RE: mouse-autoselect-window with menu pane

[Marshall, Simon]

>       A very quick simple
>     test shows the behaviour seems consistent if the function run by the
timer
>     cancels the timer, when the function is run later due to buffering up
of
>     timers, in the scenario you describe.
> 
> I can't understand those words at all.

I think Martin was asking what would happen if the function run by a
repeating timer cancelled the same timer, in the case where the repeating
timer is delayed more than REPEAT*2 seconds due to a popup or something.  I
was trying to explain that it does not result in the timer being triggered
more (or less) than the case where the timer that runs the function is not
delayed; the timer cancellation behaviour is consistent between the two
cases.

My take on the current timer behaviour is that if you ask it to repeat every
R seconds, then the timer will typically run N times in the period N*R, if N
is large enough.  However, Emacs cannot guarantee that the delay between
timer n and timer n+1 is always at least R seconds (though it can guarantee
that timer n will not run before n*R seconds).  An alternative behaviour
(which I think Martin was alluding to) is to not guarantee that the timer
will run N times in the period N*R, but instead guarantee that the delay
between timer n and timer n+1 is always at least R seconds.

I'm not arguing that the current behaviour is wrong, but I could argue that
the name REPEAT for the delay arg implies the latter behaviour is
implemented.