Different suggestions for neovim and copilot.el ?

[jkbjh]

Thanks for copilot.el, nice to be able to try this out in emacs.

I'm experiencing some interesting behaviour comparing completions in neovim vs. copilot.el;
with the following code in neovim I get ten different solutions while copilot.el gives me a single one.

import jax
import haiku as hk

class Auto

(Completing after "Auto")

Is this an expected behaviour?

[zerolfx]

(Please update copilot.el, there is a related fix.)

Have you tried copilot-next-completion? I could get two or three solutions in this way.

How did you get those ten solutions? If it was by opening a side panel, I would say it is an unimplemented feature.

[jkbjh]

Yes, the ten solutions was with :Copilot panel in neovim.
However, I still get only one completion;
I also see

(:jsonrpc "2.0" :method "LogMessage" :params
	  (:level 1 :message "[INFO] [post-insertion] [2022-08-13T16:06:55.042Z] Could not get document for /tmp/test2.py. Maybe it was closed by the editor." :metadataStr "[INFO] [post-insertion] [2022-08-13T16:06:55.042Z]" :extra
		  ["Could not get document for /tmp/test2.py. Maybe it was closed by the editor."]))

in the *copilot events*, maybe this is related?

To update I've executed straight-pull-package for copilot, and also checked the repository in ~/.emacs.d/straight/repos/copilot.el.

Emacs version is Emacs 27.1.

[zerolfx]

The side panel feature is the only missing feature compared to copilot.vim, but I have no plan to implement it.

I believe you can cycle solutions via copilot-next-completion or it will tell you only one completion is available.

[zerolfx]

I investigated that log before, but I found copilot.vim had similar log messages, so I think it's okay.

[jkbjh]

Sometimes there are more completions available, then copilot-next-completion works for me.
for this minimal snippet

import jax
import haiku as hk

class Autoencoder

cycling through completions in nvim gives me three variants, copilot.el only one.

For your example,

import jax
import haiku as hk

class S

both nvim and copilot.el give two completions each, but different ones.
Quite interesting.

Now I retried it with copilot.el five minutes later and I'm back to getting only one.

[zerolfx]

I observed the same behavior.

You can examine the *copilot events* buffer, especially getCompletionsCycling requests and corresponding responses. I think the requests are the same, but Copilot gives different results.

[jkbjh]

Interesting, then this is probably just how copilot currently works.