constrainedcategories alternatives and similar packages
Based on the "Control" category.
Alternatively, view constrainedcategories alternatives based on common mentions on social networks and blogs.

fusedeffects
A fast, flexible, fused effect system for Haskell 
transient
A full stack, reactive architecture for general purpose programming. Algebraic and monadically composable primitives for concurrency, parallelism, event handling, transactions, multithreading, Web, and distributed computing with complete deinversion of control (No callbacks, no blocking, pure state) 
apecs
a fast, extensible, type driven Haskell ECS framework for games 
recursionschemes
Generalized bananas, lenses and barbed wire 
capability
Extensional capabilities and deriving combinators 
distributedclosure
Serializable closures for distributed programming. 
classyprelude
Type classes for mapping, folding, and traversing monomorphic containers 
classypreludeyesod
Type classes for mapping, folding, and traversing monomorphic containers 
unliftio
The MonadUnliftIO typeclass for unlifting monads to IO 
extensibleeffects
Extensible Effects: An Alternative to Monad Transformers 
these
An eitherorboth data type, with corresponding hybrid error/writer monad transformer. 
auto
Haskell DSL and platform providing denotational, compositional api for discretestep, locally stateful, interactive programs, games & automations. http://hackage.haskell.org/package/auto 
safeexceptions
Safe, consistent, and easy exception handling 
selective
Selective Applicative Functors: Declare Your Effects Statically, Select Which to Execute Dynamically 
ComonadSheet
A library for expressing "spreadsheetlike" computations with absolute and relative references, using fixedpoints of ndimensional comonads. 
hask
Category theory for Haskell with a lens flavor (you need GHC 7.8.3, not 7.8.2 to build this!) 
abstractpar
Type classes generalizing the functionality of the 'monadpar' library. 
transientuniverse
A Cloud monad based on transient for the creation of Web and reactive distributed applications that are fully composable, where Web browsers are first class nodes in the cloud 
cloudhaskell
This is an umbrella development repository for Cloud Haskell 
distributedfork
A distributed data processing framework in Haskell. 
distributedprocessplatform
DEPRECATED (Cloud Haskell Platform) in favor of distributedprocessextras, distributedprocessasync, distributedprocessclientserver, distributedprocessregistry, distributedprocesssupervisor, distributedprocesstask and distributedprocessexecution 
monadcontrol
Lift control operations, like exception catching, through monad transformers 
monadvalidate
(NOTE: REPOSITORY MOVED TO NEW OWNER: https://github.com/lexilambda/monadvalidate) A Haskell monad transformer library for data validation 
ixmonad
Provides 'graded monads' and 'parameterised monads' to Haskell, enabling finegrained reasoning about effects. 
freereffects
An implementation of "Freer Monads, More Extensible Effects". 
effectmonad
Provides 'graded monads' and 'parameterised monads' to Haskell, enabling finegrained reasoning about effects. 
operational
Implement monads by specifying instructions and their desired operational semantics. 
monadtime
Type class for monads which carry the notion of the current time. 
lenstutorial
The missing tutorial module for the lens library
Access the most powerful time series database as a service
Do you think we are missing an alternative of constrainedcategories or a related project?
README
Before browsing the Hackage documentation (which looks a bit scary in places), you may want to read this first.
About
The introduction of ConstraintKinds
in GHC has mainly received attention for making one type of issue easy to solve.
The perhaps most obvious example, as given in the original blog post on ConstraintKinds
, is a monad instance for Set
:
instance RMonad S.Set where
type RMonadCtxt S.Set a = Ord a
return = S.singleton
mx >>= fxmy = S.fromList [y  x < S.toList mx, y < S.toList (fxmy x)]
where RMonad
is the Monad
class plus an added Constraint
RMonadCtxt
, which allows the Set
instance to constrict return
and >>=
to Ord a => S.Set a
rather than general forall a. S.Set a
.
This pattern of defining a "constrained version" of some existing type class adhoc (with a constrain specifically for types turning up in that class's methods), while applicable, is bound to lead to many incompatible instances. It does not really obey the mathematical concepts from category theory.
With this library we attempt to better things, by actually starting right from the Category
class. This is not a new idea, and has not intrinsically much to do with ConstraintKinds
. But it turns out the combination of redoing the categorytheory stuff with focus on ConstraintKinds
works quite well.
