Testing

[Hal Murray]

We should start collecting a list of manual tests to run in preparation for a 
release or after a major change.

One obvious one is to make sure that all of the refclocks work.  Some of them 
have several modes and options so this can get complicated.

Another one from a recent discussion.  We should deliberately introduce an 
error (using bumb in ntpfrob) and verify that the response is sane.  We 
should probably do this on as many OSes as we can.  And starting at different 
polling intervals.  ...

There are two or three regimes.  For a small offset, ntpd will slew the 
clock.  For a larger offset, it will step the clock.  For a huge offset it 
will panic and let a human sort things out.

Just going through that sort of testing has the benefit of getting people to 
look somewhat carefully at graphs

-- 
These are my opinions.  I hate spam.