> However, I'm not sure I believe the comment anymore: it has
> not changed since Postgres95 and I can see that quite a bit of work has
> been done on the duplicate-key logic since then. Furthermore findsplitloc
> itself sometimes ignores the claimed requirement: when it does the
> split-in-the-middle case quoted above, it does not pay attention to
> whether it is splitting in the middle of a group of duplicates. (But
Ops. This is bug.
> that path is taken infrequently enough that it's possible it's just
> plain broken, and we haven't noticed.)
>
> Does anyone know whether this comment still describes the btree
> equal-key logic accurately?
I think so.
Vadim