Re: Possible savepoint bug

I wrote:
> Michael Paesold <mpaesold(at)gmx(dot)at> writes:
>> I am seeing a similar unique index bug here...
>> This is PostgreSQL 8.1.1 on RHEL 3, Intel Xeon (i686).

> It looks like the problem is that index entries are being inserted out
> of order.

After further investigation, it seems that the original sort order of
the index was not C-locale, but something else --- I can reproduce the
current index ordering except for a small number of new-ish tuples if
I sort the data in en_US.

We go out of our way to prevent the backend's locale from changing after
initdb. Did you do something to override that?

Another theory is that this is a manifestation of the known problem with
plperl sometimes changing the backend's locale setting. Is it possible
that the index was created in a session that had previously run some
plperl functions?

regards, tom lane