Vladislav Vaintroub wrote:
> This seems to retain the original sorting when keys for memcmp based
> comparison and making NULL less then everything else.
> I think this fixed lot of issues. But, if we go for it, we need to issue a
> big red "warning, incompatible change" in the release notes
Err, I don't see the need for any big read warnings, let alone marking
tablespaces so they continue the old wrong behavior.
Perhaps I've missed something important, but the old way got the wrong
answer, right? And the new way gets the right answer, yes? And if you
use the new test on the old indexes, you still get the wrong answer,
but it doesn't crash. New entries using the new algorithm in an old
index will continue to get the wrong answer for old entries but will
get the right answer for new entries, right? Failing to get the old
wrong answer with the new code doesn't seem like a serious compatibility
issue to me.
That sounds like a bug fix to me. I'd document that if you have
compound indexes that allow nulls in leading segments, you should
rebuild those indexes.
That seems fine for alpha.