>>
 >> Do you get an assertion failure already before the C-x 5 2?
 >
 > No.

Thanks.  Did you ever get an assertion failure for the original frame?

If not, then we should try to understand why not.  Please with pristine
master replace the initial assignment of 'frame-size-history' in frame.c
from

   frame_size_history = Qnil;

to

   frame_size_history = Fcons (make_fixnum (100), Qnil);

start Emacs -Q and tell me the contents of the *frame-size-history*
buffer.

Can you try not using any scaling experimentally?  Jan always complained
that GTK does not handle scaling orderly (that is, in our sense).