danielg / ghc-mod

Happy Haskell Hacking for editors. DEPRECATED

ghc-mod: Happy Haskell Hacking

Legacy

Please note that using ghc-mod as a user facing tool for IDE/Editor integration is no longer supported or maintained. We are in the process of refocusing our efforts around haskell-ide-engine as the main user/IDE facing program.

Currently haskell-ide-engine still uses ghc-mod as a library for GHC session setup, so this part will still be maintained. However moving forward we're hoping to move most of the functionality which remains in ghc-mod (the library) into GHC upstream rendering ghc-mod unceccesary. Meanwhile ghc-mod (the library) will remain.

If someone feels like taking over maintainership of ghc-mod as a standalone development tool feel free to contact the maintainer. However I must warn you: a mountain of legacy baggage and scattered Editor/IDE plugins awaits. Your time is probably better spent improving haskell-ide-engine.

For more information on what ghc-mod used to be have a look at the old README.

Reporting Bugs

Please report bugs on the GitHub issue tracker for ghc-mod: https://github.com/DanielG/ghc-mod/issues

Including general environment information like the operating system (distribution, version) you're using and the output of $ ghc-mod debug run in your project directory is probably a good idea.

IRC

If you have any problems, suggestions, comments swing by #ghc-mod (web client) on Freenode. If you're reporting a bug please also create an issue here (GitHub issue tracker) so we have a way to contact you if you don't have time to stay.

Do hang around for a while if no one answers, and repeat your question if you still haven't gotten any answer after a day or so (the maintainer was probably asleep). You're most likely to get an answer during the day in GMT+1.

ghc-mod's Issues

ghc-mod check fails with "GHCi runtime linker: fatal error" on OS X

Hi,

ghc-mod check fails with following error, although it can be built with cabal.
I afraid ghc-mod is not the actual cause but I couldn't find why it happens.

$ ghc-mod check Main.hs
GHCi runtime linker: fatal error: I found a duplicate definition for symbol
   _fps_count
whilst processing object file
   /Users/fujimura/.cabal/lib/bytestring-0.10.0.1/ghc-7.4.2/HSbytestring-0.10.0.1.o
This could be caused by:
   * Loading two different object files which export the same symbol
   * Specifying the same object file twice on the GHCi command line
   * An incorrect `package.conf' entry, causing some object to be
     loaded twice.
GHCi cannot safely continue in this situation.  Exiting now.  Sorry.

I'm using OS X 10.8.2, Xcode 4.5.2 and latest haskell-platform.

reproducible(at least for my machine) code is here:
https://gist.github.com/4140026

ghc-show-type doesn't work on a file with errors

Since Info.hs tries to load the entire file to get the type, ghc-show-type fails when the file contains errors. I think this is a rather big limitation, since often you want convenient type information when trying to debug some piece of code that doesn't typecheck.

Unfortunately, partially loading the file, ignoring the parts with errors, has not been implemented yet and seems to be somewhat difficult, and such a subset would probably not be unique. A different approach would be to load just the imports, since most types of locally defined functions/constants should be visible, since they are in the same buffer (assuming the programmer wrote type signatures).
In addition to that, it might be possible to have ghc-mod output type information for locally defined function when compiling succeeds and caching those on the emacs side.

type insertion for qualified imports

If I import Bytestring qualified as B, then M-t to get the type of a function that takes a bytestring, ghc-mod will insert Bytestring rather than B.Bytestring, causing a compile failure.

Some errors in the website

Hi.

I didn't know where to post this, so I'm putting it here. I've found two mistakes on the website of ghc-mod. On the last line of http://www.mew.org/~kazu/proj/ghc-mod/en/install.html it says M-C-t, where it should be M-C-i. I've spent an amount of time until I realize this was wrong. Also, in http://www.mew.org/~kazu/proj/ghc-mod/en/, in the description of C-c> there's a ! that should not be there.

By the way, thank you for this amazing emacs mode. It makes Emacs fell like an IDE, for me it is so good that it stimulates me to use Haskell even more, and to forget about considering switching to Eclipse of Leksah. Congratulations!

