Vinyl is a general solution to the records problem in Haskell using type level strings and other modern GHC features, featuring static structural typing (with a subtyping relation), and automatic row-polymorphic lenses. All this is possible without Template Haskell.

Let's work through a quick example. We'll need to enable some language extensions first:

>>> :set -XDataKinds
>>> :set -XPolyKinds
>>> :set -XTypeApplications
>>> :set -XTypeOperators
>>> :set -XTypeFamilies
>>> :set -XFlexibleContexts
>>> :set -XFlexibleInstances
>>> :set -XNoMonomorphismRestriction
>>> :set -XGADTs
>>> :set -XTypeSynonymInstances
>>> :set -XTemplateHaskell
>>> :set -XStandaloneDeriving

>>> import Data.Vinyl
>>> import Data.Vinyl.Functor
>>> import Control.Applicative
>>> import Control.Lens hiding (Identity)
>>> import Control.Lens.TH
>>> import Data.Char
>>> import Test.DocTest
>>> import Data.Singletons.TH (genSingletons)
>>> import Data.Maybe

Let's define a universe of fields which we want to use.

First of all, we need a data type defining the field labels:

>>> data Fields = Name | Age | Sleeping | Master deriving Show

Any record can be now described by a type-level list of these labels. The DataKinds extension must be enabled to automatically turn all the constructors of the Field type into types.

>>> type LifeForm = [Name, Age, Sleeping]

Now, we need a way to map our labels to concrete types. We use a type family for this purpose. Unfortunately, type families aren't first class in Haskell. That's why we also need a data type, with which we will parametrise Rec. We also generate the necessary singletons for each field label using Template Haskell.

>>> :{
type family ElF (f :: Fields) :: * where
  ElF Name = String
  ElF Age = Int
  ElF Sleeping = Bool
  ElF Master = Rec Attr LifeForm
newtype Attr f = Attr { _unAttr :: ElF f }
makeLenses ''Attr
genSingletons [ ''Fields ]
instance Show (Attr Name) where show (Attr x) = "name: " ++ show x
instance Show (Attr Age) where show (Attr x) = "age: " ++ show x
instance Show (Attr Sleeping) where show (Attr x) = "sleeping: " ++ show x
instance Show (Attr Master) where show (Attr x) = "master: " ++ show x
:}

To make field construction easier, we define an operator. The first argument of this operator is a singleton - a constructor bringing the data-kinded field label type into the data level. It's needed because there can be multiple labels with the same field type, so by just supplying a value of type ElF f there would be no way to deduce the correct "f".

>>> :{
let (=::) :: sing f -> ElF f -> Attr f
    _ =:: x = Attr x
:}

Now, let's try to make an entity that represents a human:

>>> :{
let jon = (SName =:: "jon")
       :& (SAge =:: 23)
       :& (SSleeping =:: False)
       :& RNil
:}

Automatically, we can show the record:

>>> print jon
{name: "jon", age: 23, sleeping: False}

And its types are all inferred with no problem. Now, make a dog! Dogs are life-forms, but unlike humans, they have masters. So, let’s build my dog:

>>> :{
let tucker = (SName =:: "tucker")
          :& (SAge =:: 9)
          :& (SSleeping =:: True)
          :& (SMaster =:: jon)
          :& RNil
:}

Now, if we want to wake entities up, we don't want to have to write a separate wake-up function for both dogs and humans (even though they are of different type). Luckily, we can use the built-in lenses to focus on a particular field in the record for access and update, without losing additional information:

>>> :{
let wakeUp :: (Sleeping ∈ fields) => Rec Attr fields -> Rec Attr fields
    wakeUp = rput $ SSleeping =:: False
:}

Now, the type annotation on wakeUp was not necessary; I just wanted to show how intuitive the type is. Basically, it takes as an input any record that has a Bool field labelled sleeping, and modifies that specific field in the record accordingly.

>>> let tucker' = wakeUp tucker
>>> let jon' = wakeUp jon

>>> tucker' ^. rlens @Sleeping
sleeping: False

>>> tucker ^. rlens @Sleeping
sleeping: True

>>> jon' ^. rlens @Sleeping
sleeping: False

We can also access the entire lens for a field using the rLens function; since lenses are composable, it’s super easy to do deep update on a record:

>>> let masterSleeping = rlens @Master . unAttr . rlens @Sleeping
>>> let tucker'' = masterSleeping .~ (SSleeping =:: True) $ tucker'

>>> tucker'' ^. masterSleeping
sleeping: True

A record Rec f xs is a subtype of a record Rec f ys if ys ⊆ xs; that is to say, if one record can do everything that another record can, the former is a subtype of the latter. As such, we should be able to provide an upcast operator which "forgets" whatever makes one record different from another (whether it be extra data, or different order).

