[Discussion] REPL vs IDE

[gracjan]

We feel that there are two important use cases for a running Haskell interpreter process:

1. REPL-based: load/reload only on command, any interactive command allowed, interpreter window interacts with the user
2. IDE-based: load-on-save/visit, -Wall by default for warning overlays, interacts with user only with predefined set of commands, results are parsed and presented

The discussion was spawned by @deepfire comment: #795 (comment).

Lets discuss this avenue here.

[gracjan]

We have a bit more information:

DanielG/ghc-mod#544

In this case we should settle for the following:

1. ghc-mod serves the IDE-mode workflow
2. ghci serves the REPL-mode workflow

[deepfire]

Does this mean we will have two haskell processes running?

[gracjan]

ghc-mod should be running to get error highlights, navigate to definition, show types, etc. Repl should be running if you actually want to run the program. You are free to skip either of those. Or both. Those two use cases do not go well together, in my opinion. There are rumors that ghc-mod should also be able to 'run' things, but I do not know the details of it.

[DanielG]

I'm planning to add interactive code evaluation support to ghc-mod so then both of these would happen at the same time. In ghc-mod we first need to add server-client RPC support though as this would have to run as a separate process probably.

[gracjan]

Great to hear! Would situation like infinite loops, garbage collection, Ctrl-C, terminal handling not get in the way of the IDE functions of ghc-mod? I would hate to hang Emacs just because my Haskell program hangs.

[alanz]

I think if the right/separate comms channels are established between the ide and ghc-mod this shouldn't be a problem, even if it is one physical channel with streams multiplexed onto it.

[gracjan]

I have had bad experiences with such setups. Just look at Chrome and each-tab-is-a-process-so-we-can-kill-it-separatelly as a very successful paradigm.

[DanielG]

The problem is that the memory consumption of a single ghci process is already insane in the right project and running ghc-mod and ghci at the same time is just a huge waste.

[gracjan]

Can't deny that. Well, I just like the feeling of safety when I can kill this or the other should it go berserk.

[acowley]

I'd definitely like it to be one process, but it should of course be a top priority to avoid hanging the editor.

@DanielG is there are branch where you are working on the design of adding interactive evaluation to ghc-mod?

[gracjan]

Looks like HIE is better forum to discuss this.