Actually, the False Lead kind of led to something in the end. When pg8000 established a connection, messages end in either a READY_FOR_QUERY or a ERROR_RESPONSE, whereas in the query phase, messages only ever end with a READY_FOR_QUERY. So I've separated out the two bits of code, and each is simpler than the combined one. Hopefully, this will help in tracking down the long-standing bugs (or even solve them!).

In other news, there's been a request that pg8000 be packaged in the .whl format on the cheese shop. I did a bit of reading, and it seems like a really good idea. So that's my next task...