Ivy completion-in-region not working correctly (I think?)

[dieggsy]

If I drop down to eval-expression with M-: and type e.g. (reg-quo TAB, I get (reg-regexp-quote, instead of the expected regexp-quote (which works just fine with emacs -q).

[abo-abo]

Can't reproduce with make plain on Emacs 26.

[dieggsy]

Huh, I should mention I'm using emacs built from really recent git (specifically emacs-mirror/emacs@7367ea4), and can still reproduce with make plain. Guess you can close as unsupported, if you wish?

[basil-conto]

I can't reproduce it from make plain with Emacs <= 25, but I can with 27 (can't test 26 at the moment). Ivy doesn't (and shouldn't) specify a maximum supported version, so this issue should remain open until future versions of Emacs are supported.

[abo-abo]

So was completing-read-function API broken between versions 26 and 27? I'll try to bisect.

[abo-abo]

Reproducible for Emacs 27, commit 0ffd3dbce76c1a967522dbe9ec6f2dffe94ee886.

[basil-conto]

I don't think it's because of an API change, but I have a feeling it may be related to the completion-common-part face being applied differently across Emacs versions by completion-all-completions.

[abo-abo]

I think applying completion-common-part is part of the API.

[basil-conto]

I think applying completions-common-part is part of the API.

I disagree:

1. completions-common-part is completely undocumented in the Elisp manual.
2. completions-common-part pertains to fontification, whereas we're talking about completion.

[basil-conto]

Relying on a face for completion is unsound, especially given that Emacs offers multiple completion-styles which change from version to version. Having said that, I'm not 100% certain this face is to blame, but it sounds like it could be related.

[abo-abo]

(defface completions-common-part '((t nil))
  "Face for the common prefix substring in completions.
The idea of this face is that you can use it to make the common parts
less visible than normal, so that the differing parts are emphasized
by contrast.
See also the face `completions-first-difference'.")

I think this face is defined not only for aesthetic reasons, but also as an output parameter of completion-all-completions. If we can't use completion-all-completions or a suitable replacement, we would have to implement a fork of minibuffer.el with the equivalent functionality. I'd like to avoid doing that.

If you can think of some better way of implementing ivy-completion-common-length, I'd like to hear it. I'm not happy with looking at face properties, but I haven't seen anything better.

[abo-abo]

Found the commit: 325ef57b0e3977f9509f1049c826999e8b7c226d. It's possible to recover in ivy-completion-common-length since the info returned by completion-all-completions still has the info we need, only in a different format.

[basil-conto]

I'm not happy with looking at face properties, but I haven't seen anything better.

I appreciate that, and there's no obvious alternative as far as I know. I need to familiarise myself with Ivy and Emacs in-buffer completion before looking into this, but at least it's not highest priority given Ivy still works for all released Emacs versions.

[dieggsy]

at least it's not highest priority given Ivy still works for all released Emacs versions.

agreed, for what it's worth :)

[basil-conto]

It's possible to recover in ivy-completion-common-length since the info returned by completion-all-completions still has the info we need, only in a different format.

Great. I just have a quick suggestion for a simplification in ivy-completion-common-length while we're at it: let-binding char-property-alias-alist to '((face font-lock-face)) as font-lock-mode does means you only need to query for the face property, which will automatically be translated to font-lock-face if applicable. In other words, the following simplification is possible in the current version of ivy-completion-common-length:

diff --git a/ivy.el b/ivy.el
index 3538b92..2b4ea10 100644
--- a/ivy.el
+++ b/ivy.el
@@ -2029,19 +2029,14 @@ ivy-completion-common-length
   "Return the length of the first `completions-common-part' face in STR."
   (let ((pos 0)
         (len (length str))
-        face-sym)
+        (char-property-alias-alist '((face font-lock-face))))
     (while (and (<= pos len)
-                (let ((prop (or (prog1 (get-text-property
-                                        pos 'face str)
-                                  (setq face-sym 'face))
-                                (prog1 (get-text-property
-                                        pos 'font-lock-face str)
-                                  (setq face-sym 'font-lock-face)))))
+                (let ((prop (get-text-property pos 'face str)))
                   (not (eq 'completions-common-part
-                           (if (listp prop) (car prop) prop)))))
+                           (or (car-safe prop) prop)))))
       (setq pos (1+ pos)))
     (if (< pos len)
-        (or (next-single-property-change pos face-sym str) len)
+        (or (next-single-property-change pos 'face str) len)
       0)))
 
 (defun ivy-completion-in-region (start end collection &optional predicate)