Robert Pluim <rpluim@gmail.com> writes:

> Dmitry Gutov <dgutov@yandex.ru> writes:
>
>> That is the description of my experience with Robert's patch
>> applied. Without it applied, the "bad frame" appears last (at position
>> 40-100, approximately).
>>
>> Feng Shu has changed his code since to call set-frame-position much
>> less often, and it seems to work correctly *without* Robert's
>> patch. Quite weird.
>
> Yes. I have an improved version of the patch on the way, which should
> also fix `frameset-restore' and a few other cases of incorrect frame
> positioning. Hopefully by the end of the week.

frameset-restore turned out to be a completely separate fix.

The attached fixes the cases that I could track down where we weren't
applying scaling properly. All Emacs pixel coordinates remain unscaled
after this, the values are just scaled when passing them into GTK. The
xwidget code almost certainly requires the same type of adjustments.

Strongly recommended for emacs-26. People who don't use scaling will
see no side-effects.

Robert