Support for generic programming

GHC includes a new (in 2010) mechanism to let you write generic functions. It is described in paper A generic deriving mechanism for Haskell. This page sketches the specifics of the implementation; we assume you have read the paper. The HaskellWiki page gives a more general overview.

This mechanism replaces the previous generic classes implementation. What we describe until the "Kind polymorphic overhaul" section is implemented and released in GHC 7.2.1.


Use Keyword = Generics to ensure that a ticket ends up on this auto-generated list

Open Tickets:

Deriving Generic of a big type takes a long time and lots of space
Generic1 deriving: Can we replace Rec1 f with f :.: Par1?
Add (->) representation and the Invariant class to GHC.Generics
Derive some generic representation for GADTs
The example for GHC Generics is kinda broken
compile-time performance regression (probably due to Generics)
Make the runtime reflection API for names, modules, locations more systematic
DefaultSignatures: error message mentions internal name
Generic for existential types
Make Generic/Generic1 methods inlinable
Generic type class has type family; leads to big dep_finsts
Prohibit user-defined Generic and Generic1 instances
Standalone-derived anyclass instances aren't as permissive as empty instances

Closed Tickets:

Duplicate symbols generated when Generics flag and syb-with-class "derive" used simultaneously
Generics compilation failure in combination with Data.Accessor deriving
GHC internal error when missing -XFlexibleContexts with generics
Large space usage when deriving Generic
Deriving clause for arbitrary classes
Change the way the extra bindings for the generic representation are generated
GHC panics while trying to derive a Generic instance for Complex a
Support for data families in generics
Standalone deriving Generic on type with instantiated arguments
`deriving instance Generic (HsExpr Id)` raises panic
Wrong suggestion when deriving Generic on an instantiated type
Add derived Show instances to GHC.Generics
Allow the use of `deriving` for GHC generics
GHC panics while trying to derive Generic for GADT with kind-lifted phantom parameter
Update documentation regarding derivability of Generic1
deriving Generic does not work with TypeLits
Deriving Generic1 for a type containing Either
Allow to differentiate between newtypes and datatypes when using Generics
Panic when using DerivingGeneric with hs-boot
GHC.Generics needs more documentation
ghc panic in deriving Generic1 on Array#
Generic1 crashes with tyConStupidTheta when using associated type families
Generics instances for monoid and applicative newtypes
Deriving Generics broken
Add missing type class instances for data types in GHC.Generics
Typo in GHC Generics documentation
Add missing Generic instances in base
Add Generic instances for Language.Haskell.TH
Support for deriving Generic1 for data families
Use TypeLits in the meta-data encoding of GHC.Generics
DeriveAnyClass support for higher-kinded classes + some more comments
DeriveAnyClass fails on multi-parameter type classes
packageName for GHC.Generics.Datatype
DeriveAnyClass does not fill in associated type defaults
DeriveGeneric breaks when the same data name is used in different modules
Generic instances missing for Int32, Word64 etc.
ghc 7.6.3 Compiler panic with Generics
DeriveAnyClass and GND don't work well together
Make Generic1 kind polymorphic
Missing Generic instances for base types.
Metadata in GHC.Generic should give access to strictness annotation
Enable PolyKinds in GHC.Generics
ghc 7.8.4 on arm - panic: Simplifier ticks exhausted
Make GHC generics capable of handling unboxed types
Traversable can't be derived for datatypes with unboxed arguments
Generics unboxed types: TEST=GEq1 WAY=optasm is failing
Regression when deriving Generic1 on poly-kinded data family
GHC generics has differing conFixity behavior between 7.10.3 and 8.1
pandoc-types fails to build on 4 GB machine
Segmentation fault/internal error: evacuate: strange closure type 15 with GHC.Generics
Deriving Generic1 interacts poorly with TypeInType
singleton classes in ghc.generics are defined but not exported
Generics deriving is quadratic
GHC panic when using DeriveAnyClass with functor-like class and datatype with a type variable in a contravariant position
Commit adding instances to GHC.Generics regression compiler performance
Panic with DeriveAnyClass
DeriveAnyClass fails to derive some classes
High memory usage during compilation
Error in GHC.Generics documentation
DeriveAnyClass regression involving a rigid type variable
StandaloneDeriving and DeriveAnyClass don't work together

Main components

  • TcDeriv.tcDeriving now allows deriving Generic instances.
  • The representation types and core functionality of the library live on GHC.Generics (on the ghc-prim package).
  • Many names have been added as known in prelude/PrelNames
  • Most of the code generation is handled by types/Generics

Things that have been removed

  • All of the generic classes stuff. In particular, the following have been removed:
    • hasGenerics field from TyCon;
    • HsNumTy constructor from HsType;
    • TypePat constructor from Pat.
  • The -XGenerics flag is now deprecated.

