implicit-params: Named and unnamed implicit parameters with defaults.

[ bsd3, data, library ] [ Propose Tags ] [ Report a vulnerability ]

Named and unnamed implicit parameters with defaults using type classes and constraint hacks. For examples, see Data.Implicit.


[Skip to Readme]

Modules

[Index]

Downloads

Maintainer's Corner

Package maintainers

For package maintainers and hackage trustees

Candidates

  • No Candidates
Versions [RSS] 0.1, 0.2, 0.2.1
Dependencies base (>=4 && <5), data-default-class (>0 && <0.1) [details]
License BSD-3-Clause
Author Shane O'Brien
Maintainer shane@duairc.com
Category Data
Home page http://github.com/duairc/implicit-params
Bug tracker http://github.com/duairc/implicit-params/issues
Source repo head: git clone http://github.com/duairc/implicit-params
Uploaded by ShaneOBrien at 2013-12-23T18:52:41Z
Distributions
Reverse Dependencies 1 direct, 0 indirect [details]
Downloads 2562 total (8 in the last 30 days)
Rating (no votes yet) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Docs available [build log]
Successful builds reported [all 1 reports]

Readme for implicit-params-0.2.1

[back to package description]

Implicit parameters

There are certain types of applications which are configurable where it makes sense to model this configurability as a global or semi-global set of configuration values that some or all parts of the program can "implicitly" access. These configuration values are called "implicit parameters".

ImplicitParams in Haskell

Haskell already has support for implicit parameters via the ImplicitParams extension. However, ImplicitParams has several flaws and is barely used at all in modern Haskell code. Many Haskellers consider its (ab)use of let/where bindings to pass implicit parameters to be ugly. Also, it's questionable how "implicit" the implicit parameters of ImplicitParams actually are, as they show up in the context of the type signature of any function which uses them. There's also no way you can call a function taking an implicit parameter without passing it that parameter if it isn't already in context: i.e., there is no way to specify a "default" value for an implicit parameter if none is passed.

Motivating example

implicit-params solves some of these problems and introduces new problems of its own. However, there is one particular use case which motivated me to develop implicit-params that isn't supported by the existing ImplicitParams extension. Imagine you have the following code:

app :: Config -> IO ()
app config = doStuffWith config

defaultConfig :: Config
defaultConfig = ...

Which is used by a program as follows:

main = app defaultConfig

There are two problems with this code. One is that app has to manually plumb the Config value it was given around everywhere. One solution here would be for app to use a Reader monad internally, but that can complicate the code in some ways and it seems like overkill. If it used the ImplicitParams extension, the above code would look like this:

app :: (?config :: Config) => IO ()
app = doStuffWith ?config

defaultConfig :: Config
defaultConfig = ...

main = let ?config = defaultConfig in app

You can see why ImplicitParams isn't very highly regarded: it made the type signature of app longer, as well as the code that uses it. However, at least the internals of app will be somewhat nicer now as the Config value won't have to be manually plumbed around everywhere.

data-default

The data-default package provides a type class Default which represents the class of types which have a "default" value. It has a single operation def which returns the default value for a given type (the type is given by type inference). Using Default the above code could be made a little nicer:

app :: (?config :: Config) => IO ()
app = doStuffWith ?config

instance Default Config where def = ...

main = let ?config = def in app

However, the above code is still so ugly considering what we're trying to do: all we want to do is run app with the defaults. This should be as simple as typing app, and only if we're overriding the defaults should the code need to be any longer than this. This is exactly what implicit-params does. If an implicit parameter is not explicitly given to a function which requires it, its value is given by def for the Default instance for the type of the parameter. And if the type does not have a Default instance, then it is a type error to call that function without explicitly setting the implicit parameter (but it will work fine if you do set it). This is how the above code looks using implicit-params:

app :: Implicit_ Config => IO ()
app = doStuffWith param_

instance Default Config where def = defaultConfig

main = app

Perfect! What if we want to pass a non-default config to app? That's easy too:

main = setParam_ (def {option = 1}) app

setParam_ even has an infix synonym $~ which makes the above code even nicer:

main = app $~ def {option = 1}

(Bonus points for not abusing let/where bindings.)

Named implicit parameters

The above code uses unnamed implicit parameters, which will suffice for most code. Sometimes you might want to pass more than one implicit parameter of the same type to a single function, and for this you need some way of selecting the particular implicit parameter on which to operate. implicit-params uses type level symbols for this, which require the DataKinds extension.

Implicit_, which is used in the examples above, denotes an unnamed implicit parameter. Implicit "foo" can be used to denote a named implicit parameter named "foo". Named implicit parameters are slightly more awkward to use because they require passing Proxy parameters to the param and setParam functions to specify the names of the implicit parameters on which they are to operate. See the Haddock documentation of the Data.Implicit module for more details.

Acknowledgements

This wouldn't be possible without techniques that I learnt from Edward Kmett and Philip JF. In particular, this package uses ideas from Edward's packages tagged, constraints and reflection and Philip's blog posts Haskell Supports First-Class Instances and Using Compiler Bugs for Fun and Profit: Introducing Cartesian Closed Constraints.