It appears that the mingw faction of the buildfarm doesn't like this
patch:
http://archives.postgresql.org/pgsql-committers/2008-11/msg00289.php
The failures look like the regression tests are being run with
intervalstyle = iso (the default) rather than postgres_verbose
as they ought to be. Which means that there's something wrong
with the new code in pg_regress.c that tries to set that via
PGOPTIONS instead of via PGINTERVALSTYLE. But what, and why is
it only happening on those machines?
regards, tom lane