Re: [HACKERS] path toward faster partition pruning

Alvaro Herrera wrote:

> Yeah.  Looking at this function, I noticed it tests for BooleanTest, and
> falls back to checking "not_clause" and a few equals.  Does it make
> sense if the clause is a SAOP?  I added this assert:
>     Assert(IsA(clause, BooleanTest) ||
>            IsA(clause, BoolExpr) ||
>            IsA(clause, RelabelType));
> 
> and it failed:
> #3  0x0000556cf04505db in match_boolean_partition_clause (partopfamily=424, 
>     clause=0x556cf1041670, partkey=0x556cf1042218, rightop=0x7ffe520ec068)
>     at /pgsql/source/master/src/backend/optimizer/util/partprune.c:2159
> 2159        Assert(IsA(clause, BooleanTest) ||
> (gdb) print *clause
> $1 = {type = T_ScalarArrayOpExpr}
> 
> I'm not sure whether or not this function can trust that what's incoming
> must absolutely be only those node types.

So this is what I need for current regression tests not to crash
anymore:

    Assert(IsA(clause, BooleanTest) ||
           IsA(clause, BoolExpr) ||
           IsA(clause, RelabelType) ||
           IsA(clause, ScalarArrayOpExpr) ||
           IsA(clause, OpExpr) ||
           IsA(clause, Var));

I'm not confident in my ability to write code to handle all possible
cases right now (obviously there must be more cases that are not covered
by current regression tests), so I'll leave it without the assert since
it handles a couple of the useful cases, but I suspect it could stand
some more improvement.

I guess the question is, how interesting is boolean partitioning?  I bet
it has its uses.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services