doctest-extract: Alternative doctest implementation that extracts comments to modules

[ bsd3, program, testing ] [ Propose Tags ] [ Report a vulnerability ]

doctest-extract lets you write test examples and QuickCheck properties in Haddock comments and extracts them to test modules. It means that the user sees your tests in the documentation and knows that the examples and properties are machine-tested, or at least, she can run the tests herself.

I found the barrier to write tests much lower when I do not need to write new test modules but just add some lines to the Haddock comments. I do not need to think of test names or filling test data structures. The test identifier is the module name and the line number and if a test fails I can easily jump to the failing code.

Differences to the original GHCi-based implementation of doctest:

Pros:

  • Package versions for tests are consistent with tested library

  • Tests run much faster, especially QuickCheck property tests

  • No dependency on GHCi or GHC-as-library

  • The tested package need not be ready for compilation. Our simple parser requires only clearly recognizable Haskell comments.

  • QuickCheck properties do not cause confusing type error messages when actually only identifiers are missing.

  • You can inspect extracted modules

  • doctest collects tests from the transitive hull of imports of the specified modules. This might help you to keep the list of modules short. doctest-extract only processes the specified modules and thus allows you to focus on a module for development of tests.

  • With option --verbose test source path and line number are formatted such that Emacs allows you to click and jump to the test definition.

  • Report success only of real tests. doctest reports successful imports and definition of helper types and functions as successful tests. This makes it hard to monitor the number of real tests, e.g. whether some tests have been dropped by accident.

Cons:

  • Cannot test for output of IO functions or error messages from partial functions.

  • All free variables in QuickCheck properties must be all-quantified using lambda. (Could be even seen as an advantage for the reader of your doctests.)

  • No support for a single-line let (as in a do-block) as an example.

  • The Test module does not automatically import modules that the tested module imports. Thus, you usually have to add a setup section with required imports.

  • You need tools additional to Cabal, e.g. make and a Makefile, in order generate test modules.

See packages utility-ht, apportionment or pathtype for packages with working setups of doctest-extract.

Alternatives: cabal-docspec, cabal-doctest


[Skip to Readme]

Downloads

Maintainer's Corner

Package maintainers

For package maintainers and hackage trustees

Candidates

Versions [RSS] 0.1, 0.1.0.1, 0.1.1, 0.1.1.1, 0.1.2
Dependencies base (>=4.5 && <5), doctest-lib (>=0.0 && <0.2), non-empty (>=0.3.3 && <0.4), optparse-applicative (>=0.11 && <0.19), pathtype (>=0.8 && <0.9), semigroups (>=0.18.5 && <0.21), transformers (>=0.5.6 && <0.7), utility-ht (>=0.0.16 && <0.1) [details]
Tested with ghc ==7.4.2, ghc ==8.6.5
License BSD-3-Clause
Author Henning Thielemann <haskell@henning-thielemann.de>
Maintainer Henning Thielemann <haskell@henning-thielemann.de>
Category Testing
Home page https://hub.darcs.net/thielema/doctest-extract/
Source repo this: darcs get https://hub.darcs.net/thielema/doctest-extract/ --tag 0.1.2
head: darcs get https://hub.darcs.net/thielema/doctest-extract/
Uploaded by HenningThielemann at 2023-12-22T19:58:52Z
Distributions LTSHaskell:0.1.2, NixOS:0.1.2, Stackage:0.1.2
Executables doctest-extract-0.1
Downloads 541 total (24 in the last 30 days)
Rating (no votes yet) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Docs not available [build log]
Last success reported on 2023-12-22 [all 1 reports]

Readme for doctest-extract-0.1.2

[back to package description]

Known Issues

  • For compatibility with original doctest parser you cannot write

    {-# LANGUAGE MyPreferredExtension #-}
    

    Instead you must write

    :set -XMyPreferredExtension
    
  • In Literal Haskell files only \\begin{code} ... \\end{code} blocks are scanned, but not bird style code blocks.

  • prop> supports multi-line code, but both original doctest and haddock do not support it.

  • IO tests are not supported as doctest examples, so far. We need a syntactic distinction for IO tests, because doctest-extract does not employ a type-checker. We could mark IO tests with a specific id function, as in ioTest $ runMyTest or a type annotation, as in runMyTest :: IO ().

Tipps and Tricks

How to disable selected tests?

For focussing on certain tests it can be useful to disable other ones. We have not implemented a mechanism to disable parts of the test suite in doctest-extract, because this would require to implement a way to identify tests. You can still disable some of the tests without explicit support by doctest-extraxt.

  • If you want to disable whole modules, you may make a copy of the auto-generated Test/Main.hs and remove the modules that you want to skip.

  • For disabling all tests on a function you may turn a Haddock comment into a plain comment by removing the bar after the opening of the comment.

  • For disabling individual tests you may prefix >>> and prop> with an asterisk or the like.

These tricks work best in conjunction with a revision control systen, such that it always reminds you that there are tests disabled temporarily.