Home

Awesome

HLint Hackage version Stackage version Build status

HLint is a tool for suggesting possible improvements to Haskell code. These suggestions include ideas such as using alternative functions, simplifying code and spotting redundancies. This document is structured as follows:

Bugs and limitations

Bugs can be reported on the bug tracker. There are some issues that I do not intend to fix:

Installing and running HLint

Installation follows the standard pattern of any Haskell library or program: type cabal update to update your local hackage database, then cabal install hlint to install HLint.

Once HLint is installed, run hlint source where source is either a Haskell file, or a directory containing Haskell files. A directory will be searched recursively for any files ending with .hs or .lhs. For example, running HLint over darcs would give:

$ hlint darcs-2.1.2

darcs-2.1.2\src\CommandLine.lhs:94:1: Warning: Use concatMap
Found:
    concat $ map escapeC s
Perhaps:
    concatMap escapeC s

darcs-2.1.2\src\CommandLine.lhs:103:1: Suggestion: Use fewer brackets
Found:
    ftable ++ (map (\ (c, x) -> (toUpper c, urlEncode x)) ftable)
Perhaps:
    ftable ++ map (\ (c, x) -> (toUpper c, urlEncode x)) ftable

darcs-2.1.2\src\Darcs\Patch\Test.lhs:306:1: Warning: Use a more efficient monadic variant
Found:
    mapM (delete_line (fn2fp f) line) old
Perhaps:
    mapM_ (delete_line (fn2fp f) line) old

... lots more hints ...

Each hint says which file/line the hint relates to, how serious an issue it is, a description of the hint, what it found, and what you might want to replace it with. In the case of the first hint, it has suggested that instead of applying concat and map separately, it would be better to use the combination function concatMap.

The first hint is marked as an warning, because using concatMap in preference to the two separate functions is always desirable. In contrast, the removal of brackets is probably a good idea, but not always. Reasons that a hint might be a suggestion include requiring an additional import, something not everyone agrees on, and functions only available in more recent versions of the base library.

Any configuration can be done via .hlint.yaml file. Any other file name, such as .hlint.yml, can only be used explicitly with the --hint=FILE option.

Bug reports: The suggested replacement should be equivalent - please report all incorrect suggestions not mentioned as known limitations.

Suggested usage

HLint usage tends to proceed in three distinct phases:

  1. Initially, run hlint . --report to generate report.html containing a list of all issues HLint has found. Fix those you think are worth fixing and keep repeating.
  2. Once you are happy, run hlint . --default > .hlint.yaml, which will generate a settings file ignoring all the hints currently outstanding. Over time you may wish to edit the list.
  3. For larger projects, add custom hints or rules.

Most hints are intended to be a good idea in most circumstances, but not universally - judgement is required. When contributing to someone else's project, HLint can identify pieces of code to look at, but only make changes you consider improvements - not merely to adhere to HLint rules.

Running with Continuous Integration

On CI you might wish to run hlint . (or hlint src if you only want to check the src directory). To avoid the cost of compilation you may wish to fetch the latest HLint binary release.

For the CI systems Travis, Appveyor and Azure Pipelines add the line:

curl -sSL https://raw.github.com/ndmitchell/hlint/master/misc/run.sh | sh -s .

The arguments after -s are passed to hlint, so modify the final . if you want other arguments. This command works on Windows, Mac and Linux.

Integrations

HLint is integrated into lots of places:

Automatically Applying Hints

HLint can automatically apply some suggestions using the --refactor flag. If passed, instead of printing out the hints, HLint will output the refactored file on stdout. For --refactor to work it is necessary to have the refactor executable from the apply-refact package on your $PATH. HLint uses that tool to perform the refactoring.

When using --refactor you can pass additional options to the refactor binary using --refactor-options flag. Some useful flags include -i (which replaces the original file) and -s (which asks for confirmation before performing a hint). The --with-refactor flag can be used to specify an alternative location for the refactor binary. Simple bindings for Vim, Emacs and Atom are available.

While the --refactor flag is useful, not all hints support refactoring. See hints.md for which hints don't support refactoring.

Reports

HLint can generate a lot of information, making it difficult to search for particular types of errors. The --report flag will cause HLint to generate a report file in HTML, which can be viewed interactively. Reports are recommended when there are more than a handful of hints.

Language Extensions

HLint enables most Haskell extensions, disabling only those which steal too much syntax (e.g. Arrows, TransformListComp and TypeApplications). Individual extensions can be enabled or disabled with, for instance, -XArrows, or -XNoMagicHash. The flag -XHaskell2010 selects Haskell 2010 compatibility. You can also pass them via .hlint.yaml file. For example: - arguments: [-XArrows].

