Awesome
HLint
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:
- HLint operates on each module at a time in isolation, as a result HLint does not know about types or which names are in scope. This decision is deliberate, allowing HLint to parallelise and be used incrementally on code that may not type-check. If fixities are required to parse the code properly, they can be supplied.
- The presence of
seq
may cause some hints (i.e. eta-reduction) to change the semantics of a program. - Some transformed programs may require additional type signatures, particularly if the transformations trigger the monomorphism restriction or involve rank-2 types. In rare cases, there might be nowhere to write the required type signature.
- Sometimes HLint will change the code in a way that causes values to default to different types, which may change the behaviour.
- HLint assumes duplicate identical expressions within in a single expression are used at the same type.
- The
RebindableSyntax
extension can cause HLint to suggest incorrect changes. - HLint can be configured with knowledge of C Pre Processor flags, but it can only see one conditional set of code at a time.
- HLint turns on many language extensions so it can parse more documents, occasionally some break otherwise legal syntax - e.g.
{-#INLINE foo#-}
doesn't work withMagicHash
,foo $bar
means something different withTemplateHaskell
. These extensions can be disabled with-XNoMagicHash
or-XNoTemplateHaskell
etc. - HLint doesn't run any custom preprocessors, e.g. markdown-unlit or record-dot-preprocessor, so code making use of them will usually fail to parse.
- Some hints, like
Use const
, don't work for non-lifted (i.e. unlifted and unboxed) types. - Some language extensions like
Strict
can cause certain hints (e.g. eta reduction) to be incorrect.
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:
- Initially, run
hlint . --report
to generatereport.html
containing a list of all issues HLint has found. Fix those you think are worth fixing and keep repeating. - 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. - 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:
- Lots of editors have HLint plugins (quite a few have more than one HLint plugin).
- HLint is part of the multiple Haskell IDEs, haskell-language-server, ghc-mod and Intero.
- HLint Source Plugin makes HLint available as a GHC plugin.
- Splint is another source plugin that doesn't require reparsing the GHC source if you are on the latest GHC version.
- Code Climate is a CI for analysis which integrates HLint.
- Danger can be used to automatically comment on pull requests with HLint suggestions.
- Restyled includes an HLint Restyler to automatically run
hlint --refactor
on files changed in GitHub Pull Requests. - hlint-test helps you write a small test runner with HLint.
- hint-man automatically submits reviews to opened pull requests in your repositories with inline hints.
- CircleCI has a plugin to run HLint more easily.
- haskell/actions has
hlint-setup
andhlint-run
actions for GitHub. - haskell-actions/hlint-scan is a GitHub action using HLint for code scanning.
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:
- HLint will only check one branch of an
#if
, based on which macros have been defined. - Any missing
#include
files will produce a warning on the console, but no information in the reports.
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:
- Increases laziness - for example
foldl (&&) True
suggestsand
including this note. The new code will work on infinite lists, while the old code would not. Increasing laziness is usually a good idea. - Decreases laziness - for example
(fst a, snd a)
suggestsa
including this note. On evaluation the new code will raise an error if a is an error, while the old code would produce a pair containing two error values. Only a small number of hints decrease laziness, and anyone relying on the laziness of the original code would be advised to include a comment. - Removes error - for example
foldr1 (&&)
suggestsand
including the noteRemoves error on []
. The new code will produceTrue
on the empty list, while the old code would raise an error. Unless you are relying on the exception thrown by the empty list, this hint is safe - and if you do rely on the exception, you would be advised to add a comment.
What is the difference between error/warning/suggestion?
Every hint has a severity level:
- Error - by default only used for parse errors.
- Warning - for example
concat (map f x)
suggestsconcatMap f x
as a "warning" severity hint. From a style point of view, you should always replace a combination ofconcat
andmap
withconcatMap
. - Suggestion - for example
x !! 0
suggestshead x
as a "suggestion" severity hint. Typicallyhead
is a simpler way of expressing the first element of a list, especially if you are treating the list inductively. However, in the expressionf (x !! 4) (x !! 0) (x !! 7)
, replacing the middle argument withhead
makes it harder to follow the pattern, and is probably a bad idea. Suggestion hints are often worthwhile, but should not be applied blindly.
