Re: PG15 beta1 sort performance regression due to Generation context change

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> I don't want to take the position that we ought necessarily to commit
> your patch, because I don't really have a clear sense of what the wins
> and losses actually are.

Yeah, we don't have any hard data here. It could be that it's a win to
switch to a rule that chunks must present an offset (instead of a pointer)
back to a block header, which'd then be required to contain a link to the
actual context, meaning that every context has to do something like what
I proposed for generation.c. But nobody's coded that up let alone done
any testing on it, and I feel like it's too late in the v15 cycle for
changes as invasive as that. Quite aside from that change in itself,
you wouldn't get any actual space savings in aset.c contexts unless
you did something with the chunk-size field too, and that seems a lot
messier.

Right now my vote would be to leave things as they stand for v15 ---
the performance loss that started this thread occurs in a narrow
enough set of circumstances that I don't feel too much angst about
it being the price of winning in most other circumstances. We can
investigate these options at leisure for v16 or later.

regards, tom lane