Re: Rangejoin rebased

On Sat, Jan 6, 2018 at 10:38 AM, Simon Riggs <simon(at)2ndquadrant(dot)com> wrote:
> For this to be useful, it needs to include some details of how to use
> it when people have NOT used range datatypes in their tables.

Good idea. I added an example that doesn't have range types in the base table.

> If we can see some examples with StartDate and EndDate cast to a
> tzrange, plus some "don't write it like this" anti-patterns that would
> help.

By "anti-patterns", I assume you mean implementing range intersection
by hand in SQL over non-range types. Such an example would be quite a
long query, and might add to the confusion -- it seems strange to
spend more text explaining what *not* to do than what to do.

I see what you are saying, but perhaps I'm just not coming up with the
right text to explain it succinctly in the manual, so for now I just
added one example. Let me know if you think it needs more.

> We can then make it clear that this is a huge performance benefit for
> these important cases.

Done.

> Just to emphasise why we want this, it might be better for the EXPLAIN
> to say "Time Range Join" when the ranges being joined are Time Ranges,
> and for other cases to just say "Range Join". The use of the word
> Merge doesn't help much there.

I don't care for special-casing the word "time" in there, because no
other type would benefit. It seems a little too magical. I also do
like leaving "merge" in there because it helps the user understand why
their inputs are being sorted.

Regards,
Jeff Davis