What already works

  • Generic and Generic1 instances can be derived when -XDeriveGeneric is enabled.
  • The default keyword can used for generic default method signatures when -XDefaultSignatures is enabled.
  • Generic defaults are properly instantiated when giving an instance without defining the generic default method.
  • Base types like [], Maybe, tuples, come with Generic instances.


  • Tests are available under the generics directory of the testsuite.

Differences from original paper

GHC generics was originally based off of the paper A Generic Deriving Mechanism for Haskell (referred to henceforth as GDMH), but GHC generics has diverged from the original presentation in the paper since then. Here are some of the differences:

  • What GDMH called Representable0 and Representable1 are now called Generic and Generic1
  • GDMH has a type Par0 for denoting occurrences of the last type parameter of a datatype in a Generic instance. GHC generics, however, does not use Par0 anymore in derived instances (after all, you can have Generic instances for datatypes without type parameters!), and the Par0 type has been marked as deprecated.
  • GDMH uses auxiliary datatypes to encode metadata about datatypes, constructors, and selectors. In GHC 8.0, these auxiliary datatypes were scrapped in favor of a type-level encoding of metadata.
  • GHC generics has much more metadata than what was originally presented in GDMH (for example, Selector now encodes strictness information)
  • GHC generics has an extra representation type (URec) for certain unlifted types
  • GHC generics alters the algorithm proposed in GDMH for implementing to(1)/from(1) for performance reasons

Kind polymorphic overhaul

With the new -XPolyKinds functionality we can make the support for generic programming better typed. The basic idea is to define the universe codes (M1, :+:, etc.) as constructors of a datatype. Promotion then lifts these constructors to types, which we can use as before, only that now we have them all classified under a new kind. The overhaul of the main module is explained below; for easier comparison with the current approach, names are kept the same whenever possible.

Generic representation universe

m is the only real parameter here. f and x are there because we can't write kinds directly, since Universe is also a datatype (even if we're only interested in its promoted version). So we pass f and x only to set them to * -> * and *, respectively, in Interprt. m is different: it stands for the kind of metadata representation types, and we really want to be polymorphic over that, since each user datatype will introduce a new metadata kind.

data Universe f x m = 
  -- Void (used for datatypes without constructors)
  -- Unit
  | UU
  -- The parameter
  | PAR
  -- Recursion into a type of kind * -> *
  | REC f
  -- Constants (either other parameters or recursion into types of kind *)
  | KK Constant x
  -- Metadata
  | MM MetaData m (Universe f x m)
  -- Sum, product, composition
  | Universe f x m :++: Universe f x m
  | Universe f x m :**: Universe f x m
  | f :..: Universe f x m
  -- Note that we always compose a concrete type on the left (like []) with
  -- a generic representation on the right

infixr 5 :++:
infixr 6 :**:
infixr 6 :*:
infixr 7 :..:

-- Some shortcuts
data MetaData = CC | DD | SS
data Constant = PP | RR

data ConstantV (c :: Constant) where
  P :: ConstantV PP
  R :: ConstantV RR
data MetaDataV (m :: MetaData) where
  C :: MetaDataV CC
  D :: MetaDataV DD
  S :: MetaDataV SS

Universe interpretation

As promised, we set f to * -> * and x to *. Unfortunately we don't have explicit kind variable annotations yet, so we cannot leave m polymorphic! So this code doesn't compile:

