GHCJS support?

[NickSeagull]

I'm currently trying to use ghcid for a project, while for the backend it works perfectly, the frontend ghcjsi doesnt seem to be reloaded on file changes. Is there a reason for this?

What would it take to make ghcid work with GHCJS? I'd be happy to submit a PR to fix it :smile:

[ndmitchell]

I don't know what the problem is, but would accept a PR to fix it. If you pass --verbose to ghcid it should say what it's sending, which might give you some clues.

[NickSeagull]

It spits out some errors:

Loading cabal new-repl frontend ...
%STDERR: Warning: don't know how to find change monitoring files for the installed
%STDOUT2: Warning: don't know how to find change monitoring files for the installed
Warning: don't know how to find change monitoring files for the installed
%STDERR: package databases for ghcjs
%STDOUT2: package databases for ghcjs
package databases for ghcjs
%STDOUT: Resolving dependencies...
%STDOUT2: Resolving dependencies...
Resolving dependencies...
%STDOUT: Build profile: -w ghcjs-0.2.0 -O1
%STDOUT2: Build profile: -w ghcjs-0.2.0 -O1
Build profile: -w ghcjs-0.2.0 -O1

Any pointers about where should I start to look @ndmitchell ?

[ndmitchell]

It spits out some messages, but not sure any of those are fatal - Ghcid is just echoing them from the underlying command. I'd look elsewhere - or perhaps share a complete log.

[NickSeagull]

Taking a go with the complete log, perhaps you can see something obvious I cant :smile:

full log

[ndmitchell]

Does :set prompt foo work in new-repl? How much like normal ghci is the experience? I would be expecting stdin lines to be in the output, but I can't find them. Which version of ghcid are you using?

For info, the prefix of my session says:

C:\Neil\hlint>ghcid --verbose
Loading ghci -fno-code -fno-break-on-exception -fno-break-on-error -v1 -ferror-spans -j ...
%STDOUT: GHCi, version 8.2.2: http://www.haskell.org/ghc/  :? for help
%STDOUT2: GHCi, version 8.2.2: http://www.haskell.org/ghc/  :? for help
%STDIN: import qualified System.IO as INTERNAL_GHCID
%STDIN: :set prompt #~GHCID-START~#
%STDIN: :set -fno-break-on-exception
%STDIN: :set -fno-break-on-error
%STDIN: :set -v1
%STDIN: :set -fno-hide-source-paths
%STDIN: INTERNAL_GHCID.putStrLn ['#','~','G','H','C','I','D','-','F','I','N','I','S','H','-','1','~','#']
INTERNAL_GHCID.hPutStrLn INTERNAL_GHCID.stderr ['#','~','G','H','C','I','D','-','F','I','N','I','S','H','-','1','~','#']
GHCi, version 8.2.2: http://www.haskell.org/ghc/  :? for help

e.g. full of STDIN.

[NickSeagull]

Yes, setting the prompt works.

cabal new-repl is exactly like cabal repl except that it uses the package management capabilities of cabal to behave like stack.

Looks like the problem is definitely ghcjs, as if I run ghcid -c 'cabal new-repl' on the backend cabal project, it works.

[ndmitchell]

Can you confirm you are on the latest ghcid version? And which version that is? ghcid --version should tell you.

Does putStrLn "foo" work in ghcjs, or does the output go to a web browser?

[NickSeagull]

Forgot to mention the version, I got ghcid v0.7 from Stackage. Should I stack install the one in this repo?

On the other hand, yes, it does print to STDOUT, the output doesnt go to the browser.

[ndmitchell]

0.7 is fine. I'm not really sure why it doesn't work, or why mine shows the stdin lines. Does yours show stdin lines in the cases it does work?

[NickSeagull]

Yep, with regular ghc it shows STDIN

[ndmitchell]

Reading the logs more closely, I found why it wasn't printing out stdout (just read the diff above and it will be obvious!). Can you compile HEAD and confirm that it works in practice?

[NickSeagull]

Now the STDIN part shows up! :smile:

Although it still doesnt reload the code :(

Here's the log:

https://gist.github.com/NickSeagull/66b00654d9304e935b54150e3723dd95

[ndmitchell]

Progress!

Next problem is it seems that INTERNAL_GHCID.hPutStrLn INTERNAL_GHCID.stderr ['#','~','G','H','C','I','D','-','F','I','N','I','S','H','-','1','~','#'] doesn't result in a message being put on stderr. Can you try that by hand (with just hPutStrLn stderr "hello") and confirm?

[NickSeagull]

Looks like it worked properly:

*Main System.IO> hPutStrLn stderr "hello"
hello

[ndmitchell]

Can we confirm the message ends up on stderr and not stdout? Perhaps run ghcijs piping stderr to a file, eg "ghcijs 2> stderr.txt" - then we can see if it ends up on the screen or not.

[NickSeagull]

Looks like stderr ends up in stdout as it prints the string on screen. Any ideas what could be done next?

[ndmitchell]

Could that be fixed in ghcjs? I can think of a workaround, but always better to fix at the source.

[NickSeagull]

Ok, let me see if I can open an issue for them

[ndmitchell]

Cheers, please cc me on the issue as well.

[ndmitchell]

See https://github.com/ghcjs/ghcjs/issues/671