ALE's completion source for deoplete prevents you from pressing enter sometimes

[edgard]

I'm currently using deoplete for my completion needs. While it mostly works together with ale, there are still some annoyances that happens when the two are loaded in.

For example, right now, if a complete is started, i can't double-enter to just enter the word without any completion. ALE hijacks the menu and keeps it open.

Is there a way to complete disable all completion integration/functions on ALE?

[w0rp]

Use let g:ale_completion_enabled = 0, which is the default.

[edgard]

This is not really working. Even with the completion_enabled = 0, there seems to be some kind of hijacking happening on the completion menu.

https://asciinema.org/a/qmNGrVXR260LOYSxMUNJEiBtH

You can see that with ALE enabled when i type "set" and enter, it doesn't matter how many times i press enter, it never leaves the line and do the enter. Even with the option set. As soon as I disable ALE, it works.

[w0rp]

Which version of Vim are you using? That shouldn't happen.

[edgard]

Neovim 0.3.5

[w0rp]

Does it stop happening when you use :ALEDisable?

[edgard]

Nope, still happens. :(

[w0rp]

Have you tried setting let g:ale_completion_enabled = 0 before ALE is loaded? I recommend getting on #vim-ale on Freenode to talk about this more, as that will be faster.

[w0rp]

We spoke about this on IRC. A minimal config for repeating this bug is here: https://gist.github.com/edgard/8bb9c4d7596cc20a645f1f690c19007d

The problem appears to be caused by context['is_async'] = True in the completion source for deoplete, which is what the deoplete documentation tells you to do. This doesn't need to be done for languages ALE can't prevent completion data for, but it still breaks it anyway. Replacing return [] with return None at the end seems to fix the problem, even though the documentation tells you not to do that.

I've pushed that change now. Let me know if it fixes the bug.

[w0rp]

If that fixes the bug, then I'll also throw in a check later to disable the source if there aren't any language server linters available for the current filetype that ALE can use to provide completion information. That will prevent some more issues in the future.

[edgard]

Yep! Updated from repo and re-added noinsert to completeopt and now it seems to be working properly! Awesome! Thanks!

[w0rp]

Great! Thank you for letting me know about the bug. Later on, someone can tweak the completion source to not even try to provide completion data if the buffer doesn't have any ALE linters enabled which can provide completion data.

[dsifford]

So.... Hate to be the bearer of bad news, but... This update for whatever reason bricked ale's completion completely for me. Completion from ale is no longer listed at all.

[dsifford]

And I just confirmed that reverting the return None back to return [] fixes the issue.

[w0rp]

@dsifford Okay, do you want to fix it? Try to make it work without also causing the issue brought up here.

[dsifford]

Can't dig into it until maybe sometime next week. But just for the time being, can you or @edgard confirm that you're even able to get ale to completions through deoplete anymore?

Because if we "fixed" the issue of having to hit enter twice to insert a newline by bricking the ale completions completely, I'd consider that more of a regression than a fix and it should be reverted immediately until the real issue can be found.

[w0rp]

I'd rather not include Deoplete integration than have it cause problems for people. I think the easiest way to fix this would be to add the check to see if the buffer has linters enabled, and then return None if there aren't any. I'll do that soon, and then both things should work.

[w0rp]

@edgard @dsifford Okay, git pull again and try that out. Now None should be returned when there's nothing that ALE could get completion information from, and [] will be returned when ALE is able to ask for some completion data. Hopefully that should work.

[dsifford]

@w0rp You are the master. 🥇

After trying it for a minute or two, it seems to be working as expected again for me. I'll report back if something turns up, but looks like it should be good now. 🙏

[w0rp]

Great! 👍 I'll wait for a response from @edgard before closing this.

[edgard]

Just to confirm that it still works! Thank you!! :)

[w0rp]

Okay, thank you for letting me know. 👍

[moritzheiber]

I'm hitting this bug again right now, after having updated my plugins. As far as I can see the latest commit which is supposedly "fixing" this issue was included in the update. I also run Deoplete and Ale side-by-side. Forcefully disabling Ale (just commenting Plug w0rp/ale) "fixes" it, although that would probably not be my preferred solution. I'd be happy for any pointers.

[w0rp]

Would you like to try and fix it?

[moritzheiber]

@w0rp Absolutely, what should I do?

[w0rp]

Okay, I'm going to remove the Deoplete completion source until someone can fix it.

[moritzheiber]

For context, we had a chat on IRC and "solved" it by disabling deoplete as a completion source

[w0rp]

I remove Deoplete support completely for now, until this can be fixed.