data Interprt :: Universe (* -> *) * m -> * -> * where

  -- No interpretation for VV, as it shouldn't map to any value
  -- Unit
  U1     :: Interprt UU p
  -- The parameter
  Par1   :: p -> Interprt PAR p
  -- Recursion into a type of kind * -> *
  Rec1   :: r p -> Interprt (REC r) p
  -- Constants
  K1     :: x -> Interprt (KK c x) p
  -- Constants shortcuts
  Par0   :: x -> Interprt (KK PP x) p
  Rec0   :: x -> Interprt (KK RR x) p
  -- Metadata
  M1     :: Interprt x p -> Interprt (MM m c x) p
  -- Metadata shortcuts
  D1     :: Interprt x p -> Interprt (MM DD c x) p
  C1     :: Interprt x p -> Interprt (MM CC c x) p
  S1     :: Interprt x p -> Interprt (MM SS c x) p
  -- Sum, product, and composition
  L1     :: Interprt a r -> Interprt (a :++: b) r
  R1     :: Interprt b r -> Interprt (a :++: b) r
  (:*:)  :: Interprt a r -> Interprt b r -> Interprt (a :**: b) r
  Comp1  :: f (Interprt g r) -> Interprt (f :..: g) r


As an aside, note that we have to come up with names like UU and KK for the Universe even though we really just wanted to use U1 and K1, like before. Then we would have a type and a constructor with the same name, but that's ok. However, Universe defines both a type (with constructors) and a kind (with types). So if we were to use U1 in the Universe constructors, then we could no longer use that name in the Interprt constructors. It's a bit annoying, because we are never really interested in the type Universe and its constructors: we're only interested in its promoted variant. This is a slight annoyance of automatic promotion: when you define a "singleton type" (like our GADT Interprt for Universe) you cannot reuse the constructor names.

Metadata representation

data Proxy d = Proxy -- kind polymorphic

-- Meta data classes
class Datatype d where -- kind polymorphic
  -- The name of the datatype, fully qualified
  datatypeName :: Proxy d -> String

There's more of these, but they don't add any new concerns.

Conversion between user datatypes and generic representation

We now get a more precise kind for Rep:

-- Representable types of kind *
class Generic a where
  type Rep a :: Universe (* -> *) * m
  from :: a -> Interprt (Rep a) x
  to   :: Interprt (Rep a) x -> a
-- Representable types of kind * -> *
class Generic1 (f :: * -> *) where
  type Rep1 f :: Universe (* -> *) * m
  from1  :: f a -> Interprt (Rep1 f) a
  to1    :: Interprt (Rep1 f) a -> f a

Example generic function: fmap (kind * -> *)

User-visible class, exported:

class Functor (f :: * -> *) where
  fmap :: (a -> b) -> f a -> f b
  default fmap :: (Generic1 f, GFunctor (Rep1 f)) => (a -> b) -> f a -> f b
  fmap f = to1 . gfmap f . from1  

Defined by the generic programmer, not exported:

class GFunctor (f :: Universe (* -> *) * m) where
  gfmap :: (a -> b) -> Interprt f a -> Interprt f b
instance GFunctor UU where
  gfmap _ U1 = U1
instance GFunctor PAR where
  gfmap f (Par1 a) = Par1 (f a)

instance GFunctor (KK i c) where
  gfmap _ (K1 a) = K1 a

instance (Functor f) => GFunctor (REC f) where
  gfmap f (Rec1 a) = Rec1 (fmap f a)

instance (GFunctor f) => GFunctor (MM m c f) where
  gfmap f (M1 a) = M1 (gfmap f a)

instance (GFunctor f, GFunctor g) => GFunctor (f :++: g) where
  gfmap f (L1 a) = L1 (gfmap f a)
  gfmap f (R1 a) = R1 (gfmap f a)

instance (GFunctor f, GFunctor g) => GFunctor (f :**: g) where
  gfmap f (a :*: b) = gfmap f a :*: gfmap f b

instance (Functor f, GFunctor g) => GFunctor (f :..: g) where
  gfmap f (Comp1 x) = Comp1 (fmap (gfmap f) x)

Note that previously Functor and GFunctor had exactly the same types. Now we can make clear what the difference between them is.

Example generic function: show (kind *, uses metadata)

User-visible class, exported:

class Show (a :: *) where
  show :: a -> String
  default show :: (Generic a, GShow (Rep a)) => a -> String
  show = gshow . from

Defined by the generic programmer, not exported:

class GShow (f :: Universe (* -> *) * m) where
  gshow :: Interprt f x -> String
instance GShow UU where
  gshow U1 = ""
instance (P.Show c) => GShow (KK i c) where
  gshow (K1 a) = a
instance (Datatype c, GShow f) => GShow (MM DD c f) where
  gshow (M1 x) = datatypeName (Proxy :: Proxy c) ++ " " ++ gshow x

The other cases do not add any further complexity.