Therefore, the following works:

>>> :{
let upcastedTucker :: Rec Attr LifeForm
    upcastedTucker = rcast tucker
:}

The subtyping relationship between record types is expressed with the <: constraint; so, rcast is of the following type:

rcast :: r1 <: r2 => Rec f r2 -> Rec f r1

Also provided is a "≅" constraint which indicates record congruence (that is, two record types differ only in the order of their fields).

In fact, rcast is actually given as a special case of the lens rsubset, which lets you modify entire (possibly non-contiguous) slices of a record!

Consider the following declaration:

data Rec :: (u -> *) -> [u] -> * where
  RNil :: Rec f '[]
  (:&) :: f r -> Rec f rs -> Rec f (r ': rs)

Records are implicitly parameterized over a kind u, which stands for the "universe" or key space. Keys (inhabitants of u) are then interpreted into the types of their values by the first parameter to Rec, f. An extremely powerful aspect of Vinyl records is that you can construct natural transformations between different interpretation functors f,g, or postcompose some other functor onto the stack. This can be used to immerse each field of a record in some particular effect modality, and then the library functions can be used to traverse and accumulate these effects.

Let's imagine that we want to do validation on a record that represents a name and an age:

>>> type Person = [Name, Age]

We've decided that names must be alphabetic, and ages must be positive. For validation, we'll use Maybe for now, though you should use a left-accumulating Validation type (the module Data.Either.Validation from the either package provides such a type, though we do not cover it here).

>>> :{
let goodPerson :: Rec Attr Person
    goodPerson = (SName =:: "Jon")
              :& (SAge =:: 20)
              :& RNil
:}

>>> :{
let badPerson = (SName =:: "J#@#$on")
             :& (SAge =:: 20)
             :& RNil
:}

We'll give validation a (rather poor) shot.

>>> :{
let
    validatePerson :: Rec Attr Person -> Maybe (Rec Attr Person)
    validatePerson p = (\n a -> (SName =:: n) :& (SAge =:: a) :& RNil) <$> vName <*> vAge
      where
      vName = validateName $ p ^. rlens @Name . unAttr
      vAge  = validateAge $ p ^. rlens @Age . unAttr
      validateName str | all isAlpha str = Just str
      validateName _ = Nothing
      validateAge i | i >= 0 = Just i
      validateAge _ = Nothing
:}

Let's try it out:

>>> isJust $ validatePerson goodPerson
True

>>> isJust $ validatePerson badPerson
False

The results are as expected (Just for goodPerson, and a Nothing for badPerson); but this was not very fun to build.

Further, it would be nice to have some notion of a partial record; that is, if part of it can't be validated, it would still be nice to be able to access the rest. What if we could make a version of this record where the elements themselves were validation functions, and then that record could be applied to a plain one, to get a record of validated fields? That's what we’re going to do.

>>> type Validator f = Lift (->) f (Maybe :. f)

Let's parameterize a record by it: when we do, then an element of type a should be a function Identity a -> Result e a:

>>> :{
let lift f = Lift $ Compose . f
    validateName (Attr str) | all isAlpha str = Just (Attr str)
    validateName _ = Nothing
    validateAge (Attr i) | i >= 0 = Just (Attr i)
    validateAge _ = Nothing
    vperson :: Rec (Validator Attr) Person
    vperson = lift validateName :& lift validateAge :& RNil
:}

And we can use the special application operator <<*>> (which is analogous to <*>, but generalized a bit) to use this to validate a record:

>>> let goodPersonResult = vperson <<*>> goodPerson
>>> let badPersonResult  = vperson <<*>> badPerson

>>> isJust . getCompose $ goodPersonResult ^. rlens @Name
True

>>> isJust . getCompose $ goodPersonResult ^. rlens @Age
True

>>> isJust . getCompose $ badPersonResult ^. rlens @Name
False

>>> isJust . getCompose $ badPersonResult ^. rlens @Age
True

So now we have a partial record, and we can still do stuff with its contents. Next, we can even recover the original behavior of the validator (that is, to give us a value of type Maybe (Rec Attr Person)) using rtraverse:

>>> :{
let mgoodPerson :: Maybe (Rec Attr Person)
    mgoodPerson = rtraverse getCompose goodPersonResult
:}

>>> let mbadPerson  = rtraverse getCompose badPersonResult

>>> isJust mgoodPerson
True

>>> isJust mbadPerson
False