On Tue, Aug 20, 2013 at 7:53 PM, Heikki Linnakangas
<hlinnakangas@vmware.com> wrote:
> I've been browsing the code to figure out how we could use protocol-level
> cursors, instead of DECLARE/FETCH, to make the default behavior of the
> driver nicer. However, all the different ways a query can be prepared and
> executed is making my head explode. The driver supports many completely
> obsoleted ways to do things, for the sake of supporting 7.3 era servers
> which didn't support the extended query protocol. There is even some code to
> support 6.2-era servers.
>
> I'd like to rip out all the support for pre-7.4 servers, including protocol
> version 2, to simplify the code. Any objections? If someone is running a
> really old server version, I think it's fair to insist that they also use a
> somewhat old version of the driver. (the only thing anyone should be doing
> with 7.3 servers today is pg_dump, to be honest)
>
> I pushed a patch to github to do that, to see what the effect is to the
> code:
> https://github.com/hlinnaka/psqlodbc/commit/d5cbe6e62a72d87200217b73360a2c515a2fec79
+1. It looks like a good plan to refresh this code tree by removing
old protocols, and just by seeing the first patch you sent, code is
simplified in many places.
> That's just the beginning though. I think we can get rid of much of the
> "premature execution" stuff, replacing it with the v3 protocol-level
> Describe functionality. The driver already uses that in many cases, but not
> all. That requires some more investigation and thorough testing of all the
> corner cases, though.
Regression tests and precise documentation would be a nice addition as
well when doing this implementation.
> Timingwise, I'm thinking we would do this in the first 9.4 version of the
> driver. The server guys will put out the 9.3 release in the next couple of
> weeks, so we should put out a stable 9.3 version of the ODBC driver very
> soon. But after that, we can do more drastic changes again.
Agreed.
--
Michael
