Re: Calling for a replacement committer for GROUPING SETS

>>>>> "Greg" == Greg Stark <stark(at)mit(dot)edu> writes:

Greg> Has there been anything controversial?

The major controversy is the idea of processing multiple sort orderings
using a stacked chain of aggregate and sort nodes. I'll follow up
shortly with links or a summary of the most significant criticisms; but
note that the patch has already been fixed to keep the memory usage
bounded regardless of the depth of the chain, so the issue is more about
code and design aesthetics than about actually working.

There is a description of the current approach in the block comment at
the start of executor/nodeAgg.c.

A less major issue is whether the introduction of GroupedVar was a good
idea; I'm happy with ripping that out again, but if so the code needed
to replace it depends on the resolution of the above point, so I've left
it as originally submitted pending a decision on that.

Greg> What was causing it to take so long.

Tom claimed the reviewer/committer spot for himself five or six months
ago, and other than the inadequate review and subsequent discussion in
December there has been no progress. We posted the latest patch shortly
after that (there was some slight delay thanks to the holiday season).

Greg> I have time to work on it now

Great. I'm actually about to post the latest patch with some tiny
updates to match the recent changes in MemoryContextReset, but those are
as much cosmetic as anything else (just removing no-longer-needed calls
to MemoryContextDeleteChildren).

--
Andrew (irc:RhodiumToad)