bug#77928: Regression in Emacs pretest 31.0.90

> From: Roshan Shariff <[email protected]>
> Date: Mon, 27 Jul 2026 08:39:42 -0600
> Cc: [email protected], [email protected], [email protected], 
> 	[email protected]
> 
> On Mon, 27 Jul 2026 at 08:20, Eli Zaretskii <[email protected]> wrote:
> > I'm not even sure we all agree that it's a "regression".  That's why I
> > asked for opinions in the first place.
> 
> I won't go any further, because this is starting to seem Sisyphean.
> But I will point out that this bug report has extremely limited reach.
> Steven Allen very clearly noticed the change of behaviour, spent a
> great deal of time explaining it, and wrote a set of patches to try to
> correct it. Andy Moreton and Kristoffer Balintona also thought this
> was a regression. It could simply be that they're busy right now,
> especially after their earlier efforts didn't get much response (and
> the latter two are not specifically CC'd) . Except for Michael
> Shields, the patch author, everyone else who has an opinion preferred
> the status quo of previous Emacs versions. You yourself just said you
> wanted more clarification, which I attempted (and unfortunately,
> failed) to provide. I suspect, but of course can't know, that more
> people will report this as a bug when Emacs 31 is released.

Nevertheless, I would hesitate reverting to previous behavior for
Emacs 31.

However, I could perhaps agree to try this proposal of yours:

  Concretely, I'm proposing that the use-package :custom-face form
  should put its argument into the 'face-override-spec property. You can
  look at the implementation of face-spec-recalc to see what exactly
  that means, but briefly it is applied "on top of" either the defface
  spec or the user theme + customization. That broadly seems to me like
  the right semantics.

on the master branch.  If it produces good results, maybe we could
consider it for backporting in Emacs 31.2.