Set enable-recursive-minibuffers automatically

[ashton314]

Expected Behavior

Let's say I want to evaluate the variable magit-completing-read-function from the minibuffer. I'd hit M-: to bring up the Eval: prompt, then I'd type magit- and expect to see a list of viable variable candidates.

Current Behavior

After bringing up the Eval: prompt, I type magit-, hit TAB and I get an error saying Command attempted to use minibuffer while in minibuffer.

Steps to reproduce

1. Start Emacs with emacs -q
2. Load Selectrum (I ran load-file and found selectrum.el in my .emacs.d/ directory)
3. Type M-: ema TAB (I'm not sure what the completion candidates are here

It seems to only happen if there are candidates to complete. If you type a string like asdfasdf it will just say no matches in the minibuffer.

[Raphus-cucullatus]

I met the same issue. Here is the backtrace (with some long line truncated):

Debugger entered--Lisp error: (error "Command attempted to use minibuffer while in minibuffer")
  read-from-minibuffer("Completion: " nil (keymap (9 . selectrum-insert-current-candidate) (10 . selectrum-submit-exact...
  selectrum-read("Completion: " (... ... ... ... ... ... ... ... ... ... ... ... ... ... ...) :may-modify-candidates t)...
  selectrum-completion-in-region(#<marker at 8 in  *Minibuf-1*> 11 [0 0 ansi-color-apply-overlay-face isearch--state-po...
  completion-in-region(#<marker at 8 in  *Minibuf-1*> 11 [0 0 ansi-color-apply-overlay-face isearch--state-pop-fun--cma...
  completion-at-point()
  funcall-interactively(completion-at-point)
  call-interactively(completion-at-point nil nil)
  command-execute(completion-at-point)
  read-from-minibuffer("Eval: " nil (keymap (9 . completion-at-point) (27 keymap (9 . completion-at-point)) keymap (rem...
  read--expression("Eval: ")
  byte-code("\301\302!\303\10!B\207" [current-prefix-arg read--expression "Eval: " eval-expression-get-print-arguments]...
  call-interactively(eval-expression nil nil)
  command-execute(eval-expression)

Completion in minibuffer incurs selectrum-completion-in-region who calls selecttrum-read who itself uses the minibuffer.

Completion in minibuffer is useful: when type elisp with"M-:", when enter shell command with "M-!", etc. So it is quite frustrating that selectrum makes completion in minibuffer not work.

As a temporary workaround, I checkout to commit prior to 8857a8f. Hope this can be fixed soon.

[clemera]

I think you both forgot to set enable-recursive-minibuffers to t? @raxod502 Maybe selectrum-mode should do this by default?

[Raphus-cucullatus]

You are right! After setting enable-recursive-minibuffers to t, it works. Thanks!

[Raphus-cucullatus]

Hi @clemera , but here comes another issue when in a remote buffer (TRAMP),

• If I checkout 2b49ab4 (so that no selectrum-completion-in-region), if I use "M-! ls ~/" then "<TAB>", home files on this remote machine are shown. This is expected.
• But if I use master and set enable-recursive-minibuffers to t, home files on my local machine are shown.

Do you experience this, or it is again due to my incorrect config?

[clemera]

Thanks for discovering this, that seems to be a bug with file completion for selectrum-completion-in-region. Could you open another issue for that?

[Raphus-cucullatus]

OK, thanks again for your timely reply!

[raxod502]

I am a little reluctant to automatically bind enable-recursive-minibuffers. After all, some people want the default behavior (perhaps it was Stefan Monnier who said he liked it?), and if we bind the variable to non-nil automatically, then there's no way for the user to enforce that setting.

On the other hand, I suppose we could have a variable selectrum-enable-recursive-minibuffers which, if non-nil, means enable-recursive-minibuffers gets set to non-nil. I think that would be a fine idea, and quite easy to add. The value of this new variable could be non-nil by default.

[ashton314]

I think that's a good idea. If I get some time tomorrow I can try my hand at creating a PR for that.

[DivineDominion]

Ran into this today and was puzzled :) Couldn't selectrum catch the error defensively (checking if that's enabled) and then display a more helpful error message in the minibuffer? Like

Requires enable-recursive-minibuffers to be non-nil for completion to work here

[raxod502]

Yes, that seems like a good idea, I agree.

[mathrick]

I just ran into this issue now that I've switched over to selectrum. Couldn't selectrum automatically bind it to t when it's asked to do completion, ie. if I'm inside M-: and press <TAB>, I clearly want to get completion. I think at this point selectrum should simply make sure that won't error out with (let ((enable-recursive-minibuffers t)) ...), so that it can provide the user with the requested functionality without changing the defaults elsewhere.

Alternatively, maybe use a popup buffer if recursive minibuffers are disabled, the way default completion (and icomplete) do in this situation? I don't know how much of a rewrite making selectrum able to use a non-minibuffer buffer would be.

[clemera]

Thanks @mathrick for bringing my attention to this issue again. With #397 the value is let bound for the commands which need it, this is also what the built-in describe-... commands do.