Hi Simon.

 > Hi Martin, I occasionally get this (taken from *Messages*):
 >
 > mouse-autoselect-window-start: Wrong type argument: numberp, nil
 >
 > I just caught it with debug-on-error, though I'm not sure quite how to read
 > it as it looks like 2 errors have occurred.  The way I read it, we are in
 > the process of switching to the *Backtrace* window when the error
 > (presumably) happened again:
 >
 > Debugger entered--Lisp error: (wrong-type-argument numberp nil)
 >   window-at(nil nil #<frame test2/CONTRACT.in 0xd1a200>)
 >   mouse-autoselect-window-start()
 >   handle-select-window((select-window (#<window 47 on *Backtrace*>)))
 >   call-interactively(handle-select-window)
 >
 > But it is kind-of what I guessed, that the error is thrown by window-at
 > because mouse-position returns nil for X and Y.
 >
 > I can fairly easily reproduce this with emacs -Q if I move from the selected
 > first frame to a second frame and then off that second frame quickly when
 > (a) the second frame has split windows, (b) the second frame's selected
 > window (ie, when the second frame is selected) and its point is covered by
 > the first frame.

I can obtain the error too, even when running two distinct Emacs
processes with one frame each.  Obviously, `mouse-position' _should_
return nil whenever it's not able to determine a meaningful position,
hence I can't put the blame on that.

 > The amount of time spent in the second frame does not seem to depend on
 > mouse-autoselect-window.

I suppose you tell that from using an autoselect timeout in your window
manager that differs from `mouse-autoselect-window'.

 > Maybe it has more to do with whether Emacs gets
 > the chance to get some info out of X before the mouse has moved off the
 > second frame.  I use Solaris CDE on a PC running Exceed, which may
 > complicate/slow things.

Honestly, I expected many more problems when switching frames than you
encountered so far.

 > Maybe the only fix is to be tolerant of nil for X and Y from mouse-position.

The only thing I can offer is to wrap this in a `condition-case' and
have `handle-select-window' do it's usual stuff when
`mouse-autoselect-window-start' fails.

martin