Sometimes, but not always, the pkd tests will fail because they
close the socket at hand a bit too early for the client. The
client in turn may exit non-zero when that happens.
Split up the final close loop so that pkd waits to receive a
channel close from the client, and then socket close, before
finally returning.
With this change I observe that tests are now passing in
environments that would previously tickle the above race
and fail.
Signed-off-by: Jon Simons <jon@jonsimons.org>
Reviewed-by: Andreas Schneider <asn@cryptomilk.org>
8d7563a9c3