Emacs Integration

Emacs integration has been provided by Alex Ott. The integration is similar to compilation-mode, allowing navigation between errors. The script is at hs-lint.el, and a copy is installed locally in the data directory. To use, add the following code to the Emacs init file:

(require 'hs-lint)
(defun my-haskell-mode-hook ()
    (local-set-key "\C-cl" 'hs-lint))
(add-hook 'haskell-mode-hook 'my-haskell-mode-hook)

GHCi Integration

GHCi integration has been provided by Gwern Branwen. The integration allows running :hlint from the GHCi prompt. The script is at hlint.ghci, and a copy is installed locally in the data directory. To use, add the contents to your GHCi startup file.

Parallel Operation

To run HLint on 4 processors append the flags -j4. HLint will usually perform fastest if n is equal to the number of physical processors, which can be done with -j alone.

If your version of GHC does not support the GHC threaded runtime then install with the command: cabal install --flags="-threaded"

C preprocessor support

HLint runs the cpphs C preprocessor over all input files, by default using the current directory as the include path with no defined macros. These settings can be modified using the flags --cpp-include and --cpp-define. To disable the C preprocessor use the flag -XNoCPP. There are a number of limitations to the C preprocessor support:

FAQ

Usage

Why doesn't the compiler automatically apply the optimisations?

HLint doesn't suggest optimisations, it suggests code improvements - the intention is to make the code simpler, rather than making the code perform faster. The GHC compiler automatically applies many of the rules suggested by HLint, so HLint suggestions will rarely improve performance.

Why do I sometimes get a "Note" with my hint?

Most hints are perfect substitutions, and these are displayed without any notes. However, some hints change the semantics of your program - typically in irrelevant ways - but HLint shows a warning note. HLint does not warn when assuming typeclass laws (such as == being symmetric). Some notes you may see include:

What is the difference between error/warning/suggestion?

Every hint has a severity level:

The difference between warning and suggestion is one of personal taste, typically my personal taste. If you already have a well developed sense of Haskell style, you should ignore the difference. If you are a beginner Haskell programmer you may wish to focus on warning hints before suggestion hints.

Why do I get a parse error?

HLint enables/disables a set of extensions designed to allow as many files to parse as possible, but sometimes you'll need to enable an additional extension (e.g. Arrows, QuasiQuotes, ...), or disable some (e.g. MagicHash) to enable your code to parse.

You can enable extensions by specifying additional command line arguments in .hlint.yaml, e.g.: - arguments: [-XQuasiQuotes].

How do I only run hlint on changed files?

If you're using git, it may be helpful to only run hlint on changed files. This can be a considerable speedup on very large codebases.

{ git diff --diff-filter=d --name-only $(git merge-base HEAD origin/master) -- "***.hs" && git ls-files -o --exclude-standard -- "***.hs"; } | xargs hlint

Because hlint's --refactor option only works when you pass a single file, this approach is also helpful to enable refactoring many files in a single command:

{ git diff --diff-filter=d --name-only $(git merge-base HEAD origin/master) -- "***.hs" && git ls-files -o --exclude-standard -- "***.hs"; } | xargs -I file hlint file --refactor --refactor-options="--inplace --step"

Configuration

Why doesn't HLint know the fixity for my custom !@%$ operator?

HLint knows the fixities for all the operators in the base library, as well as operators whose fixities are declared in the module being linted, but no others. HLint works on a single file at a time, and does not resolve imports, so cannot see fixity declarations from imported modules. You can tell HLint about fixities by putting them in a hint file named .hlint.yaml with the syntax - fixity: "infixr 5 !@%$". You can also use --find to automatically produce a list of fixity declarations in a file.

Which hints are ignored?

Some hints are off-by-default. Some are ignored by the configuration settings. To see all hints pass --show. This feature is often useful in conjunction with --report which shows the hints in an interactive web page, allowing them to be browsed broken down by hint.

Which hints are used?

HLint uses the hlint.yaml file it ships with by default (containing things like the concatMap hint above), along with the first .hlint.yaml file it finds in the current directory or any parent thereof. To include other hints, pass --hint=filename.yaml.

Are there any extra hints available?

There are a few groups of hints that are shipped with HLint, but disabled by default. These are:

These can be enabled by passing --with-group=future or adding the following to your .hlint.yaml file:

- group: {name: future, enabled: true}

Design

Why are hints not applied recursively?

Consider:

foo xs = concat (map op xs)

This will suggest eta reduction to concat . map op, and then after making that change and running HLint again, will suggest use of concatMap. Many people wonder why HLint doesn't directly suggest concatMap op. There are a number of reasons:

Is it possible to use pragma annotations in code that is read by ghci (conflicts with OverloadedStrings)?

Short answer: yes, it is!