Example datatype encoding: lists (derived by the compiler)

instance Generic [a] where
  type Rep [a] = MM DD DList 
                   (MM CC DList_Nil UU :++: 
                    MM CC DList_Cons (KK PP a :**: KK RR [a]))

  from [] = D1 (L1 (C1 U1))
  from (h:t) = D1 (R1 (C1 (Par0 h :*: Rec0 t)))
  to (D1 (L1 (C1 U1))) = []
  to (D1 (R1 (C1 (Par0 h :*: Rec0 t)))) = h:t
-- Metadata
data List_Meta = DList | DList_Nil | DList_Cons

Note that we use only one datatype; more correct would be to use 3, one for DList, another for the constructors, and yet another for the selectors (or maybe even n datatypes for the selectors, one for each constructor?) But we don't do that because Universe is polymorphic only over m, so a single metadata representation type. If we want a more fine-grained distinction then we would need more parameters in Universe, and also to split the MM case.

instance Datatype DList where datatypeName _ = "[]"


Even better would be to index the metadata representation types over the type they refer to. Something like:

  data family MetaTypes a -- kind polymorphic
  data instance MetaTypes [] = DList | DList_Nil | DList_Cons

But now we are basically asking for promotion of data families, since we want to use promoted DList. Also, the case for MM in Universe would then be something like:

  | MM MetaData (MetaTypes m) (Universe f x m)

But I'm not entirely sure about this.

GHC 8.0 and later

Type-level metadata encoding

Because what we've described so far is rather backwards-incompatible, we wanted to at least try to improve the encoding of metadata, which was currently rather clunky prior to GHC 8.0 (giving rise to lots of empty, compiler-generated datatypes and respective instances). We can accomplished that by changing M1 to keep the meta-information at the type level:

newtype M1 i (c :: Meta) f p = M1 { unM1 :: f p }

data Meta = MetaData Symbol Symbol  Bool
          | MetaCons Symbol FixityI Bool
          | MetaSel  Symbol SourceUnpackedness SourceStrictness DecidedStrictness

data Fixity  = Prefix  | Infix  Associativity Int
data FixityI = PrefixI | InfixI Associativity Nat

data Associativity = LeftAssociative
                   | RightAssociative
                   | NotAssociative

data SourceUnpackedness = NoSourceUnpackedness
                        | SourceNoUnpack
                        | SourceUnpack

data SourceStrictness = NoSourceStrictness
                      | SourceLazy
                      | SourceStrict

data DecidedStrictness = DecidedLazy
                       | DecidedStrict
                       | DecidedUnpack

Why did we need to add FixityI? Because Fixity does not promote. Yet, we wanted to expose Fixity to the user, not FixityI. Note that the meta-data classes remained mostly unchanged (aside from some enhancements to Datatype and Selector):

class Datatype d where
  datatypeName :: t d (f :: * -> *) a -> [Char]
  moduleName   :: t d (f :: * -> *) a -> [Char]
  packageName  :: t d (f :: * -> *) a -> [Char]
  isNewtype    :: t d (f :: * -> *) a -> Bool

class Constructor c where
  conName     :: t c (f :: * -> *) a -> [Char]
  conFixity   :: t c (f :: * -> *) a -> Fixity
  conIsRecord :: t c (f :: * -> *) a -> Bool

class Selector s where
  selName               :: t s (f :: * -> *) a -> [Char]
  selSourceUnpackedness :: t s (f :: * -> *) a -> SourceUnpackedness
  selSourceStrictness   :: t s (f :: * -> *) a -> SourceStrictness
  selDecidedStrictness  :: t s (f :: * -> *) a -> DecidedStrictness

But now, using the magic of singletons, we give one single instance for each of these classes, instead of having to instantiate them each time a user derives Generic:

instance (KnownSymbol n, KnownSymbol m, KnownSymbol p, SingI nt)
    => Datatype ('MetaData n m p nt) where
  datatypeName _ = symbolVal (Proxy :: Proxy n)
  moduleName   _ = symbolVal (Proxy :: Proxy m)
  packageName  _ = symbolVal (Proxy :: Proxy p)
  isNewtype    _ = fromSing  (sing  :: Sing nt)

instance (KnownSymbol n, SingI f, SingI r)
    => Constructor ('MetaCons n f r) where
  conName     _ = symbolVal (Proxy :: Proxy n)
  conFixity   _ = fromSing  (sing  :: Sing f)
  conIsRecord _ = fromSing  (sing  :: Sing r)

instance (SingI mn, SingI su, SingI ss, SingI ds)
    => Selector ('MetaSel mn su ss ds) where
  selName               _ = fromMaybe "" (fromSing (sing :: Sing mn))
  selSourceUnpackedness _ = fromSing (sing :: Sing su)
  selSourceStrictness   _ = fromSing (sing :: Sing ss)
  selDecidedStrictness  _ = fromSing (sing :: Sing ds)

