On tis, 2011-01-25 at 16:19 -0500, Tom Lane wrote:
> What would probably be reasonable to do is a one-time run over a lot
> of locales, and then collate the results to see how many distinct
> outputs we see and over what sets of locales. Then we could make some
> reasoned decisions about which cases are worth carrying variant
> expected files for.
We already did that and the current state is the result of that.
http://archives.postgresql.org/message-id/200901111254.03722.peter_e@gmx.net