[evlatests] Bug in executor

[Ken Sowinski]

On Wed, 22 May 2019, Barry Clark wrote:

> There is a race condition in Executor which primarily affects scans
> or subscans of length five seconds.  It can occur in scans of length
> between five and ten seconds.  It should not occur for scans shorter
> than five seconds or longer than ten seconds.  If the race is won by
> the wrong thread, extra pointing, and for VLA, delay commands are sent,
> whose effects range from harmless to wrong pointing/delay for the entire
> scan.
>
> In the simulator, the race went to the wrong thread about 15% of the
> time with five second scans and about 1.5% of the time with seven
> second scans (your mileage may vary).
>
> I am currently testing a fix in the simulator, will commit it tomorrow.

I examined a few weeks worth of delay packets addressed to one
antenna looking for successive packets with identical "activation" times.
There were three clasees of scripts which suffered from this:

1.  Test scripts created by Vivek.
2.  The Sysstart? scripts.
3.  Any VLBI phased array script.

The last of these has lots of short scans for phasing.  No examples
of duplicated activation times was seen for any ordinary OPT/m2s
generated scripts.