Naturally, we require singletons for Bool, Maybe, FixityI, Associativity, SourceUnpackedness, SourceStrictness, and DecidedStrictness, but that is one time boilerplate code, and is not visible for the user. (In particular, this is where we encode that the demotion of (the kind) FixityI is (the type) Fixity.)

I believe this change is almost fully backwards-compatible, and lets us simplify the code for deriving Generic in GHC. Furthermore, I suspect it will be useful to writers of generic functions, who can now match at the type-level on things such as whether a constructor is a record or not.

I say "almost fully backwards-compatible" because handwritten Generic instances might break with this change. But we've never recommended doing this, and I think users who do this are more than aware that they shouldn't rely on it working across different versions of GHC.


Before GHC 8.0, the following declaration:

data Example = Example deriving Generic1

Would have generated all of this:

instance Generic Example where
  type Rep Example = D1 D1Example (C1 C1_0Example (S1 NoSelector U1))

data D1Example
data C1_0Example

instance Datatype D1Example where
  datatypeName _ = "Example"
  moduleName   _ = "Module"
  isNewtype    _ = False

instance Constructor C1_0Example where
  conName     _ = "Example"
  conFixity   _ = Prefix
  conIsRecord _ = False

But on GHC 8.0 and later, this is all that is generated (assuming it was compiled with no strictness optimizations):

instance Generic Example where
  type Rep Example =
    D1 ('MetaData "Example" "Module" "package" 'False)
      (C1 ('MetaCons "Example" 'PrefixI 'False)
        (S1 ('MetaSel 'Nothing 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy)

Not bad!


The Selector class now looks like this:

class Selector s where
  selName               :: t s (f :: * -> *) a -> [Char]
  selSourceUnpackedness :: t s (f :: * -> *) a -> SourceUnpackedness
  selSourceStrictness   :: t s (f :: * -> *) a -> SourceStrictness
  selDecidedStrictness  :: t s (f :: * -> *) a -> DecidedStrictness

data SourceUnpackedness = NoSourceUnpackedness
                        | SourceNoUnpack
                        | SourceUnpack

data SourceStrictness = NoSourceStrictness
                      | SourceLazy
                      | SourceStrict

data DecidedStrictness = DecidedLazy
                       | DecidedStrict
                       | DecidedUnpack

This design draws much inspiration from the way Template Haskell handles strictness as of GHC 8.0 (see here for what motivated the change). We make a distinction between the source strictness annotations and the strictness GHC actually decides during compilation. To illustrate the difference, consider the following data type:

data T = T {-# UNPACK #-} !Int
  deriving Generic

If we were to encode the source unpackedness and strictness of each of T's fields, they were be SourceUnpack/SourceStrict, NoSourceUnpackedness/SourceStrict, and NoSourceUnpackedness/NoSourceStrictness, no matter what. Source unpackedness/strictness is a purely syntactic property.

The strictness that the user writes, however, may be different from the strictness that GHC decides during compilation. For instance, if we were to compile T with no optimizations, the decided strictness of each field would be DecidedStrict, DecidedStrict, and DecidedLazy. If we enabled -O2, however, they would be DecidedUnpack, DecidedStrict, and DecidedLazy.

Things become even more interesting when -XStrict and -O2 are enabled. Then the strictness that GHC would decided is DecidedUnpack, DecidedStrict, and DecidedStrict. And if you enable -XStrict, -O2, and -funbox-strict-fields, then the decided strictness is DecidedUnpack, DecidedUnpack, and DecidedUnpack.

The variety of possible DecidedStrictness combinations demonstrates that strictness is more just annotation—it's also a dynamic property of the compiled code. Both facets of strictness can be useful to a generic programmer, so we include both in the above design. That way, you get more bang for your buck.

