Eval plugin: mark exceptions

[xsebek]

This change will* format exception results in the same way as doctest:

>>> if error "-1" then 1 else 0
------------------
-- current:
-1
------------------
-- new:
*** Exception: -1

*) EDIT: if the configuration option haskell.plugin.eval.config.exception is set to true.

• Fixes Eval Plugin: different exception rendering than doctest #2353

[xsebek]

Hi @jneira, there seems to be no code owner of the eval plugin - do you know who might be familiar with the code? 🙂

[pepeiborra]

Looks good to me.

[Ailrun]

This no "Exception" message is intended AFAIK: https://github.com/haskell/haskell-language-server/tree/master/plugins/hls-eval-plugin#multiline-output

I think we cannot change this at least before we have somewhat "proper" handling of IO.

[Ailrun]

C.C. @tittoassini

[xsebek]

This no "Exception" message is intended AFAIK: https://github.com/haskell/haskell-language-server/tree/master/plugins/hls-eval-plugin#multiline-output

Thanks, @Ailrun, I was not aware of this hack. I do find the missing message a bit confusing, but if someone is using error to emulate putStr then I can let this rest for a while.

[pepeiborra]

I was not aware of this hack either! We should strive to make this hack unnecessary, some options come to mind:

• Capturing output can be solved by using the external interpreter (-fexternal-interpreter). We should revisit using it for the eval plugin, but this might be challenging or even impossible.
• We could wrap all the user code in a custom exception type with a crafted Show instance that doesn't escape output. Then we catch all other exception types and add the "*** Exception" markup.

What do you think @Ailrun ?

[Ailrun]

@pepeiborra I think the second would be more feasible. If we can handle IO better, that would be better, but that may take quite effort

[xsebek]

Hi @pepeiborra, I am not entirely sure what you mean by the second option?

I imagine this could be hacked around by recognizing some prefix like "PRETEND_I_AM_IO: ".

If supporting this hack should involve more effort, I would rather attempt to fix the #1977 instead 🙂

[pepeiborra]

Sorry I wasn't clear: The only way to properly fix #1977 is by using the external interpreter, which is far from obvious.

[xsebek]

Yes, I understood that the first option is fixing #1977 with an external interpreter though I have no idea how to do that (yet).

What I can not imagine is how to "wrap all the user code" which you suggested as the second option. 🤔

[pepeiborra]

Yes, I can see now that the 2nd option is not trivial either.

[xsebek]

Another solution that comes to mind would be to make this configurable by the user.

Currently, the eval plugin has the #diff property for WAS/NOW (which is passed to runTests so it's almost there).

So I suggest adding an #exception property (False by default) with the hint:

Enable marking exceptions with *** Exception: similarly to doctest and GHCi.

What do you think, @pepeiborra, @Ailrun? It seems relatively low effort and once we get better IO support, we could just turn it on by default or keep it off. 🙂

[xsebek]

OK, the behaviour is now configurable via haskell.plugin.eval.config.exception (default false) and there is one more test for this configuration option.

Does this resolve your concerns @Ailrun? Btw. thanks for the scaffolding around diff. 🙂

[Ailrun]

LGTM :) Here is one minor comment for documentation.

[Ailrun]

It looks like a link is missing?

[xsebek]

Thanks, I was about to look up how to link sections and got distracted. 😅

Fixed ✔️

[xsebek]

At the last minute, I remembered to check the docs for configuration options and added the eval section. 😅

I wonder if it needs to be added to vscode-haskell too?

[michaelpj]

Docs look fine to me. The way the config options are documented is suboptimal, but this doesn't make it worse!

I don't know what the deal is with exposing options in vscode-haskell...