The outset
Haskell has, and makes great use of, powerful facilities from category theory – basically various variants of functors.
However, all those are just endofunctors in Hask, the category of all Haskell types with functions as morphisms. Which is sufficient for container / control structures that you want to be able to handle any type of data, but otherwise a bit limiting, seeing as there are (in maths, science etc.) many categories that cannot properly be represented this way. Commonly used libraries such as vectorspace thus make little use of the fact that the objects they deal with actually form a category. Instead, just specialised versions of the operations are defined.
Haskell sure can do better than that! package/categories rolls up the category theory much more rigorously, with all the correct and general notions as would be used in a mathematical treatment. Functors are not limited to any particular category, functions are generalised to closed cartesian category morphisms, and rather than just having a monad type class thown in you get a finegrained hierarchy and define monoidal endofunctors. datacategory is even more elaborate, defining functors themselves as morphisms in the category Cat.
While this may be in principle the ideal way to go about this, it is neither simple to learn nor adapt to actual Haskell programs, with their somewhat sloppy but effectiveproven notion of functors and monads in particular.
On the other side of the spectrum, there are packages like ConstraintKinds, which focus on just allowing more instances of the functor through monad classes to be defined.
This allows for some practically useful things like the Monad
instance of the standard Set
container as discussed above, without introducing much to be learned or changed about existing Haskell programs; but it hardly generalises the classes themselves and certainly does not open up new opportunities for using specific categories in a canonical manner. Moreover, it is not readily clear how exactly these generalised classes relate to the old plain Hask endofunctors.
The solution
This package attempts to offer a good compromise between the two approaches. It is very similar to packages/hask or packages/subhasksee note below, but keeps a lot closer to the terminology used in the base
modules.
Two crucial changes are made, relative to the established standard classes:
 Functors are parameterised on the two categories they map between.
If both are
(>)
, we have an ordinary Haskendofunctor and all works like what we're used to. In fact, this parameterisation allows there to be a safe (nonoverlapping) generic instance: all instancesPrelude.Functor f
give rise toinstance Control.Functor.Constrained.Functor f (>) (>)
. Likewise forApplicative
andMonad
. As a result, it is trivial to adapt Haskell code to the classes from this package if you just import the "Constrained
" modules. Any of the numerous monad etc. instances, from any package, can immediately be used as Haskspecific instances of the generalMonad
class.
But functors can as well map between other categories!
 A category is classified by its morphisms. The objects on the other hand remain ordinary types of kind
*
, but with one important constraint, namelyObject
. This allows a category to narrow its objects down to e.g. totally ordered types, or vector spaces, or whatever your morphisms and functors require.
Short HowTo
First on importing the modules. The classes are designed to have good backwardscompatibility, so you basically don't need the Prelude
versions at all anymore. Here's the recommended import list:
import Prelude ()
import Control.Category.Constrained.Prelude
import qualified Control.Category.Hask as Hask
import Control.Monad.Constrained
Most of your existing code should still work after this import change. The compiler will silently replace the Prelude.>>=
operators with Control.Monad.Constrained.>>=
and so on, which, though the signatures look much more complicated at first glance, are completely equivalent within the (>)
category.
What will not work is if you
 Write generic
Functor
/Applicative
/Monad
routines with explicit signatures. Those classes are nowMultiParamTypeClasses
; if you put them in a constraint you need to add what category you desire; so if it's Hask you should write(Functor f (>) (>), Num n) => ...
. Define your own instances for these classes. You might now be tempted to make it
instance Functor YourFunctor (>) (>) where fmap = ...
however, this would give an OverlappingInstances
clash with the generic instance: any old (or new!) Hask endofunctor is automatically a Functor f (>) (>)
. So the correct way is actually to define
instance Hask.Functor YourFunctor where
Hask.fmap = ...
where Hask.Functor
is just the plain old Functor
class from the prelude, as reimported through Control.Category.Hask
.
As for actually doing stuff you couldn't have also done with Prelude.Monad
etc. – There are mainly two such things:
Constrain a given category to a particular class of objects.
This would again apply to the Set
example, where you need the objects in a set to be Ord
.
The Category
module includes the ConstrainedCategory
newtype wrapper for this purpose. The functor classes have easy helper functions to facilitate defining instances withing such a constrained class. Look up the Set
example to see how to do it.