Fail to build using GHC 7.2.1

Resolving dependencies...
Configuring ghc-mod-0.6.1...
Preprocessing executables for ghc-mod-0.6.1...
Building ghc-mod-0.6.1...
[1 of 9] Compiling Types ( Types.hs, dist/build/ghc-mod/ghc-mod-tmp/Types.o )
[2 of 9] Compiling Cabal ( Cabal.hs, dist/build/ghc-mod/ghc-mod-tmp/Cabal.o )

Cabal.hs:37:23: Not in scope: `H.liftIO'

Cabal.hs:66:5: Not in scope: `H.liftIO'

Cabal.hs:71:13: Not in scope: `H.liftIO'

Cabal.hs:79:13: Not in scope: `H.liftIO'
cabal: Error: some packages failed to install:
ghc-mod-0.6.1 failed during the building phase. The exception was:
ExitFailure 1

tangboyun@tangboyun-MXC062:/tmp/ghc-mod$ ghc --version
The Glorious Glasgow Haskell Compilation System, version 7.2.1

How hard is it to integrate with the auto-complete engine?

Auto-complete is also hosted on github: http://github.com/m2ym/auto-complete. It conveniently shows a popup when completion is available.

Currently I'm using http://madscientist.jp/~ikegami/diary/20090215.html, which while usable, is a unextensible hack.

Use -fno-code by default

I managed to make syntastic much faster with this:

let g:syntastic_haskell_checker_args = '--ghcOpt="-fno-code"'

This makes GHC type check the code without actually compiling it.

I propose to make this the default in ghc-mod itself, or even hardcode it and not make it configurable, since ghc-mod isn't really tasked with compilation at all.

Could not find module 'Control.Parallel'

Hi,
I've encountered an issue about modules. When I use ghc-mod check par.hs to check a source file it gives me 'Cannot find module Control.Parallel' error, while the file can be compiled by ghc par.hs without any problems.

if a cabal package is not installed and I try to import it, I don't get an error.

➜  ghc-mod git:(master) ✗ cat Test.hs
module Test where
import Data.Attoparsec.Char8
foo = bar
➜  ghc-mod git:(master) ✗ ghc-mod check Test.hs
# no output
➜  ghc-mod git:(master) ✗ cabal install --user --disable-shared --disable-library-profiling --disable-executable-profiling attoparsec
<snip details>
➜  ghc-mod git:(master) ✗ ghc-mod check Test.hs
# ah, and now we have it
Test.hs:3:6:Not in scope: `bar'

My expected behaviour would be to complain that it can't find those modules.
(do you have a preferred setup for submitting tests? HUnit, test-framework, shell commands... ?)

ghc-mod 1.11.1 broken with ghc 6.12

Hi,

I'm using ghc 6.12.1 with Cabal library 1.8.0.2. The following happens:

$ cabal install ghc-mod
Resolving dependencies...
Configuring ghc-mod-1.11.1...
Preprocessing executables for ghc-mod-1.11.1...
Building ghc-mod-1.11.1...
[ 1 of 16] Compiling GHCChoice        ( GHCChoice.hs, dist/build/ghc-mod/ghc-mod-tmp/GHCChoice.o )
[ 2 of 16] Compiling Gap              ( Gap.hs, dist/build/ghc-mod/ghc-mod-tmp/Gap.o )
[ 3 of 16] Compiling ErrMsg           ( ErrMsg.hs, dist/build/ghc-mod/ghc-mod-tmp/ErrMsg.o )
[ 4 of 16] Compiling Types            ( Types.hs, dist/build/ghc-mod/ghc-mod-tmp/Types.o )
[ 5 of 16] Compiling GHCApi           ( GHCApi.hs, dist/build/ghc-mod/ghc-mod-tmp/GHCApi.o )
[ 6 of 16] Compiling Cabal            ( Cabal.hs, dist/build/ghc-mod/ghc-mod-tmp/Cabal.o )

Cabal.hs:37:56: Not in scope: `usedExtensions'

Cabal.hs:38:53: Not in scope: `defaultLanguage'
cabal: Error: some packages failed to install:
ghc-mod-1.11.1 failed during the building phase. The exception was:
ExitFailure 1

I believe that ghc-mod needs a more recent version of Cabal, but in ghc-mod's dependencies Cabal is listed as -any.