If the language extension OverloadedStrings is enabled, ghci may however report error messages such as:

Ambiguous type variable ‘t0’ arising from an annotation
prevents the constraint ‘(Data.Data.Data t0)’ from being solved.

In this case, a solution is to add the :: String type annotation. For example:

{-# ANN someFunc ("HLint: ignore Use fmap" :: String) #-}

See discussion in issue #372.

Customizing the hints

To customize the hints given by HLint, create a file .hlint.yaml in the root of your project. For a suitable default run:

hlint --default > .hlint.yaml

This default configuration shows lots of examples (as # comments) of how to:

You can see the output of --default for a clean lint here but for a dirty project --default output includes an extra warnings section that counts and ignores any hints it finds:

# Warnings currently triggered by your code
- ignore: {name: "Redundant $"} # 20 hints
- ignore: {name: "Unused LANGUAGE pragma"} # 29 hints

If you wish to use the Dhall configuration language to customize HLint, there is an example and type definition.

Finding the name of a hint

Hints are named with the string they display in their help message

For example, if hlints outputs a warning like

./backend/tests/api-tests/src/Main.hs:116:51: Warning: Redundant ==
Found:
  regIsEnabled rr == True
Perhaps:
  regIsEnabled rr

the name of the lint is Redundant ==.

You can use that name to refer to the lint in the configuration file and ANN pragmas, see the following sections.

Ignoring hints

Some of the hints are subjective, and some users believe they should be ignored. Some hints are applicable usually, but occasionally don't always make sense. The ignoring mechanism provides features for suppressing certain hints. Ignore directives can either be written as pragmas in the file being analysed, or in the hint files. Examples of pragmas are:

For ANN pragmas it is important to put them after any import statements. If you have the OverloadedStrings extension enabled you will need to give an explicit type to the annotation, e.g. {-# ANN myDef ("HLint: ignore" :: String) #-}. The ANN pragmas can also increase compile times or cause more recompilation than otherwise required, since they are evaluated by TemplateHaskell.

For {-# HLINT #-} pragmas GHC may give a warning about an unrecognised pragma, which can be suppressed with -Wno-unrecognised-pragmas.

For {- HLINT -} comments they are likely to be treated as comments in syntax highlighting, which can lead to them being overlooked.

Ignore directives can also be written in the hint files:

These directives are applied in the order they are given, with later hints overriding earlier ones.

You can choose to ignore all hints with - ignore: {} then selectively enable the ones you want (e.g. - warn: {name: Use const}), but it isn't a totally smooth experience (see #747 and #748).

Finally, hlint defines the __HLINT__ preprocessor definition (with value 1), so problematic definitions (including those that don't parse) can be hidden with:

#ifndef __HLINT__
foo = ( -- HLint would fail to parse this
#endif

Adding hints

The hint suggesting concatMap can be defined as:

- warn: {lhs: concat (map f x), rhs: concatMap f x}

This line can be read as replace concat (map f x) with concatMap f x. All single-letter variables are treated as substitution parameters. For examples of more complex hints see the supplied hlint.yaml file in the data directory. This hint will automatically match concat . map f and concat $ map f x, so there is no need to give eta-reduced variants of the hints. Hints may tagged with error, warn or suggest to denote how severe they are by default. In addition, hint is a synonym for suggest. If you come up with interesting hints, please submit them for inclusion.

You can search for possible hints to add from a source file with the --find flag, for example:

$ hlint --find=src/Utils.hs
-- hints found in src/Util.hs
- warn: {lhs: "null (intersect a b)", rhs: "disjoint a b"}
- warn: {lhs: "dropWhile isSpace", rhs: "trimStart"}
- fixity: "infixr 5 !:"

These hints are suitable for inclusion in a custom hint file. You can also include Haskell fixity declarations in a hint file, and these will also be extracted. If you pass only --find flags then the hints will be written out, if you also pass files/folders to check, then the found hints will be automatically used when checking.

Hints can specify more advanced aspects, with names and side conditions. To see examples and descriptions of these features look at the default hint file and the hint interpretation module comments.

Restricting items

HLint can restrict what Haskell code is allowed, which is particularly useful for larger projects which wish to enforce coding standards - there is a short example in the HLint repo itself. As an example of restricting extensions:

- extensions:
  - default: false
  - name: [DeriveDataTypeable, GeneralizedNewtypeDeriving]
  - {name: CPP, within: CompatLayer}

The above block declares that GHC extensions are not allowed by default, apart from DeriveDataTypeable and GeneralizedNewtypeDeriving which are available everywhere. The CPP extension is only allowed in the module CompatLayer. Much like extensions, you can use flags to limit the GHC_OPTIONS flags that are allowed to occur. You can also ban certain functions:

- functions:
  - {name: nub, within: []}
  - {name: unsafePerformIO, within: CompatLayer}

This declares that the nub function can't be used in any modules, and thus is banned from the code. That's probably a good idea, as most people should use an alternative that isn't O(n^2) (e.g. nubOrd). We also whitelist where unsafePerformIO can occur, ensuring that there can be a centrally reviewed location to declare all such instances. Function names can be given qualified, e.g. Data.List.head, but note that functions available through multiple exports (e.g. head is also available from Prelude) should be listed through all paths they are likely to be obtained, as the HLint qualified matching is unaware of re-exports.

