Quitting a completing-read performed in the middle of a completion will break Company

[vemv]

CIDER can perform a (completing-read) while a Company popup is being shown. That is accomplished here:

https://github.com/clojure-emacs/cider/blob/5edfbc0bedef4c751f2e753477eb3ca4986bf0f0/cider-client.el#L484

Here's a sample of it working:

What it accomplishes is to disambiguate the candidate, as it may come from different Java classes. This way, Company will show the right documentation.

This techique works, except if the user aborts with C-g. In that case, completion will no longer work until the buffer is closed and opened again.

I will see in *messages*: Error in post-command-hook (company-post-command): (quit)

It's as if some internal state wasn't cleaned up afterwards?

Thanks - V

[dgutov]

Sounds like this will need a condition-case to catch the quit signal. Around doc-buffer, right? This doesn't happen for meta?

[vemv]

Around doc-buffer, right?

Maybe, if you provide a patch or branch I'll happily try it out

This doesn't happen for meta?

What does that mean?

[dgutov]

What does that mean?

Could this happen in the :company-docsig handler as well?

Maybe, if you provide a patch or branch I'll happily try it out

This might suffice:

diff --git a/company.el b/company.el
index a650cbc..5d732e2 100644
--- a/company.el
+++ b/company.el
@@ -2232,7 +2232,8 @@ For more details see `company-insertion-on-trigger' and
                 (when company-auto-update-doc
                   (condition-case nil
                       (company-show-doc-buffer)
-                    (user-error nil))))
+                    (user-error nil)
+                    (quit nil))))
             (let ((delay (company--idle-delay)))
              (and (numberp delay)
                   (not defining-kbd-macro)

[vemv]

Thanks much for the prompt response!

I can confirm that the patch fixes it - after applying it, I can proceed after C-g. No other changes were needed.

Could this happen in the :company-docsig handler as well?

Not that I could see. company-echo-metadata-frontend kept working.

[dgutov]

I can confirm that the patch fixes it - after applying it, I can proceed after C-g. No other changes were needed.

Thanks for testing, I can apply and push it now.

Something else to consider: do you want these prompts? CIDER could consult some variable (e.g. non-essential) and skip the prompt when the doc-buffer display is initiated not by a user command, but by the automatic post-command action. Then it would either choose the first alternative among the set, or concatenate all of the possible docstrings in the buffer.

[vemv]

Cheers.

I'm pretty confident that we want these prompts. These Java docstrings can be wildly long and different across classes, so a guess can misinform the user, and a concatenation wouldn't be quite readable.

I appreciate the concern though. I created clojure-emacs/cider#3501 for making the experience more streamlined (only for the particular Company + IDO combination though) - feel free to check it out.

[dgutov]

These Java docstrings can be wildly long and different across classes, so a guess can misinform the user, and a concatenation wouldn't be quite readable.

Makes sense. But in the linked issue you're referring to the same problem (flooring <down> being made impossible, or C-s, or etc).

What I'm suggesting is for cider to display the first alternative anyway, but you would be able to press C-h explicitly, then see the prompt and choose the right class. This is slightly less obvious for the user (not seeing the prompt originally), but it could OTOH work across completion systems, and you wouldn't worry about which bindings need the override.

Anyway, I'll follow the other discussion, let's see what you settle on.