I can't seem to import GHC modules

I have the line

import GHC(parseStaticFlags, noLoc)

and flymake is choking on it, because it's in the hidden package ghc-6.12.1. Is there a place I can customise the packages flymake sees? It's in the cabal file for the package..

ghc-mode saves the file every time I hit enter when flymake mode is on.

I know flymake needs something to work off, but saving the file breaks the intuition of the user that it's safe to experiment then not save when it doesn't work out. could you perhaps maintain a copy of the file somewhere and save & check that?

guilty code in ghc-flymake.el

(defun ghc-flymake-init ()
  (let ((after-save-hook nil))
    (save-buffer))
  (let ((file (file-name-nondirectory (buffer-file-name))))
    (list ghc-module-command (append (ghc-module-command-args)
                     (list "check" file)))))

Odd warnings against using Template Haskell

When I use "ghc-mod check" on my code using Template Haskell, I got "This binding for `...' shadows the existing binding" warnings. If I run ghc normally, I get no warnings.

The following URL is a code to reproduce;
https://gist.github.com/3082903

I tried to fix this problem and read source codes, but I couldn't finish it :/

Key binding issue

ghc.el try to bind the following keys

• M-t
• C-M-d
• C-x C-s
• M-p
• M-n

but these keys are reserved by GNU Emacs or are not expected to be overrided by some users.

The conservative or standard way to make mode-specific key bindings is to add the prefix C-c to each key bindings.

Changelog

I really like ghc-mod.

It would be cool if you could keep some small changelog that informs of the latest and greatest changes in new versions ;)

plugin/ghcmod.vim accesses autoload on startup

plugin/ghcmod.vim invokes ghcmod#check_version on its startup. That kills the whole point of autoload feature, i.e. speed up of startup time.

ignores cabal file

If I use a module that's defined in two places like Control.Monad.Trans, and specify the package in the cabal file, everything builds fine but ghc-mod check complains. I assume this is because it's not referring to the cabal file, so it doesn't know which packages to add. Is this a feature that's on the roadmap for implementation?

ghc-indent.el duplicates the functionality of indent-rigidly (C-x TAB)

Buffer-local completion

What do you think about making the completion buffer-local? The following advice does the trick for me, but I think this would also be a sane default:

(defadvice ghc-init
    (before dhl-ghc-mod-local-completion first () activate)
  "Makes ghc-mod completions buffer local."
  (make-local-variable 'ghc-loaded-module)
  (make-local-variable 'ghc-merged-keyword))

Allow user to not use ghc-import-module

ghc-comp.el contains this line of code at the top level:

(add-hook 'find-file-hook 'ghc-import-module)

This means that I cannot choose not to have this hook defined. This should be moved to the ghc-init function, so I can choose to define my own ghc-init function that does not add this hook.

ghc-browse-document could work with functions too

With C-c C-i I can get the information of where a function was defined, and then I can browse the documentation with M-C-d. It would be good if this process was in a keybinding.

generating etags with hasktags

haskell-emacs has a cool feature: generatng etags using hasktags program.
This allows autocompletion on the project level (not only local buffer)
How hard would it be to add it to ghc-mode ?

ghc-mod fails with FFI declarations

ghc-mod info Ffi.hs "" foo fails with this file as Ffi.hs:

{-# LANGUAGE ForeignFunctionInterface #-}
module HelloFFI where
import Foreign.C.Types

foreign export ccall foo :: CUInt
foo :: CUInt
foo = 123

with this message

Dummy:0:0:Error:Could not find module `HelloFFI'
It is not a module in the current program, or in any known package.

however, ghc-mod works without the foreign declaration and provides the correct type for foo.