Define categories where the morphisms aren't just Haskell functions but some other type of arrow.
If you have an entirely new category,
data YourCat a b = ...
and want to use it with this library, you need to define a couple of class instances. The classes in this library form a rather fine hierarchy, much finer than the standard Arrow
; here's a guide you may use to determine which instances to define and which not, for your category:
Category
corresponds exactly to the homonymous standard class.Cartesian
allows to do “obvious” transformations on tuples, like swapping elements. You probably want this for almost all categories you define.Curry
does what it says: you can curry and uncurry morphisms. This is actually a pretty strong operation, only possible in cartesian closed categories. You may want to skip this for the beginning.Morphism
represents one half of the standardArrow
class. This is still almost as basic asCartesian
.PreArrow
is the other, substantially stronger half. Unlike withMorphism
, some interesting categories aren't instances of this class, namely all categories which are “informationpreserving”, sincePreArrow
allows both duplicating (&&&
) and “forgetting” (fst
/snd
) information. In particular, this class therefore can't contain truely invertible arrows.WellPointed
is yet more specific; together withCurry
it completes cartesian closedness.
CoCartesian
, likeCartesian
, expresses obvious isomorphisms, but on sum types (i.e.Either
) instead of products (tuples). This also includes booleans andMaybe
, which Haskell traditionally expresses as separate ADTs, though they are basically just specific instantiations ofEither
.MorphChoice
is dual toMorphism
.PreArrChoice
is dual toPreArrow
. A class that's all ofWellPointed
,PreArrChoice
andCurry
is bicartesian closed, an extremely strong property – such categories can essentially do anything you can do in Hask, or generally in lambda calculi.
SPDistribute
combines the product and sumclasses in the algebraic sense: it offers the isomorphism a · (b + c) = a · b + a · c, in Haskell(a, Either b c) <> Either (a,b) (a,c)
.HasAgent
, [CartesianAgent
](hackage.haskell.org/package/constrainedcategories0.2.0.0/docs/ControlArrowConstrained.html#t:CartesianAgent) and [PointAgent
](hackage.haskell.org/package/constrainedcategories0.2.0.0/docs/ControlArrowConstrained.html#t:PointAgent) have little to do with category theory as such; they just offer a useful little eDSL for defining arrows in general categories in a similar way as you would in Hask, by “pretending” the “results” of your computation are Hask objects. The idea is similar to the syntactic sugar offered by arrow notation.
Pros & cons
This project differs in particular from Hask in that the constraints are used explicitly in all the generic functions which somehow compose morphisms. That, frankly, makes the signatures of even simple generic operations look awful, because they internally use compositions with tuples and functorapplications. For instance,
guard :: (MonadPlus m k, Arrow k (>), Function k, UnitObject k ~ (), Object k Bool)
=> Bool `k` m ()
forever :: (Monad m k, Function k, Arrow k (>), Object k a, Object k b, Object k (m a), Object k (m (m a)), ObjectPoint k (m b), Object k (m (m b)))
=> m a `k` m b
The good thing is that all these constraints can be automatically inferred (and, hopefully, inlined & optimised) by the compiler. For any actual instantiation, you shouldn't need to worry, because these combined constraints are trivially fulfilled if so are the components.
Trying to generalise random existing code to the classes in this package works empirically kind of well (see this example), only when writing more generic stuff the compiler will ask for a lot of things along the lines Could not infer Object k (c (m a, Bool)) from ...
. This can generally be fixed easily, pretty much by pasting the requested constraints in the signature. Obviously though, this isn't altogether nice and scales badly, in terms of codeamount.
In Hask, the Op
constraint is only required for generating “new” arrows; the general philosophy seems to be to pack constraints in GADT dictionaries as much a possible. All in all, that is probably the better solution, but off the top it does require more structure (e.g. observe
) that differs substantially from the standard modules.
So all in all, this project may be seen as a bruteforce hack, generalising the traditional baseclasses as much as possible without actually changing the interface.
Deprecation
The subhask
package has a very similar focus and tackles the subject in an arguably more consequent manner. It is very possible that this package will in the future be deprecated in favour of subhask, but probably not while that package has its own problems. In particular, it depends on hmatrix, which I consider quite problematic for such a generalpurpose project.