One key difference between the way Template Haskell handles strictness and the way GHC generics would handle strictness is that in the former, source and decided strictness are decoupled, whereas the latter lumps them all into MetaSel. This is a result of Template Haskell allowing programmers to splice in TH ASTs directly into Haskell source code. Imagine if you wanted to splice in a datatype that had not been declared yet (i.e., one GHC had not compiled). You'd probably write something like this

$([d| ... Bang NoSourceUnpackedness NoSourceStrictness ??? ... |])

and become stuck. What do you put in place of ???? After all, you'd need a decided strictness, but there's no way to know what GHC had decided yet—the datatype hasn't been compiled!

GHC generics does not suffer from this problem, since you can only reify metadata with generics, not splice metadata back into code. Therefore, we can bundle both source and decided strictness with MetaSel without issue.

Using standard deriving for generic functions

Currently, users can instantiate a generic function to their datatypes by doing the following:

  1. Attaching a deriving Generic clause to their datatype declaration:
    data MyDatatype = .... deriving Generic
  2. Giving an empty instance:
    instance GBinary MyDatatype

However, it would be more natural and concise to simply be able to write:

data MyDatatype = .... deriving (Generic, GBinary)

Ultimately, this might even allow us to replace the generated code by classes such as Eq, Ord, etc. with plain generic code.

The ticket for this request is #5462. We refer to this new feature as DeriveAnyClass.

Which classes can be derived?

How do we figure out which classes will now be allowed as part of a deriving clause? We don't; we allow any class! Although the feature makes more sense for classes whose `MINIMAL` set is empty, it'll work for any other class too. It simply generates an instance without any method definitions.

Interaction with GeneralizedNewtypeDeriving

GeneralizedNewtypeDeriving (GND) already allows many non-standard classes to be derived, but:

  1. It means something different from ordinary deriving. Instead of generating new code for the instance, GHC simply uses the instance for the type inside the newtype.
  2. As such, this only works when deriving newtypes, and
  3. The classes Read, Show, Typeable, Data, and Generic even with GND on, give rise to standard instances. That is, in the following deriving clause
    newtype MyInt = MyInt Int deriving (Eq, Num, Show)

Eq and Num operations will simply be lifted through the newtype coercion, but show (MyInt 3) will still print MyInt 3, not just 3.

GND interacts with DeriveAnyClass because both are about generating instances for arbitrary classes, but they do it in different ways. There are multiple ways to handle this.

1. Disallow DeriveAnyClass for newtypes

In this case we simply do not allow GeneralizedDeriving for newtypes. It's simple to implement, but it's a bit limiting, in that if you happen to be using newtypes for which you want generic instances, you have to go back to manually writing down the empty instances.

2. Try both approaches, fail if there's ambiguity

In many cases it will be clear whether the user wants GND or the new DeriveAnyClass:

  1. If GND is on, and DeriveAnyClass is off, then we do GND.
  2. Conversely, if GND is off, and DeriveAnyClass is on, then we do the latter.
  3. If both are on, we pick the DeriveAnyClass approach.

The problem with this approach is that in situation 3 users are left in a bad situation if they actually wanted GND; they'll have to manually write a Coercible instance, which is not nice.

3. Let class authors specify how their classes should be derived

Right now we already treat some classes specially for GND: Read, Show, Typeable, Data, and Generic are derivable, but not via GND. We could make it a property of the class to say "do not use GND" in deriving clauses. In that case, we know that we can use the DeriveAnyClass strategy for newtypes instead.

Maybe we don't even need to make this user-controllable; a handful of built-in classes may suffice.

4. Let users annotate their deriving clauses

We could even let users specify which of the two types of deriving they want:

newtype MyInt = MyInt Int
  deriving newtype  GBinary
  deriving anyclass GShow

Here, we use the keywords newtype and anyclass to specify that the GBinary instance should be created using GND (so that we do not have the unnecessary extra tag in the binary encoding), but GShow should use the "generic" instance (so that we still print "MyInt"). See the deriving strategies page for information.

Standard vs. Standalone deriving

There's no reason to restrict this new deriving form to be attached to a datatype. Users should be free to write the following, if they wish:

deriving instance (GBinary a) => GBinary (Maybe a)

... even though the following would do exactly the same with 9 characters less:

instance (GBinary a) => GBinary (Maybe a)

Figuring out the context

The only non-trivial part of actually generating the instances from a deriving clause is that now we have to figure out the context. Previously the user would have written

data MyMaybe a ... deriving Generic
instance (GBinary a) => GBinary (MyMaybe a)

Now they'll simply write

data MyMaybe a ... deriving (Generic, GBinary)

So it's now up to us to find out that the instance needs a GBinary a => head.