Although I'll try to fix this, I think I'm not able to fix this on my own, so any help is appreciated. Thanks!

edit: okay, it fails at Gap.setCtx.
edit2: sigh ghci itself fails with ghci Ffi.hs:

[1 of 1] Compiling HelloFFI         ( Ffi.hs, interpreted )
Ffi.hs:5:1:
    Illegal foreign declaration: requires via-C, llvm (-fllvm) or native code generation (-fvia-C)
    When checking declaration: foreign export ccall "foo" foo :: CUInt
Failed, modules loaded: none.

good news: ghci -fobject-code Ffi.hs works. bad news: I don't know how to tell ghc-mod about that. maybe via DynFlags?

Failed to launch syntax check process 'ghc-mod' wth args (-f nil lint ****)

Sometimes I got error message "Flymake: Failed to launch syntax check process 'ghc-mod' with args (-f nil lint some_filename.hs): Wrong type argument: stringp, nil. Flymake will be switched OFF" when I press "C-x C-s".

If I invoke ghc-mod -f nil lint some_filename.hs or ghc-mod lint some_filename.hs in a console it returns error code 0.

my ~/.emacs:

(add-to-list 'load-path "~/.emacs.d/ghc-mod/")
(autoload 'ghc-init "ghc" nil t)
(add-hook 'haskell-mode-hook (lambda () (ghc-init) (flymake-mode)))

flymake does not play nice with delete-trailing-space

If I add

(add-hook 'before-save-hook 'delete-trailing-whitespace)

to my .emacs it becomes difficult to edit, as the buffer is continuously being saved and deleting the trailing whitespace.

Would it be possible to modify ghc-save-buffer to call a hook which could have the delete-trailing-whitespace in it.

Or is it simpler to just run a separate process on the project to clean the files periodically?

Type information is incomplete.

When using the type features of ghc-mod, it never returns any TypeClass constraints or even the right namespaces. As a result, the types are not 'usable'. For example the type of the following function

import Data.Vector.Unboxed as U
test = U.Singleton (0 :: Int)

is given as Vector Int, but this won't compile because Vector is not in scope at this point, only U.Vector.

A second issue arises with TypeClass constraints. For example, let's require a simple one like so:

fractionalOrElse a b = a / b

ghc-mod reports the type as a -> a -> a, but if we add that type annotation it fails because a is not an instance of Fractional. What we really need is Fractional a => a -> a-> a. Is there anything we can do about this?

no output with LANGUAGE TemplateHaskell

rhino ➜ ~/projects/dojo/ninjablocks-zeromq-workers/src/channels git:(refactoring_channels) ✗ ghc-mod check T.hs
rhino ➜ ~/projects/dojo/ninjablocks-zeromq-workers/src/channels git:(refactoring_channels) ✗ ghc-mod check T2.hs
T2.hs:2:3:No instance for (Num [Char]) arising from the literal 1'Possible fix: add an instance declaration for (Num [Char])In the first argument of(==)', namely 1'In the expression: 1 == "1"In an equation forl': l = 1 == "1"
rhino ➜ ~/projects/dojo/ninjablocks-zeromq-workers/src/channels git:(refactoring_channels) ✗ cat T.hs
{-# LANGUAGE TemplateHaskell #-}
module T where
l=1=="1"
rhino ➜ ~/projects/dojo/ninjablocks-zeromq-workers/src/channels git:(refactoring_channels) ✗ cat T2.hs
module T2 where
l=1=="1"

Frontend-backend split?

Would you be interested in splitting ghc-mod into a backend, which actually loads the various source files, and a frontend which queries the backend? The reason for doing this would be that the backend wouldn't need to continually restart, so ghc-mod would only have to reparse files when something in the source tree changes, making ghc-mod much more responsive.

I'm not saying this needs to be done right now; this is more of a wishlist. But I might do it, and I was wondering if that might be something you'd like.

ghc-which duplicates the functionality of executable-find

Just FYI

Parse error with multiple where clauses

The following code compiles in ghc 7.2.1.

foo :: Int -> Int
foo x = x + y where
    y = z where
    z = 2

main :: IO ()
main = do
    print $ foo 2
    return ()

ghc-mod check reports no errors, but ghc-mod lint produces the cryptic:

%ghc-mod lint minerror.hs
minerror.hs:6:1: Parse errorError message:  Parse error: ;Code:        z = 2      > main :: IO ()    main = do        print $ foo 2

Removing the second where from foo eliminates the parse error. I would expect the above code to either produce no error from lint, or to complain about the second where rather than showing a parse error in an unrelated location.

unused values should probably be a warning, not an error

if you're developing bottom up, it's very distracting to see it flash red for variables you're not yet using. A warning is fine, but it shouldn't be an outright error.

