Re: BUG #8516: Calling VOLATILE from STABLE function

On Oct 11, 2013, at 9:21 AM, Dimitri Fontaine <dimitri(at)2ndQuadrant(dot)fr> wrote:

> Inter function dependencies is a hard topic indeed. I still would like
> to see some kind of progress being made someday. The general case is
> turing complete tho, because you can use EXECUTE against programatically
> generated SQL.
>
> You could even generate a CREATE FUNCTION command from within a PL
> function and EXECUTE it then call the created function… and I think I've
> seen people do that in the past.
>
> Still some kind of limited in scope static analysis for the cases where
> it's possible to do so would be great. With pg_depend tracking so that
> you know you're doing something wrong at DROP FUNCTION time.

I'm not very familiar with PostgreSQL internals, so I might be way off here, and I'm asking as much out of curiousity as anything else…

Would it be possible (and make sense) to solve this in a completely different way, not walking the function tree or doing static analysis, but simply setting and checking a bit during execution?

That is, when you execute a STABLE function, set a bit (clear it when function is done), and always check it when executing any VOLATILE function?

If a volatile function checks the bit, and you're "inside" a stable function, you could then raise an exception for calling volatile inside stable?

Terje