On Mon, Jan 17, 2011 at 1:53 PM, Federico Di Gregorio
<federico.digregorio@dndg.it> wrote:
> On 17/01/2011 14:35, Daniele Varrazzo wrote:
>> The problem is that postgres 9 uses by default the hex format instead
>> of the escape format to represent bytea
>> (http://www.postgresql.org/docs/9.0/static/datatype-binary.html). A
>> workaround I was thinking about could be to send "SET bytea_output TO
>> 'escape';" on connection in case of pre9-libpq/post9-server
>> combination is detected. I hate it because I was trying to drop the
>> "queries on connection" psycopg does (there used to be 3, now there is
>> only one and I think the last one can be dropped as well).
>
> Please lets not do that. psycopg can already be used with a lesser libpq
> to connect to a newer backend _unless_ changes like the one in this
> discussion. Solving this particular problem may seem a good thing but,
> IMHO, it is just putting a workaround in the wrong place.
>
>> Of course
>> bytea_output can be set in the postgresql.conf too and it would make
>> previous releases of psycopg work fine.
>
> That's the correct workaround.
A different workaround would be to have our own bytea hex format
parser if the libpq doesn't support it. It looks easy enough and
wouldn't be a security risk as it's only to receive data from the
backend: 9.0 servers understands escape format ok with both settings.
My fear is that not everybody would be able to set bytea_output to
escape in the server they have to talk to, and it looks 8.4 on the
clients is there to stay for a long while (referring to ubuntu, the
distro I'm more up-to-date, both the last LTS and the last 10.10 have
only 8.4 in the official repos, and things don't look different in the
upcoming natty - see
<http://packages.ubuntu.com/natty/postgresql-client>). If we decide
not to do anything we should at least document the toxic interaction
in a FAQ.
-- Daniele