Re: "failed to commit client_encoding" explained

Magnus Hagander <magnus@hagander.net> writes:
> Tom Lane wrote:
>> Comments?

> Certainly seems like a reasonable compromise. From what I understand,
> you'll get this "failed to commit..." message *if* you have changedf
> things in pg_conversion. I think that's acceptable - it's not like
> people modify pg_conversion all the time (at least I hope they don't).

Yeah, that's a corner case on a corner case.

A further thought here is that even though the change I propose is
pretty localized, it's still complicated enough to possibly introduce
new bugs.  And it's fixing a case that I think doesn't occur in
practice; people don't really make local-in-transaction changes of
client_encoding.  (Remember the bug was only discovered by accident.)

So my inclination is to fix it now in HEAD so it can go through a
cycle of beta testing, but leave 8.3 alone.  We can back-patch it
after testing if we get any additional complaints.
        regards, tom lane