Re: Append's first_partial_plan

On 18 April 2018 at 07:52, Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> wrote:
> While looking at this patch I became curious as to why do we even have
> first_partial_plan in the first place; it seems to require some strange
> contortions in the code. Wouldn't it be simpler to have two lists, one
> for non-partial and another for partial paths? I went to check the
> original discussion, and this design was indeed considered [1] -- but
> the idea was discarded because using the list index would lead to
> simpler code. However, now that we have pruning it seems to me that
> using the index isn't simpler anymore. Should we revisit this now?
>
> [1] /message-id/CA%2BTgmoZrjAB0bPzbtKgjP2uAP_6XAyuZenU54QuM7XGE_k2Q1g%40mail.gmail.com

I don't think having two Lists and/or two AppendState arrays would
make the pruning code anymore simple. All the pruning code in
execPartition.c would need to determine the index within the partial
or non-partial subnode array, and also communicate which array it
means. That code did take me a while to get right and be readable
too, I don't really want to have to change it again. I really don't
think it would look quite as simple as it does today either, so -1
from me for changing this.

--
David Rowley http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services