PGCLIENTENCODING behavior of current CVS source

hi,
I'm using CVS source built postgres, may be one day later
then the main site, but found one problem:

I've set PGCLIENTENCODING environment before, for easy of
typing, like export PGCLIENTENCODING=GBK in my .profile,
but after I upgrade my postgresql to current CVS, I found
problem, the database initialized using:

initdb --locale=zh_CN.utf8 ...

the database connected is UNICODE encoded, but when I
use psql to loging to one of my database, it response:

psql: FATAL: invalid value for parameter "client_encoding": "GBK"

but when I remove the PGCLIENTENCODING setting:

unset PGCLIENTENCODING,

now I can login, but when I do a:

DHY_JJG=# \dt
ERROR: invalid byte sequence for encoding "UNICODE": 0xed

but, after:

DHY_JJG=# \encoding gbk
DHY_JJG=#\dt

woule be ok. the LANG setting is zh_CN.gbk, I guess it's
a localization problem. may be the encoding of thos po files.
because while using psql -E we see the query contain the
locale string in AS clause, but don't know the best way to
fix that, may be use UNICODE to encode those po files?

regards

Laser