Alexander Korotkov <aekorotkov@gmail.com> writes:
> On Mon, Oct 21, 2013 at 11:06 PM, Heikki Linnakangas <
> hlinnakangas@vmware.com> wrote:
>> I guess it can't happen. Or is it possible that a toasted value that came
>> from disk will be passed to these functions, without detoasting them
>> somewhere along the way? Not sure.
> I will ask Teodor about it. When situation will be completely clear we
> should cleanup confusing comments.
I believe the reason GIST has compress/decompress functions is not for
TOAST (they predate that, if memory serves), but to allow the on-disk
representation of an index entry to be different from the data type's
normal representation in other ways --- think lossy storage in particular.
It's possible that the code in cube's decompress/decompress functions is
unnecessary because cube doesn't actually do any such compression; the
code might've been copy-pasted from somewhere else without adequate
thought about whether it was useful for cubes. I'd want to see some
discussion about why it's okay to change it before we do so, though.
In any case it seems separate from the claimed purpose of this patch
and thus better done in a different patch.
regards, tom lane