Panic with type variables in function type in where clause

I'm getting this error:

$ ghc-mod type Sample.hs Main 4 12
Dummy:0:0:Error:ghc-mod: panic! (the 'impossible' happened)
  (GHC version 7.4.2 for i386-apple-darwin):
    Evaluated the place holder for a PostTcKind

Please report this as a GHC bug:  http://www.haskell.org/ghc/reportabug

with the test code I've boiled down to as minimal as possible (saved in Sample.hs):

foo :: Int -> Int
foo x = foo' x where
  foo' :: a -> a
  foo' y = y

The error seems to be caused by the type variable in the signature for foo' in the embedded where clause, as either of the following work:

foo :: Int -> Int
foo x = foo' x where
--  foo' :: a -> a
  foo' y = y

 $ ghc-mod type Sample.hs Main 4 12
4 12 4 13 "t"
4 3 4 13 "t -> t"
2 1 4 13 "Int -> Int"

or

foo :: Int -> Int
foo x = foo' x where
  foo' :: Int -> Int
  foo' y = y

 $ ghc-mod type Sample.hs Main 4 12
4 12 4 13 "Int"
4 3 4 13 "Int -> Int"
2 1 4 13 "Int -> Int"

Note that ghci has no problem with any of the above.

I am on Mac 10.6.8 with 32-bit GHC 7.4.2 and ghc-mod version 1.11.2. Please let me know if there's any other information I can provide.

customize hlint

Of course I opened this issue up on ghcmod-vim first :) eagletmt/ghcmod-vim#5 (comment)

I can customize hlint by adding a HLint.hs file, but I want to have global settings. I put a script in my path that calls hlint with a --ignore option. This works well if I just invoke hlint, but it doesn't have an effect through syntastic.

What's the reasoning behind hiding GHC.* modules?

Hiding is done in the exposedModules function. Why?

Quasi Quote syntax support

vim syntastic user here. It seems that Quasi Quotes causes a syntax error.

Please don't overwrite M-t

M-t is a very useful emacs default keybinding, please don't overwrite it.

fixing things up for ghc 7.6

i'm trying to sort out getting things to build for ghc 7.6

will add details as I have time

Compile error on cabal install ghc-mod-1.10.8 with ghc 7.0.3

Info.hs:50:10:
Illegal instance declaration for HasType (LHsExpr Id)' (All instance types must be of the form (T t1 ... tn) where T is not a synonym. Use -XTypeSynonymInstances if you want to disable this.) In the instance declaration forHasType (LHsExpr Id)'

Info.hs:60:10:
Illegal instance declaration for HasType (LHsBind Id)' (All instance types must be of the form (T t1 ... tn) where T is not a synonym. Use -XTypeSynonymInstances if you want to disable this.) In the instance declaration forHasType (LHsBind Id)'

Info.hs:64:10:
Illegal instance declaration for HasType (LPat Id)' (All instance types must be of the form (T t1 ... tn) where T is not a synonym. Use -XTypeSynonymInstances if you want to disable this.) In the instance declaration forHasType (LPat Id)'
cabal: Error: some packages failed to install:
ghc-mod-1.10.8 failed during the building phase. The exception was:
ExitFailure 1

Sorting of import section

haskell-emacs has a neat feature: sorting alphabetically the import section.
How hard would it be to add it to ghc-mode ?

post-command-hook

For emacs versions which support it, probably it's better to use
(add-hook 'post-command-hook 'ghc-type-post-command-hook nil t)
in ghc-type-init instead of making local variable.
Since latter breaks several hooks, e.g.: eldoc-schedule-timer & global-hl-line.

