Eli Zaretskii <eliz@gnu.org> writes:

>> +void ns_make_frame_key_window (struct frame *f)
>> +{
>> +  [[FRAME_NS_VIEW (f) window] makeKeyWindow];
>> +}
>
> Is this new call guaranteed to exist and work well on all the
> supported OS versions where we have the NS build?  I wouldn't want to
> fix this on some systems and break it on others at the same time.

makeKeyWindow is a very old API that should be available on every macOS
and GNUstep we support, AFAIK.  I see usages of that API in other parts
of the NS build, in things as central as frame creation, and they are
not protected by any version or system check.

>
> Alternatively, can we come up with a change that does both what the
> old code did and this addition?  That old code did work at some point,
> I presume?

This is not the case of the old code breaking at some point.  It's a
difference in behavior between the GNU/Linux version of Emacs and the NS
version of Emacs.  In GNU/Linux, if you place an Emacs frame in a
separate desktop and press C-x 5 0, the window manager does not
automatically switch to the desktop where the other Emacs frames reside.

I don't know the reason why the NS build has been calling raise-frame
for so long.  I presume it was considered a good enough way to refocus
to another frame after closing one, in a world where virtual desktops
were not very common and the behavior reported by the OP was ignored.

I've attached a revised version of the patch, to fix a couple of code
style issues.