bug#81666: 31.1; browse-url fails in emacsclient -nw

> From: समीर सिंह Sameer Singh <[email protected]>
> Date: Fri, 21 Aug 2026 02:45:33 -0500
> Cc: [email protected], [email protected], [email protected], 
> 	[email protected], [email protected]
> 
> Eli Zaretskii <[email protected]> writes:
> 
> >> From: Abdulnafé Toulaïmat
> >>  <[email protected]>
> >> Cc: Po Lu <[email protected]>,  Stefan Monnier <[email protected]>,
> >>   martin rudalics <[email protected]>,  Yuuki Harano <[email protected]>,
> >>   [email protected]
> >> Date: Fri, 21 Aug 2026 08:55:36 +0200
> >>
> >> Eli Zaretskii <[email protected]> writes:
> >>
> >> > This error is specific to a PGTK build, and comes from browse-url here:
> >> >
> >> >     ;; When connected to various displays, be careful to use the display of
> >> >     ;; the currently selected frame, rather than the original start display,
> >> >     ;; which may not even exist any more.
> >> >     (let ((dpy (frame-parameter nil 'display))
> >> >           classname)
> >> >       (if (stringp dpy)
> >> >         (cond
> >> >          ((featurep 'pgtk)
> >> >           (setq classname (pgtk-backend-display-class))  <<<<<<<<<<<<<<<
> >> >           (if (equal classname "GdkWaylandDisplay")
> >> >               (progn
> >> >                 ;; The `display' frame parameter is probably wrong.
> >> >                 ;; See bug#53969 for some context.
> >> >                 ;; (setenv "WAYLAND_DISPLAY" dpy)
> >> >                 )
> >> >             (setenv "DISPLAY" dpy)))
> >> >
> >> > All this is conditioned on "(frame-parameter nil 'display)" returning
> >> > a string, which AFAIU is not expected in the -nw case.  So the first
> >> > question is: what does "(frame-parameter nil 'display)" yield in your
> >> > case, in the scenario you describe, just before you invoke browse-url
> >> > from the client frame?
> >> >
> >> > (FWIW, I tried to reproduce this in a PGTK build I have here, but
> >> > couldn't: I see no error in this case.  And frame-parameter returns
> >> > nil in my case, as expected.)
> >>
> >> Hello, thanks for looking into this.
> >>
> >> In the emacsclient TTY frame, I get ":0". This does indeed appear to be
> >> the main culprit, because in an *emacs* TTY frame, I get nil, as you
> >> describe.
> 
> With some grepping it appears to be because of commit:
> e2b64f8999f79a5820ba00d2987885d7dda492d5
> 
> which solves bug#54844

Thanks, but I still don't see why this should cause the problem in the
PGTK builds.  That change only affects GUI frames, or at least that's
my reading of the code.