Update pq3.py

[jcborras]

Prevents the behaviour in #72 from arising.

[jwp]

I'm disinclined to accept this on grounds that Amazon is out of line. Additional Startup parameters are properly handled by every other server that I know of (pgpool and pgbouncer).

[jcborras]

I respect that. Notice however that setting the param to None does not void its setting.

[jwp]

Yeah, I should probably send out an e-mail to the list. I think patching is semi-reasonable.

[chriscannon]

Why was this not merged? It was such a simple solution to the problem.

[jwp]

Got this confused with a different patch. I already commented on why I didn't merge it.

[chriscannon]

I read your rationale, but it doesn't make sense. As a user of the library, I should be able to unset any options that are passed to the Postgres database. It's a two-line change that makes the library more robust. It has nothing to do with Amazon.

[jwp]

The patch as is doesn't accommodate for empty strings which will trigger the condition. He needs to use an is None check. There are no included unit tests exercising the new functionality. My perception is that it is beyond unreasonable for Amazon to disallow the passing of the client_min_messages as part of the Startup message. It almost feels like they wrote it broken on purpose.