Gregory Stark <stark(at)enterprisedb(dot)com> writes:
> The key to the algorithm is that it uses a trie to bin rows with common
> leading prefixes together. This avoids performing redundant comparisons
> between those columns later.
Interesting, but doesn't that make it utterly useless for sorting in
non-C locales?
I'm not that thrilled with introducing datatype-specific paths into the
sort code anyway; seems like a maintenance nightmare.
regards, tom lane