Character collating order

I'm having problems with the regression tests. In particular, the CHAR test fails because of an unexpected character collating sequence:

test=> select 'a' < 'A' AS "Lowercase is low", 'a' > 'A' AS "Uppercase is low";
Lowercase is low | Uppercase is low
------------------+------------------
t | f
(1 row)

The test scripts seems to expect that 'A' < 'a', which is not a completely unreasonable assumption :-)

I'm using the SQL_ASCII encoding, and un-setting the LANG environment variable (which is the only locale variable set for the account; it defaults to en_US) for postgres when starting postmaster does not fix it.

Any suggestions on what I could try?

Allan.

bash-2.04$ rpm -qi postgresql
Name : postgresql Relocations: (not relocateable)
Version : 7.1.2 Vendor: (none)
Release : 4PGDG Build Date: Wed 13 Jun 2001 10:18:46 PM BST
Install date: Fri 29 Jun 2001 08:43:51 PM BST Build Host: lowen.wgcr.org
Group : Applications/Databases Source RPM: postgresql-7.1.2-4PGDG.src.rpm
Size : 4426307 License: BSD
URL : http://www.postgresql.org/
Summary : PostgreSQL client programs and libraries.
[...]
bash-2.04$ uname -a
Linux XXXX 2.4.2-2 #17 SMP Tue Jul 3 14:48:45 BST 2001 i686 unknown