Add SUM(BOOL) overload

[Mytherin]

This is a useful overload that allows the user to quickly count the amount of matches for a boolean condition, it is essentially syntactic sugar for:

SELECT SUM(CASE WHEN [cond] THEN 1 END) FROM tbl

As an added bonus, it is also slightly faster than the above expression:

SELECT SUM(l_extendedprice > 500) FROM lineitem;
-- 0.020s
SELECT SUM(CASE WHEN l_extendedprice > 500 THEN 1 END) FROM lineitem;
-- 0.028s

[carlopi]

I guess this becomes equivalent to:

SELECT count(l_extendedprice > 500) FROM lineitem;

but nice to provide the overload.

[Mytherin]

No, COUNT counts the number of non-NULL values, so count(l_extendedprice > 500) is equivalent to count(l_extendedprice) (which is equivalent to count(*) as lineitem does not have NULL values).

[carlopi]

Thanks, again

[soerenwolfers]

Isn't

SELECT count() FILTER (l_extendedprice > 500) FROM lineitem;

the canonical way of doing this? It's also faster than the CASE WHEN expression for me, and while more verbose has the slight advantage that there is no chance to footgun oneself into thinking the aggregation happens in the boolean algebra, i.e., modulo 2. (That being said, I did write SUM(BOOL) in the expectation it act like in this PR before and was sad it didn't exist, so I'd welcome its addition).

[Mytherin]

That's another alternative yeah, but FILTER is supported in much fewer systems. It is around as fast as the CASE statement for me (i.e. still slightly slower than the SUM(BOOL) provided in this PR, but the difference is negligible since they are both fast).

[soerenwolfers]

By the way, the macro count_if could now be rewritten to be SUM(x::BOOL) see duckdb/duckdb-web#3294

[Mytherin]

Opened a PR for that here - #15061