Prevent hie crash if hlint crashes

[fendor]

Related to #1143

Also improves stability of hie, thus related to #1146
If hlint chokes on anything, dont kill hie.
An error message might be desirable but I dont know how to do that.

[mpickering]

There are already other functions which catch exceptions in HIE. Did you look at using one of those rather than rolling your own using try? In particular catchingSomeException seems like quite a broad brush.

[fendor]

Indeed, catching SomeException is terrible.

[fendor]

Now we catch IOExceptions!

[samuelpilz]

@mpickering I am actually in favor of aggressive error-catching from hlint. I think the hie language server should not be brought down by a internal error in any plugin like hlint.

It is the best thing for hie to do is to stop the plugin hlint and possibly restart it since hie is still quite useful if one plugin is removed.

[mpickering]

@power-fungus The problem is catching async exceptions such as interrupts.

[samuelpilz]

Indeed, async exceptions should not be caught

[lukel97]

I would argue that language servers should be a bit overdramatic and crash frequently: VS Code automatically restarts the server if it dies, and from what I recall with the ghc-mod plugin, the design is to propagate as many fatal errors as possible

[fendor]

Hlint errors dont have to be fatal ones though. In that case, hlint would just not work, while other functionality is still working.

[Anrock]

I'm against being overdramatic, too. While VSCode may handle this well, LanguageClient-neovim stumbles on server crashes and i suspect other client may do this too.

As a side note: LSP should be general overall, so coding against behaviour of specific client isn't right thing to do - remember embrace&extend&extuinguish. There are plenty of VSCode-specific quirks in other clients/servers already.

[Avi-D-coder]

The problem with crashing frequently is many language clients, I have used don't restart servers. If the server can recover it should try to. If possible plugins should be restarted.
A message on stderr might also be a good idea. Does anyone know how many clients handle stderr?

[alanz]

LGTM, now it is catching IOException.