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:
future
, which suggests switchingreturn
forpure
.extra
, which suggests replacements which introduce a dependency on theextra
library.use-lens
, which suggests replacements which introduce a dependency on thelens
library.use-th-quotes
, which suggests using[| x |]
where possible.generalise
, which suggests more generic methods, e.g.fmap
instead ofmap
.generalise-for-conciseness
, which suggests more generic methods, but only when they are shorter, e.g.maybe True
becomesall
.dollar
which suggestsa $ b $ c
is replaced witha . b $ c
.teaching
which encourages a simple beginner friendly style, learning about related functions.
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:
- HLint aims to both improve code, and to teach the author better style. Doing modifications individually helps this process.
- Sometimes the steps are reasonably complex, by automatically composing them the user may become confused.
- Sometimes HLint gets transformations wrong. If suggestions are applied recursively, one error will cascade.
- Some people only make use of some of the suggestions. In the above example using concatMap is a good idea, but sometimes eta reduction isn't. By suggesting them separately, people can pick and choose.
- Sometimes a transformed expression will be large, and a further hint will apply to some small part of the result, which appears confusing.
- Consider
f $ (a b)
. There are two valid hints, either remove the $ or remove the brackets, but only one can be applied.
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:
- Add command line arguments to all runs, e.g.
--color
or-XNoMagicHash
. - Ignore certain hints, perhaps within certain modules/functions.
- Restrict the use of GHC flags/extensions/functions, e.g. banning
Arrows
andunsafePerformIO
. - Add additional project-specific hints.
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:
{-# ANN module "HLint: ignore" #-}
or{-# HLINT ignore #-}
or{- HLINT ignore -}
- ignore all hints in this module (usemodule
literally, not the name of the module).{-# ANN module "HLint: ignore Eta reduce" #-}
or{-# HLINT ignore "Eta reduce" #-}
or{- HLINT ignore "Eta reduce" -}
- ignore all eta reduction suggestions in this module.{-# ANN myDef "HLint: ignore" #-}
or{-# HLINT ignore myDef #-}
or{- HLINT ignore myDef -}
- don't give any hints in the definitionmyDef
. This may be combined with hint names,{- HLINT ignore myDef "Eta reduce" -}
, to only ignore that hint in that definition.{-# ANN myDef "HLint: error" #-}
or{-# HLINT error myDef #-}
or{- HLINT error myDef -}
- any hint in the definitionmyDef
is an error.{-# ANN module "HLint: error Use concatMap" #-}
or{-# HLINT error "Use concatMap" #-}
or{- HLINT error "Use concatMap" -}
- the hint to useconcatMap
is an error (you may also usewarn
orsuggest
in place oferror
for other severity levels).
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:
- ignore: {name: Eta reduce}
- suppress all eta reduction suggestions.- ignore: {name: Eta reduce, within: [MyModule1, MyModule2]}
- suppress eta reduction hints in theMyModule1
andMyModule2
modules.- ignore: {within: MyModule.myDef}
- don't give any hints in the definitionMyModule.myDef
.- error: {within: MyModule.myDef}
- any hint in the definitionMyModule.myDef
is an error.- error: {name: Use concatMap}
- the hint to useconcatMap
is an error (you may also usewarn
orsuggest
in place oferror
for other severity levels).
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:
- Requires that all imports of
Set
must bequalified Data.Set as Set
, enforcing consistency - Ensures the module
Control.Arrow
can't be used anywhere - Prevents explicit imports of the given identifiers from
Control.Monad.State
(e.g. to prevent people from importing reexported identifiers). - Prevents all imports from
Control.Exception
, exceptException
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:
-
asRequired
: boolean.If true,
as
alias is required. Ignored ifas
is empty. -
importStyle
: one of'qualified'
,'unqualified'
,'explicit'
,'explicitOrQualified'
,'unrestricted'
.The preferred import style.
explicitOrQualified
accepts bothimport Foo (a,b,c)
andimport qualified Foo
, but notimport Foo
orimport Foo hiding (x)
.explicit
is basically the same, but doesn't acceptimport qualified
.qualified
andunqualified
do not care about the import list at all. -
qualifiedStyle
: either'pre'
,'post'
or'unrestricted'
; how should the module be qualified? This option also affects how suggestions are formatted.
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:
- Requires
Data.Set
andData.HashSet
to be imported with aliasSet
; if imported without alias, a warning is generated. - Says that
Debug
must be imported either qualified with post-qualification, i.e.import Debug qualified
, or with an explicit import list, e.g.Debug (debugPrint)
. - Requires that
Unsafe
must always be imported qualified, and can't be aliased. - Forbids
import qualified Prelude
andimport Prelude qualified
(with or without explicit import list).
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 SDoc
s. 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.