For this we propose using the already existing strategy in GHC. Consider the following case:

class GBinary  (a :: *)      where ...
class GFunctor (f :: * -> *) where ...

data MyMaybe a ... deriving (Generic, GBinary, GFunctor)

The instance head for the GBinary (MyMaybe a) instance will be guessed in the same way as if we were deriving an Eq instance. The instance head for the GFunctor MyMaybe instance will be guessed in the same way as if we were deriving a Functor instance.

The context might end up not being exactly what the user intended. For example, the class being derived might not require instances for each of the type arguments of the datatype. We could try to figure this out in a clever way from the definition of the class being derived, but this is very hard in general. So, in this case, we just leave the user to specify the context manually via standalone deriving (or, well, just giving an empty instance with a context, as they had to do before).

Handling unlifted types

A weakness of generics (prior to GHC 8.0) was that it wasn't expressive enough to represent types of kind #. Part of the reason was due to Rec0 expecting a type of kind *, not kind #.

Having a generic # representation would be desirable to mimic GHC's built-in support for certain unlifted types when deriving Eq, Ord, and Show:

data IntHash = IntHash Int# deriving (Eq, Ord, Show)

-- Wouldn't have been allowed!
deriving instance Generic IntHash

Attempt 1: re-use Rec0

One especially simple approach to this problem would be to mark occurrences of Int# with Rec0 Int, Char# with Rec0 Char, etc. After all, data Int = I# Int#, so it would be simple to wrap up an occurrence of Int# with I#.

This approach suffers from a number of issues:

  • Ambiguity. When we encounter Rec0 Int, does it refer to Int or Int#? Without giving K1 a different tag, there'd be no way to distinguish them.
  • Legacy issues. Because of the ambiguity of the representation, previously written generic code (which was not aware of unlifted types) will start to work with data types that have unlifted arguments. This may not be desirable. For example, a programmer may not want their class to support datatypes with unlifted arguments at all.

Attempt 2: introduce new data types

To avoid reusing Int, Char, etc., we could introduce new data types such as data UInt = UInt Int#, data UChar = UChar Char#, etc. to represent occurrences of Int#, Char#, etc. in generic code. This would avoid any ambiguity issues, because we now have datatypes with the sole purpose of boxing unlifted types in generic code.

This approach is still lacking, however, because UInt, UChar, etc. aren't really representation types. To illustrate what that means, consider a simple generic program:

class GShow f where
    gshow :: f a -> String


instance Show c => GShow (K1 c) where
    gshow (K1 x) = gshow x

data Example = Example Int# deriving Generic
instance Show Example where show = gshow . from

When GHC creates the Generic Example instance, UInt appears as a type parameter of K1 in Example's representation type. This means that one can't use GShow to define how UInts should be shown. Rather, one must use Show directly! This is a bit awkward, since we'd like to provide a Show UInt instance from GHC.Generics directly, but that would enforce how all generic programmers must show UInt.

Attempt 3: separate unlifted representation types

Instead of using Rec0 at all, let's create standalone representation types to mark occurrences of unlifted types:

data UInt  (p :: *) = UInt  Int#
data UChar (p :: *) = UChar Char#

This solves the problems with the previous two approaches, since (1) unlifted types are now completely separate cases from Rec0, and (2) we can properly use a GShow UInt instance (as an example).

There's a thorny issue that arises with this method: how can we define generic code that behaves the same on every unlifted data type? For example, what if a programmer wanted the string "<unlifted>" to show whenever any unlifted data type is encountered? She could write this:

instance GShow UInt where
    gshow _ = gshowUnlifted

instance GShow UChar where
    gshow _ = gshowUnlifted


gshowUnlifted :: String
gshowUnlifted = "<unlifted>"

But then she'd be writing gshow _ = gshowUnlifted for every single unlifted data type under consideration. That's more boilerplate than any reasonable generic programmer should have to deal with!

