.

The function returns Nothing for non-newtypes, or unsaturated applications

This function does *not* look through type families, because it has no access to the type family environment. If you do have that at hand, consider to use topNormaliseType_maybe, which should be a drop-in replacement for topNormaliseNewType_maybe If topNormliseNewType_maybe ty = Just (co, ty'), then co : ty ~R ty'

A general function for normalising the top-level of a type. It continues to use the provided NormaliseStepper until that function fails, and then this function returns. The roles of the coercions produced by the NormaliseStepper must all be the same, which is the role returned from the call to topNormaliseTypeX.

Typically ev is Coercion.

If topNormaliseTypeX step plus ty = Just (ev, ty') then ty ~ev1~ t1 ~ev2~ t2 ... ~evn~ ty' and ev = ev1 plus ev2 plus ... plus evn If it returns Nothing then no newtype unwrapping could happen

This breaks a Coercion with type T A B C ~ T D E F into a list of Coercions of kinds A ~ D, B ~ E and E ~ F. Hence:

decomposeCo 3 c [r1, r2, r3] = [nth r1 0 c, nth r2 1 c, nth r3 2 c]

Extract a covar, if possible. This check is dirty. Be ashamed of yourself. (It's dirty because it cares about the structure of a coercion, which is morally reprehensible.)

Attempt to take a coercion application apart.

Like splitForAllCo_maybe, but only returns Just for tyvar binder

Like splitForAllCo_maybe, but only returns Just for covar binder

Converts a coercion to be nominal, if possible. See Note [Role twiddling functions]

Tests if this coercion is obviously a generalized reflexive coercion. Guaranteed to work very quickly.

Tests if this coercion is obviously reflexive. Guaranteed to work very quickly. Sometimes a coercion can be reflexive, but not obviously so. c.f. isReflexiveCo

Returns the type coerced if this coercion is reflexive. Guaranteed to work very quickly. Sometimes a coercion can be reflexive, but not obviously so. c.f. isReflexiveCo_maybe

Returns the type coerced if this coercion is a generalized reflexive coercion. Guaranteed to work very quickly.

Slowly checks if the coercion is reflexive. Don't call this in a loop, as it walks over the entire coercion.

Extracts the coerced type from a reflexive coercion. This potentially walks over the entire coercion, so avoid doing this in a loop.

Tests if this MCoercion is obviously generalized reflexive Guaranteed to work very quickly.

Like mkCoherenceRightCo, but with an MCoercion

Compose two MCoercions via transitivity

Cast a type by an MCoercion

Get the reverse of an MCoercion

Is this a coercion variable? Satisfies isId v ==> isCoVar v == not (isNonCoVarId v).

Get a deterministic set of the vars free in a coercion

A substitution of Coercions for CoVars

Substitute within a Coercion The substitution has to satisfy the invariants described in Note [The substitution invariant].

Substitute within several Coercions The substitution has to satisfy the invariants described in Note [The substitution invariant].

Coercion substitution, see zipTvSubst

liftCoSubst role lc ty produces a coercion (at role role) that coerces between lc_left(ty) and lc_right(ty), where lc_left is a substitution mapping type variables to the left-hand types of the mapped coercions in lc, and similar for lc_right.

Extend a lifting context with a new mapping.

Extend a lifting context with a new mapping, and extend the in-scope set

Is a var in the domain of a lifting context?

Extend the substitution component of a lifting context with a new binding for a coercion variable. Used during coercion optimisation.

Erase the environments in a lifting context

Like substForAllCoBndr, but works on a lifting context

Lookup a CoVar in the substitution in a LiftingContext

Get the InScopeSet from a LiftingContext

Apply "sym" to all coercions in a LiftCoEnv

Syntactic equality of coercions

Compare two Coercions, with respect to an RnEnv2

Tidy a Coercion

See Note [Strictness in tidyType and friends]

like mkKindCo, but aggressively & recursively optimizes to avoid using a KindCo constructor. The output role is nominal.

Assuming that two types are the same, ignoring coercions, find a nominal coercion between the types. This is useful when optimizing transitivity over coercion applications, where splitting two AppCos might yield different kinds. See Note [EtaAppCo] in GHC.Core.Coercion.Opt.

Given a coercion `co :: (t1 :: TYPE r1) ~ (t2 :: TYPE r2)` produce a coercion `rep_co :: r1 ~ r2` But actually it is possible that co :: (t1 :: CONSTRAINT r1) ~ (t2 :: CONSTRAINT r2) or co :: (t1 :: TYPE r1) ~ (t2 :: CONSTRAINT r2) or co :: (t1 :: CONSTRAINT r1) ~ (t2 :: TYPE r2) See Note [mkRuntimeRepCo]

Is there a hetero-kind coercion hole in this type? (That is, a coercion hole with ch_hetero_kind=True.) See wrinkle (EIK2) of Note [Equalities with incompatible kinds] in GHC.Tc.Solver.Equality

Is there a hetero-kind coercion hole in this coercion?

Set the type of a CoercionHole