Re: sit-for and idle timers

[Chong Yidong]

>> That is definitely not right.  Emacs really is idle when it reads an event,
>> even if there is a timeout.
>>
>> So my conclusion is that when read-event is called from an idle timer,
>> it should not change the state to idle at the beginning, and it should
>> not change the state away from idle at the end.
>
> I already checked in a different change to keyboard.c before your
> email came in.  But I think your version makes more sense.

I just thought of one inconsistency, though.  Suppose Emacs is not
idle, and is running Lisp code that calls sit-for.  With your change,
this starts idle timers:

!   if (! already_idle)
!     timer_start_idle ();

This is inconsistent with the old (built-in) behavior of sit-for,
which did not activate idle timers while waiting either.  The
reasoning, I think, is that `sit-for' means for the Lisp code to
"spin" for that period of time or until input arrives, which is not
the same as idling.  (This is similar to why `sleep-for' does not run
idle timers.)

If this is the intended behavior, then read-event with a timeout
should not start idle timers, since it only exists to be called by
sit-for.  We could simply document this behavior in the lispref
manual.