Matthew Woodcraft

Thanks. One possible complication I thought of is that the Diagnostics still wants to see a connection in order to call conn_text_from_chars(); I suppose that p...

OK.

I think now that the problem is that, because commit is called directly on the connection, there is no cursor associated with the exception. So I'm not sure...

I think diagnostics aren't working for errors which are raised from COMMIT. I suppose the PQresult is being cleared too early, but I'm not sure where.

This test...

I do have one other small worry: it's possible for a cursor to be re-used after an error, so in principle 'diag' from an existing Error object could end up repo...

Looks good to me. Perhaps it's worth saying in the docs that the PG_DIAG_*_NAME fields won't be available unless the server is PostgreSQL 9.3 or later? (Also, I...

I hope to do some more work on it tomorrow. But of course I won't have any complaints if you fix things yourself.

Ok, I've pushed it to a 2013-03-17_diagnostics branch in mattheww/psycopg2 on github:
https://github.com/mattheww/psycopg2/tree/2013-03-17_diagnostics

It's base...

I have something basically working now, using the design you described. If you like I'll publish what I've got so you can judge whether it looks suitable, and h...

Ok, that makes sense. I will have a go at writing it that way.

Last summer, when an earlier version of this Postgres patch was being considered, I had a look at adding support in Psycopg.

If you haven't started work on this...