Haskell Tutorials, a tutorial

Main image

tl;dr: Some hints on how to make great documentation for Haskell libraries.

  1. Create a Tutorial module containing nothing except documentation.
  2. Mention the Tutorial module in your cabal description
  3. Use doctest to check your documentation is up to date
  4. For more complex real world examples, link to the source of some test.

Great documentation make a big difference. A bad documentation could simply make people not using your lib.

My friend was learning Haskell. To start he tried a Haskell library to make a small application. The documentation was deprecated to the point he wasn’t able to make a basic example work. How do you believe he felt? What does he thought about Haskell in general?

So here are my hint on how to make a great documentation in Haskell.

Documentation can take many different form.

  1. Tutorials/Guides – write some prose which friendly take a user by hand and help him
  2. Examples – how to use each function
  3. Generated API Documentation – haddock



  1. Create a new module named Tutorial (or Guide.GuideTopic)
  2. Create a link to the tutorial in the cabal description
  3. Create a link to the tutorial in your README
  4. Here is an example some Tutorial module content:

To prevent obsolescence of your tutorial, use doctest.

That way when you’ll do a stack test or cabal test you’ll get errors if some example doesn’t work anymore.

Examples (doctest)

doctest is a great way to provide examples in your code documentation. These example will then be used as tests. Apparently it comes from Python community.

To use doctest, this is very simple:

And to make it works simply verify you have a test bloc in your .cabal file looking like this:

test-suite doctest
  type: exitcode-stdio-1.0
  hs-source-dirs: test
  main-is: DocTest.hs
  build-depend: base >= 4.7 && < 5
              , <YOUR_LIBRARY> 
              , Glob >= 0.7
              , doctest >= 0.9.12

and in test/DocTest.hs simply use

Now stack test or cabal test will check the validity of your documentation.


Verifying documentation coverage

  1. Install haddock stack install haddock or cabal install haddock
  2. Launch haddock without output format:
> haddock src/**/*.hs
Haddock coverage:
 100% ( 15 / 15) in 'Data.Duration'
 100% (  3 /  3) in 'Data.Duration.Tutorial'

Continuous Integration

There are plenty of alternative solution. I provide the one I believe would be used by most people. So if you use github simply create an account on travis.

Add a .travis.yml file in your repo containing the content of the file here and remove the builds you don’t need. It will build your project using a lot of different GHC versions and environemnts.

If you are afraid by such its complexity you might just want to use this one:

Don’t forget to activate your repo in travis.

For some bonus points add the build status badge in your README.md file:

Congratulation! Now if you break your documentation examples, you’ll get notified.


You could add badges to your README.md file.

Here is a list of some: shields.io



If you didn’t declared your package to stackage, please do it. It isn’t much work. Just edit a file to add your package. And you’ll could be able to add another badge:

See Stackage Badges for more informations.

Creating a new project with stack

If you use stack I suggest you to use the tasty-travis template. It will include the boilerplate for:

  • tests
  • doctest
  • benchmark
  • travis CI
  • a README file to help you start

So edit your ~/.stack/config.yaml like this:

      author-name: Your Name
      author-email: your@mail.com
      copyright: 'Copyright: (c) 2016 Your Name'
      github-username: yourusername
      category: Development

And then you can create a new projec with:

stack new my-project tasty-travis

Generated Documentation

Even not doing anything, if you submit your library to hackage, haddock should generate some API documentation for free.

But to make real documentation you need to add some manual annotations.


-- | My function description
myFunction :: T1 -- ^ arg1 description
           -> T2 -- ^ arg2 description
myFunction arg1 arg2 = ...


data MyData a b
  = C1 a b -- ^ doc for constructor C1
  | C2 a b -- ^ doc for constructor C2

data MyData a b
  = C { a :: TypeA -- ^ field a description
      , b :: TypeB -- ^ field b description


Module    : MyModule
Description: Short description
Copyright : (c)
License : MIT

Here is a longer description of this module.
With some code symbol @MyType@.
And also a block of code:

data MyData = C Int Int

myFunction :: MyData -> Int


Documentation Structure:

module MyModule (
  -- * Classes
  -- * Types
  -- ** A data type
  -- ** A record
  -- * Some functions
  f, g
  ) where

That will generate headings.

Other Random Ideas

In Haskell we have great tools like hayoo! and hoogle.

And hackage and stackage provide also a lot of informations.

But generally we lack a lot of Tutorials and Guides. This post was an attempt to help people making more of them.

But there are other good ideas to help improve the situation.

In clojure when you create a new project using lein new my-project a directory doc is created for you. It contains a file with a link to this blog post:

Having a page by function/symbol with comments

If you try to search for some clojure function on a search engine there is a big chance the first result will link to:

  • clojuredocs.org: try to search for reduce, update-in or index for example

For each symbol necessiting a documentation. You don’t only have the details and standard documentation. You’ll also get:

  • Responsive Design (sometime you want to look at documentation on a mobile)
  • Contributed Examples
  • Contributed See Also section
  • Contributed notes/comments

clojuredocs.org is an independant website from the official Clojure website.

Most of the time, if you google the function you search you end up on clojuredocs for wich there are many contributions.

Currently stackage is closer to these feature than hackage. Because on stackage you have access to the README and also some comments by package.

I believe it would be more efficient to have at least a page by module and why not a page by symbol (data, functions, typeclasses…).

For example, we could provide details about foldl for example. Also as there would be less information to display, it will make the design cleaner.

Today, if you want to help documenting, you need to make a PR to the source of some library. While if we had an equivalent to clojuredocs for Haskell, adding documentation would simply be a few clicks away:

  1. login
  2. add/edit some example, comments, see-also section

There are more than 23k people on /r/haskell. If only 1% of them would take 10 minutes adding a bit of documentation it will certainly change a lot of things in the percieved documentation quality.

And last but not least,

Design is important

Design is Important

Design is a vague word. A good design should care not only about how something look, but also how users will interact with it. For example by removing things to focus on the essential.

When I stumble upon some random blog post or random specification in the Haskell community, I had too much a feeling of old fashioned design.

If you look at node.js community lot of their web page look cleaner, easier to read and in the end, more user friendly.

Haskell is very different from node, I wouldn’t like to replace all long and precise documentation with short human unprecise concepts. I don’t want to transform scientific papers by tweets.

But like the scientific community has upgraded with the use of LaTeX, I believe we could find something similar that would make, very clean environment for most of us. A kind of look and feel that will be

  • modern
  • device friendly (either on computer, mobile, tablet)
  • efficient, focus on what is most important and is helpful