On tor, 2011-03-10 at 18:12 -0500, Tom Lane wrote:
> What I'm inclined to do about this is set "default"'s collencoding to
> -1, with the semantics of "works for any encoding", and fix the lookup
> routines to try -1 if they don't get a match with the database
> encoding. Having done that, we could also use -1 for "C" and "POSIX",
> thus avoiding having to make a bunch of duplicate entries for them.
Good idea.
> BTW, I would like to eventually have "C" and "POSIX" in there all the
> time (ie created by pg_collation.h), so that they can be used even in
> machines that don't have locale_t support. I haven't yet gotten
> around to reading the parts of the collation patch that might need to
> change to support this, so I'm not sure how much work it'd be. But
> I'd say that being able to do COLLATE "C" in an otherwise non-C
> database would cover a very large fraction of the user requests I've
> read about this, so being able to handle that case even without
> locale_t support would be really useful IMO.
That should actually already work. The relevant logic is in
varstr_cmp().
But good point. We should support this out of the box on all platforms.