[Bucardo-general] bucardo5 kick hangs

[Wim Lewis]

The git head fails its tests for me. In each test, one of the "kick" commands hangs and eventually times out. It's always the same one in a given test. For example, from t/20-postgres.t:

> ok 190 - Row with pkey of type TEXT gets removed from C.bucardo_test4 after triple delete [line: 266] [time: 1324449942]
> ok 191 - Row with pkey of type TEXT gets removed from D.bucardo_test4 after triple delete [line: 269] [time: 1324449942]
> bucardo timed out:  kick pgtest 0
> # Looks like you planned 226 tests but ran 191.

I narrowed down the problem to commit 3828b42f6, which removed a call to $dbh->commit() from near the end of bc_deeply(). And in fact re-adding that call lets the tests continue. On the other hand, it doesn't seem to me that bc_deeply() should be doing a commit in the first place; and even more to the point, it seems wrong that an unclosed transaction in the test process should be able to make the bucardo command (which has its own independent database connection) hang indefinitely. But maybe that's a consequence of postgres' NOTIFY semantics, I don't know.