I think the problem that steve is talking about is related to a pg bug where the coallating order is wrong for locales other than 'C'. This should get fixed at some point. However, it should be possible to dynamically generate the tree_encodings table so that the decimal code matches the collating order of the machine that it iss running on when the core is first loaded.
I'll see if I can fix the initialization of the tree_encodings table, so that the locale is not an issue.