Attempt 4: use a GADT

We could reformulate UInt, UAddr, etc. into a single GADT:

data URec :: * -> * -> * where
    UInt  :: Int#  -> URec Int  p
    UChar :: Char# -> URec Char p

Now the GShow example above can be dramatically simplified:

instance GShow (URec a) where
    gshow _ = "<unlifted>"

From a conceptual standpoint, however, a GADT doesn't quite capture the notion of what URec is trying to represent. A GADT implies a closed set of things that can be exhaustively pattern matched, but we certainly aren't encoding every unlifted data type in existence into URec! On the contrary, we deliberately only choose a handful of unlifted types to achieve parity with GHC's built-in deriving. Imagine if we were to support more unlifted types in the future—code which pattern-matched on a URec value would no longer be exhaustive!

Current design: data family instances

To convey the property that URec is "open", we use a data family:

data family URec a p
data instance URec (Ptr ()) p = UAddr   { uAddr#   :: Addr#   }
data instance URec Char     p = UChar   { uChar#   :: Char#   }
data instance URec Double   p = UDouble { uDouble# :: Double# }
data instance URec Int      p = UInt    { uInt#    :: Int#    }
data instance URec Float    p = UFloat  { uFloat#  :: Float#  }
data instance URec Word     p = UWord   { uWord#   :: Word#   }

type UAddr   = URec (Ptr ())
type UChar   = URec Char
type UDouble = URec Double
type UFloat  = URec Float
type UInt    = URec Int
type UWord   = URec Word

With this approach, it is still possible to write an all-encompassing unlifted type case:

instance GShow (URec a) where
    gshow _ = "<unlifted>"

Or, if a programmer desires, she can write type-specific cases:

instance GShow UInt where
    gshow (UInt i) = show (I# i)

instance GShow UChar where
    gshow (UChar c) = show (C# c)


One downside is that data families are conceptually more complicated than any of the tricks we've used in previous approaches, which may add some cognitive overhead for generic programmers. What is clever about this design, however, is that if a programmer wishes, he can simply use the provided type synonyms and pretend that the URec family doesn't exist!


The following declaration:

data IntHash = IntHash Int# deriving Generic


instance Generic IntHash where
  type Rep IntHash =
    D1 ('MetaData "IntHash" "Module" "package" 'False)
      (C1 ('MetaCons "IntHash" 'Prefix 'False)
        (S1 'MetaNoSel UInt))

Compilation performance tricks

Unfortunately, deriving Generic has been known to incur large compilation times. It is suspected that deriving Generic has both nonlinear behavior as well as a large constant overhead (see #5642 for further discussion). This section discusses some of the tricks GHC developers have used to make deriving Generic faster.

Factoring out M1

If you have a datatype like:

data Letter = A | B | C | D

then a naïve Generic instance for Letter would be:

instance Generic Letter where
  type Rep Letter = D1 (MetaData ...) ...

  to (M1 (L1 (L1 (M1 U1)))) = A
  to (M1 (L1 (R1 (M1 U1)))) = B
  to (M1 (R1 (L1 (M1 U1)))) = C
  to (M1 (R1 (R1 (M1 U1)))) = D

  from A = M1 (L1 (L1 (M1 U1)))
  from B = M1 (L1 (R1 (M1 U1)))
  from C = M1 (R1 (L1 (M1 U1)))
  from D = M1 (R1 (R1 (M1 U1)))

Notice that in every LHS pattern-match of the to definition, and in every RHS expression in the from definition, the topmost constructor is M1. This corresponds to the datatype-specific metadata (the D1 in the Rep Letter instance). But this is wasteful from a typechecking perspective, since this definition requires GHC to typecheck an application of M1 in every single case, leading to an O(n) increase in the number of coercions the typechecker has to solve, which in turn increases allocations and degrades compilation speed.

Luckily, since the topmost M1 has the exact same type across every case, we can factor it out reduce the typechecker's burden:

instance Generic Letter where
  type Rep Letter = D1 (MetaData ...) ...

  to (M1 x) = case x of
    L1 (L1 (M1 U1)) -> A
    L1 (R1 (M1 U1)) -> B
    R1 (L1 (M1 U1)) -> C
    R1 (R1 (M1 U1)) -> D

  from x = M1 (case x of
    A -> L1 (L1 (M1 U1))
    B -> L1 (R1 (M1 U1))
    C -> R1 (L1 (M1 U1))
    D -> R1 (R1 (M1 U1)))

A simple change, but one that pays off, since it turns an O(n) amount of coercions to an O(1) amount.

Last modified 7 months ago Last modified on Aug 2, 2016 10:46:35 PM