[postgis-devel] ST_Equals, immutable strict

[strk]

On Tue, Aug 31, 2010 at 07:15:38PM -0400, Paragon Corporation wrote:

> So if it concludes the content of your function is NON-STRICT, but you mark
> it as STRICT -- it can't blindly inline it.  Since the inlined version could
> return a different answer from your intended.

Right, so the problem lies in "concluding the content of your function 
is NON-STRICT", which fails in this case.

Anarazel found that postgresql considers AND and OR inherently
non-strict [1], And that this is due to :

	SELECT NULL AND false; -- giving false

But such construct really is only non-strict if any operand
in the AND is non-strict or is not using the argument.
In our case we have:

SELECT $1 && $2 AND _ST_Intersects($1,$2);
       ^^^^^^^^     ^^^^^^^^^^^^^^^^^^^^^

They are both strict (the function is, the function associated
with the operator is), so for a NULL $1 or $2 the body would
resolve in:

	NULL AND NULL

I think either the planner should do a better job OR (better)
let the function definer signal inlinability with another
switch (INLINE?)

[1] http://trac.osgeo.org/postgis/ticket/536

--strk;

  ()   Free GIS & Flash consultant/developer
  /\   http://strk.keybit.net/services.html