bug#18347: 24.3.93; Incomplete splash screen display on Cygwin-w32 build

[Eli Zaretskii <eliz@gnu.org>]

> Date: Fri, 29 Aug 2014 11:08:21 -0400
> From: Ken Brown <kbrown@cornell.edu>
> CC: 18347@debbugs.gnu.org
> 
> On 8/29/2014 2:25 AM, Eli Zaretskii wrote:
> >> Date: Thu, 28 Aug 2014 17:51:06 -0400
> >> From: Ken Brown <kbrown@cornell.edu>
> >>
> >> --- lisp/startup.el     2014-07-08 09:17:09 +0000
> >> +++ lisp/startup.el     2014-08-28 20:07:20 +0000
> >> @@ -1812,7 +1812,7 @@
> >>      (let (chosen-frame)
> >>        ;; MS-Windows needs this to have a chance to make the initial
> >>        ;; frame visible.
> >> -    (if (eq system-type 'windows-nt)
> >> +    (if (eq window-system 'w32)
> >>           (sit-for 0 t))
> >>        (dolist (frame (append (frame-list) (list (selected-frame))))
> >>          (if (and (frame-visible-p frame)
> >
> > It is fundamentally wrong to use window-system the variable in such
> > circumstances (or almost any other).  Please use the function instead.
> 
> OK.  But could you explain why?  TIA.

Because we don't want to risk assigning it a value.  Also, the
function can accept a frame argument, thus giving you a chance to
inquire about a specific frame (different frames can have different
window-system values).

And if you use window-system as a predicate (not in this case,
obviously), you should use the various display-*-p predicates instead.

> >> Is this still OK for the native Windows build?  If so, is it OK to
> >> install it in the release branch?
> >
> > Yes and yes.  Although I'd urge you to try to figure out why this is
> > needed in the cygwin-w32 build (the native-build problem which led to
> > this code manifested itself in a rather different way, see bug#16014).
> 
> Actually, the visual symptom I'm describing (failure of the logo to 
> appear) is identical to what Juanma reported in that bug.

OK, but that's not what your original description said.

> Your message in that bug discussion says, "It's again that timing
> thing with making the initial frame visible."  The "again" suggests
> that there is an earlier bug report or discussion about that.  Can
> you give me a reference or just explain what you meant by that?

Bug#14841, I think.

Basically, since creating a frame on w32 takes some time until the OS
sets up the message pump for the new frame, Lisp code that creates a
frame and immediately proceeds with manipulating the new frame might
sometimes need sit-for to let the dust settle.