Re: Avoid a possible out-of-bounds access (src/backend/optimizer/util/relnode.c)

On Tue, Sep 26, 2023 at 3:32 PM David Rowley <dgrowleyml@gmail.com> wrote:
>
> find_base_rel() could be made more robust for free by just casting the
> relid and simple_rel_array_size to uint32 while checking that relid <
> root->simple_rel_array_size.  The 0th element should be NULL anyway,
> so "if (rel)" should let relid==0 calls through and allow that to
> ERROR still. I see that just changes a "jle" to "jnb" vs adding an
> additional jump for Ranier's version. [1]

That's a good suggestion.

I am fine with find_base_rel() as it is today as well. But
future-proofing it seems to be fine too.

>
> It seems worth not making find_base_rel() more expensive than it is
> today as commonly we just reference root->simple_rel_array[n] directly
> anyway because it's cheaper. It would be nice if we didn't add further
> overhead to find_base_rel() as this would make the case for using
> PlannerInfo.simple_rel_array directly even stronger.

I am curious, is the overhead in find_base_rel() impacting overall performance?

--
Best Wishes,
Ashutosh Bapat