Daniele Varrazzo

(ops)

was saying: register_array() seems a better option. We can have a chat on the ML about this functionality (it could take an oid, or none meaning all the o...

> Duncan Burke updated this ticket at February 24th, 2012 @ 10:29 AM

> In the meantime, I took a look at the type system, and it seems quite feasible to fix thi...

Nevermind, all done.

fixed register_inet() and added plenty of documentation about creating a generic typecaster.

I've looked into the problem. The issue has been mostly fixed in 2.4.2: it has gone mostly because SET ISOLATION LEVEL is not used anymore. However, a concurren...

Fixed in my dev branch.

rownumber is still broken for named cursors though. This would be an entirely different correction, and they have been broken forever, s...

I see: It's probably a regression introduced with this change, to make itersize work as expected.

https://github.com/dvarrazzo/psycopg/commit/8473209d240bad964...

Hi Duncan, thank you for the patch.

Unfortunately the behaviour you report is common to any unknown data type, for instance:

>>> cur.execute("select array['<hi...

Ticket put on hold: I don't want to add with support until the DB-SIG decides what should be its semantic. I'm not so worried about the cursor: more about the c...

Nobody has ever said it was.

Strange race condition. However, more recent psycopg versions do without "set transaction isolation level".

What you report is strange, because these versions u...