Re: Psycopg 2.5 released

On Tue, Apr 9, 2013 at 7:59 AM, Michael van Tellingen
<michaelvantellingen@gmail.com> wrote:

> My goal when porting psycopg2 was both as an exercise as well as seeing how
> fast it could be with pypy.
> In the end the performance was okayish, for some workloads it was quite fast
> but I never actually used it in production. The cffi port is in that aspect
> much more promising from what I've read. Never tried it myself though.
>
> Regarding further updates to psycopg2-ctypes I'm afraid I just can't find
> the time for it. So if anyone is interested in taking over the development
> the please let me know!

Michael,

your past work not only has been greatly appreciated, but it's still
the base of the current ports. We are everybody fork of your repos :)

The summary of the story is: Konstantin Lopuhin has ported Michael's
psycopg2ct to cffi (https://github.com/chtd/psycopg2cffi). I've
learned from him that they are using it in production on PyPy and he's
actively maintaining it. I've also learned from Antonio Cuni that cffi
is *way* more appreciated than ctypes in PyPy world (for type safety
and jittability) so I'm convinced cffi is the way to go.

I've forked Kostia's repos
(https://github.com/dvarrazzo/psycopg2-ctypes/tree/cffi-2.5) and I'm
hacking on it to bring it at psycopg 2.5 feature level. There are only
4 tests not passing in the test suite now :)

My plan is to finish the porting, have some serious test with valgrind
and the databases grid on buildbot, after which I think we can bless
the -cffi implementation. I'm not planning to work on the -ctypes
branch anymore... but I'm not sure I want to drop entirely the idea of
a psycopg on ctypes. A pure python module only depending on the stdlib
is actually *very* appealing, but as it is now I'd derive it back from
the -cffi codebase, maybe with a compatibility layer. OTOH if cffi is
easy to install on Windows and OS X too there would be no reason for
it.

In the meantime a big thank for all your work to date.

Cheers,

-- Daniele