Add "site" sub-command

It would be useful if the ghc-mod executable could output the path to its emacs site files. That way, it is possible to automate the configuration of ghc-mod in emacs by using (add-to-list 'load-path (shell-command-to-string "ghc-mod site")).

The command should simply output its cabal data directory. Not having to find that directory using elisp would be very useful.

check broken?

➜  leafy git:(master) ✗ cat Foo.hs
module Foo where

main = putStr "foo"

with 0.3:

➜  leafy git:(master) ✗ ghc-mod check Foo.hsFoo.hs:3:0:Warning: Definition but no type signature for `main'         Inferred type: main :: IO ()
Warning: output was redirected with -o, but no output will be generated because there is no Main module.

with git head:
➜ leafy git:(master) ✗ ghc-mod check Foo.hs

Integration of Haskell Refactoring Tool

I am wondering whether it would be possible to integrate HaRe into ghc-mod. I guess that HaRe has some rough edges right now but an optional integration into ghc-mod might get more people to work on HaRe.

With GHC 7.6, flymake info is no longer useful

When I type M-? on an error, I now see this most of the time:

Line 61: 2 error(s), 0 warning(s)

34:
36:

I have to use C-c C-l to get the actual error message from ghci. I pulled and rebuild ghc-mod earlier today.

assoc obsolete

Tried to run this on the Emacs pretest 24.2.90 from http://emacsformacosx.com/builds and it gives me an error that assoc is obsolete.

Expanding Template Haskell in Yesod Hello World causes error

If I type in the Yesod Hello world example into helloworld.hs:

{-# LANGUAGE TypeFamilies, QuasiQuotes, MultiParamTypeClasses, TemplateHaskell, OverloadedStrings #-}
import Yesod

data HelloWorld = HelloWorld

mkYesod "HelloWorld" [parseRoutes|
/ HomeR GET
|]

instance Yesod HelloWorld

getHomeR :: Handler RepHtml
getHomeR = defaultLayout [whamlet|Hello World!|]

main :: IO ()
main = warpDebug 3000 HelloWorld

and put point immediately after a |], and type C-c C-e, I get:

ghc-mod:0:0:Probably mutual module import occurred

ghc-mod:0:0:Probably mutual module import occurred

ghc-mod:0:0:Probably mutual module import occurred

in the GHC Info buffer.

emacs 24.2, ghc-mod 1.11.2, ghc 7.6.1

'ghc-mod check <file>' gives no output

I have tried this with ghc 7.0.3 and ghc 7.4 on Arch Linux with ghc-mod 1.10.* . I cannot get the ghc-mod check command to produce any output. Is this a symptom of something common. ghc-mod lint works fine.

quits on incorrect .lhs files

if i run ghc-mod check on incorrect literate Haskell, the preprocessor returns 1, so ghc-mod returns 1 - this kills the emacs process and turns flymake off, which is pretty annoying, especially given that the code will be wrong a lot of the time while hacking.

example code you get this on:

  here's some doco
  some more doco
  > main = putStrLn "oh no, i didn't put a newline in!"

ghc-mod version 1.10.15

vi mode: :wq behavior and run both check + lint on each save

Great work! I really enjoy using ghc-mod in vi. Two requests for the vi integration, if possible:

• Do not run any checks if :wq is issued, i.e., if the user wants to write-and-quit. It's a waste of time running the checks since they won't be displayed anyhow.
• I think the old behavior upon a ":w" command was to run both check and lint on the haskell file. Looks like the recent changes to the vi-mode altered this behavior and it currently runs just a check upon saving; and it runs lint only when explicitly requested. This behavior sounds OK, but the old behavior of running both check and lint at each save is also quite convenient as well. Maybe this can be configured in the .vimrc file via a flag.

fails if built with profiling libraries

usually i build with executable and library profiling. this causes ghc-mod to fail:

rhino ➜ ~/projects/dojo/ninjablocks-zeromq-workers git:(master) ✗ ghc-mod check src/reup/ReUp.hs
ghc-mod: You can't call hscCompileCoreExpr in a profiled compiler

reinstalling without library and executable profiling does the right thing,