>> what's going on here, IIUC it takes some time (at least a couple of
 >> seconds) for a frame to show up on Vframe_list.
 >
 > It should not matter at all, because x_create_frame cannot make the new
 > frame known to lisp before it is added to Vframe_alist.

I suppose you mean Vframe_list.  At least on my system I can see frames
that are not returned by `frame-list'.

 >> (1) Check whether the frame passed to next_frame is in Vframe_list and
 >>     quit if it isn't.
 >>
 >> (2) Put a maximum of 100 frames investigated on the loop in next_frame
 >>     so there should be no endless looping otherwise.
 >
 > This is just doctoring the symptoms.

What did you expect me to do?  I first have to find out whether the
looping really is in next_frame.  Or do you have a definitive clue?

BTW my doctoring patch has an obvious bug which should be corrected in
the new attachment.

 > There is an invariant that *every*
 > live frame is on Vframe_alist.  If that invariant is violated then this
 > is the bug that must be fixed.

Such an invariant seems obvious but I don't see it neither formulated
nor preserved.  And why is Vframe_alist "V" prefixed but not available
in Lisp?

martin