bug#81407: 31.0.90; Crash in display_line: stale glyph_row after window-config change from menu-bar :enable eval during redisplay

> From: Sean Whitton <[email protected]>
> Cc: [email protected]
> Date: Fri, 14 Aug 2026 11:35:49 +0100
> 
> Eli Zaretskii [16/Jul  1:05pm +03] wrote:
> >> From: Aaron Iba <[email protected]>
> >> Date: Wed, 15 Jul 2026 09:23:56 -0400
> >> Cc: [email protected]
> >>
> >> > We need something smarter, and perhaps more radical.  Let me think
> >> > about this.
> >>
> >> Understood, and agreed on the infloop risk: in the recipe the second
> >> pass happens to stabilize only because the rows array moves just when
> >> the window exceeds its historical maximum height, so a retry
> >> succeeds; Lisp that alternates between two window configurations
> >> would defeat any plain retry scheme.  Happy to test whatever you
> >> settle on -- I can reproduce at will, both with the -Q recipe and
> >> with the original configuration.
> >
> > Try the patch below, it seems to avoid the trouble here.
> 
> It seems that this patch makes an improvement; shall we install it on
> the release branch?

I'm not sure.  If anything, I'd install on master.

> Also, do we have any reason to think this is a bug introduced in Emacs
> 31?  It seems like it is probably older.  I don't have a macOS machine
> to test on.

It's a very old bug.  Basically, no one could expect Lisp called from
a menu-bar's :enable form or from :eval of the mode line to start
deleting windows.