BUGS

* Server doesn't respond to subscribe request if no data currently available.
  Should instead fail connection.

    This one turns out to be a bit more tricky: in fact, a data gap *is*
    reported normally, though the timeout on the viewer was a bit too eager.
    The issue described only arises if the sniffer device can't be opened, this
    probably isn't worth fixing.


WORK IN PROGRESS

* Line analyser.  Needs to work closely with fast feedback excitation, probably
  doesn't want to be a separate IOC anymore.


IDEAS

* Experiment with SSE intrinsics, see if makes any difference to CIC speed.
    Unlikely to be worth doing.

* Compute decent filters for decimation, checking timing.
    The current 1/10 filter is ok.

* Guard against floating point exceptions.
    Current approach is to ignore them unless -E used to enable, and -E isn't
    safe against an overflow which can occur in the decimation filter (ringing
    on a very large step).  This option now removed.


FURTHER WORK

* The timestamp array returned by fa_load is dodgy at best.  It would be better
  to compute this from the knowledge available to us.  This will require server
  support.

* Want some mechanism for streaming the decimation variance information so that
  we can compute long term beam variance statistics.

* Might be worth trying to make fa-capture build under Windows.  This is a bit
  of a pain as Microsoft seem unwilling to implement C99 and of course there is
  no Posix, and make behaves differently.  Hopefully mingw will do the trick.

* Looks like falib(3) is incomplete.  Actually need to restructure falib a
  little, anyway.