Finally, we can restrict the use of modules with:

- modules:
  - {name: [Data.Set, Data.HashSet], as: Set}
  - {name: Control.Arrow, within: []}
  - {name: Control.Monad.State, badidents: [modify, get, put], message: "Use Control.Monad.State.Class instead"}
  - {name: Control.Exception, only: [Exception], message: "Use UnliftIO.Exception instead"}

This fragment adds the following hints:

You can customize the Note: for restricted modules, functions and extensions, by providing a message field (default: may break the code).

Other options are available:

For example:

- modules:
  - {name: [Data.Set, Data.HashSet], as: Set, asRequired: true}
  - {name: Debug, importStyle: explicitOrQualified}
  - {name: Unsafe, importStyle: qualified, qualifiedStyle: post, as: Unsafe}
  - {name: Prelude, importStyle: unqualified}

This:

You can match on module names using glob-style wildcards. Module names are treated like file paths, except that periods in module names are like directory separators in file paths. So **.*Spec will match Spec, PreludeSpec, Data.ListSpec, and many more. But *Spec won't match Data.ListSpec because of the separator. See the filepattern library for a more thorough description of the matching.

Restrictions are unified between wildcard and specific matches. With asRequired, importStyle and qualifiedStyle fields, the more specific option takes precedence. The list fields are merged. With multiple wildcard matches, the precedence between them is not guaranteed (but in practice, names are sorted in the reverse lexicograpic order, and the first one wins -- which hopefully means the more specific one more often than not)

If the same module is specified multiple times, for asRequired, importStyle and qualifiedStyle fields, only the first definition will take effect.

- modules:
  - {name: [Data.Map, Data.Map.*], as: Map}
  - {name: Test.Hspec, within: **.*Spec }
  - {name: '**', importStyle: post}

Hacking HLint

Contributions to HLint are most welcome, following my standard contribution guidelines.

How to run tests

You can run the tests either from within a ghci session by typing :test or by running the standalone binary's tests via cabal run -- hlint --test or stack run -- hlint --test. After changing hints, you will need to regenerate the hints.md file with hlint --generate-summary.

How to add tests

New tests for individual hints can be added directly to source and hint files by adding annotations bracketed in <TEST></TEST> code comment blocks. Here are some examples:

{-
    Tests to check the zipFrom hint works

<TEST>
zip [1..length x] x -- zipFrom 1 x
zip [1..length y] x
zip [1..length x] x -- ??? @Warning
</TEST>
-}

The general syntax is lhs -- rhs with lhs being the expression you expect to be rewritten as rhs. The absence of rhs means you expect no hints to fire. In addition ??? lets you assert a warning without a particular suggestion, while @ tags require a specific severity -- both these features are used less commonly.

Printing abstract syntax

Getting started on problems in HLint often means wanting to inspect a GHC parse tree to get a sense of what it looks like (to see how to match on it for example). Given a source program Foo.hs (say), you can get GHC to print a textual representation of Foo's AST via the -ddump-parsed-ast flag e.g. ghc -fforce-recomp -ddump-parsed-ast -c Foo.hs.

When you have an HsSyn term in your program, it's quite common to want to print it (e.g. via Debug.Trace.trace). Types in HsSyn aren't in Show. Not all types in HsSyn are Outputable but when they are you can call ppr to get SDocs. This idiom is common enough that there exists unsafePrettyPrint. The function showAstData can be called on any HsSyn term to get output like with the dump-parsed-ast flag. The showAstData approach is preferable to ppr when both choices exist in that two ASTs that differ only in fixity arrangements will render differently with the former.

Generating the hints summary

The hints summary is an auto-generated list of hlint's builtin hints. This can be generated with hlint --generate-summary, which will output the summary to hints.md.

Acknowledgements

Many improvements to this program have been made by Niklas Broberg in response to feature requests. Additionally, many people have provided help and patches, including Lennart Augustsson, Malcolm Wallace, Henk-Jan van Tuyl, Gwern Branwen, Alex Ott, Andy Stewart, Roman Leshchinskiy, Johannes Lippmann, Iustin Pop, Steve Purcell, Mitchell Rosen and others.