Fix vanilla Emacs completions

[rudolf-adamkovic]

Describe the bug

Citar does not work well with Emacs' default completions.

To Reproduce

Steps to reproduce the behavior:

1. Run citar-open
2. Type TAB

Expected behavior

The completions buffer appears, like in M-x or C-x b.

Emacs version:

29

Additional context

• When I type something at the "Reference:" prompt and then press TAB, nothing happens either. The newly typed text disappears and Citar asks again (and again).

• Everything works well with Vertico.

[bdarcus]

AFAIK, citar should work with default completion. But it sounds like there may be some incompatibility with citar--select-multiple, and maybe the related citar--setup-multiple-keymap (assuming there's some issue with the exit keybinding)?

Do you get the same behavior, for example, with citar-open-entry (which doesn't use citar--select-multiple)? I'd guess not.

I don't really have time or expertise to figure this out. Do you think you could figure it out?

[rudolf-adamkovic]

Do you get the same behavior, for example, with citar-open-entry (which doesn't use citar--select-multiple)? I'd guess not.

It behaves the same way. After I execute citar-open-entry, Citar asks "Reference:". Then, I type TAB, and Citar says in "Reference: [sole completion]". Similarly, when I type "a" followed by TAB, Citar deletes the "a" and does nothing. No completions show in either case.

I don't really have time or expertise to figure this out. Do you think you could figure it out?

I know next to nothing about the completion system in Emacs. 😞

For the back story, I had this idea to try the built-in vanilla completion system in Emacs, and after configuring it, I discovered that it works rather well everywhere, except for Citar. Hence, I opened this issue.

[bdarcus]

I really have no clue, other than that it might be some keybinding issue.

I just tried to turn off vertico-mode and see if I could reproduce. I was able to get completions to pop up in a citar command, and select candidates with the mouse. But I couldn't figure out how to move up and down the list, to test what happen with RET!

[rudolf-adamkovic]

I was able to get completions to pop up in a citar command, and select candidates with the mouse.

Interesting. Just by pressing TAB?

[bdarcus]

I think I had to type a character first, but yeah.

[rudolf-adamkovic]

Thank you @bdarcus! That helped. I investigated further, seeing that it worked for you, and found that Citar works perfectly with (setq completion-styles '(basic)) but breaks with (setq completion-styles '(orderless basic)). Not sure why this poses a problem for Citar specifically. Should I close the bug?

[bdarcus]

I investigated further, seeing that it worked for you, and found that Citar works perfectly with (setq completion-styles '(basic)) but breaks with (setq completion-styles '(orderless basic)).

@oantolin - any ideas why this might be the case?

I don't really know anything about the default completion system.

[oantolin]

This is mysterious! I do see this weird behavior where, using default completion and orerless, TAB deletes what you have written. On the other hand, if you just don't press TAB, everything else seems to work with orderless: you can run switch-to-completions (bound by default to M-v) the completions popup and seem correct, and pressing RET (or clicking with the mouse) on one of them does select it correctly. So it's actually usable with orderless, but of course we should try to figure out why TAB isn't working.

[oantolin]

It seems citar binds TAB to minibuffer-complete-and-exit. In default completion that function is usually bound to RET and TAB is usually bound to minibuffer-complete. Citar's binding does seem reasonable since citar-open is basically completing-read in a loop, until you press RET to run citar--multiple-exit.

As we've said, just pressing TAB seems to just delete text and not select anything when using citar with orderless under default completion; but I discovered that if you first run minibuffer-complete you do see the correct completion appear in the minibuffer, and if then you press TAB, it does get selected. Now, that doesn't make sense, because I'd expect minibuffer-complete-and-exit to be basically the same as first running minibuffer-complete and then exiting!

EDIT: Sorry, that's wrong! I thought I was running minibuffer-complete with my old binding for <backtab> (shift + TAB), but I was running minibuffer-force-complete. Running minibuffer-complete deletes the text too, which is a good clue.

[oantolin]

I figured it out! It's a small bug in citar's completion table. The function citar--completion-table creates a predicate to pass to complete-with-action that combines the completion system's predicate with a citar filter. The combined predicate correctly calls filter on the citation key, and should call predicate on the display string of the citation but instead mistakenly calls it on the string the user is trying to complete in the minibuffer!

I'll submit a tiny pull request to correct this.

[oantolin]

That was initially pretty mysterious and was fun to figure out. 😃

[aikrahguzar]

@oantolin, that kind of thing is very hard to debug so thanks for figuring it out. But I am surprised the issue isn't present in vertico.

Citar's multiple selection is also a bit of hack where the intent was to support vertico without depending on it. It seems like it works for default completion too which is nice and probably a consequence of vertico's commitment to consistency with the default completion. I remember trying the fido-vertical-mode and icomplete-mode when I was working on it and they didn't work with multiple completion because the rebindings in minibuffer-with-setup-hook didn't seem to work. Multiple selection was off by default back then so I didn't do investigate much. I wonder if this problem was responsible? Now that multiple selection is default maybe if we should guard it with vertico-mode or maybe check if completing-read-function is completing-read-default?