Hi,
On 2021-09-15 16:51:37 -0400, Tom Lane wrote:
> The other problem here is that a simple call-this-instead-of-that
> top-level hook doesn't seem all that useful anyway, because it leaves
> you with the task of duplicating a huge amount of functionality that
> you're then going to make some tweaks within. That's already an issue
> when you're just thinking about the grammar, and if you have to buy
> into it for parse analysis too, I doubt that it's going to be very
> practical. If, say, you'd like to support some weird function that
> requires special parsing and analysis rules, I don't see how you get
> that out of this without first duplicating a very large fraction of
> src/backend/parser/.
We do have a small amount of infrastructure around this - the hackery that
plpgsql uses. That's not going to help you with everything, but I think it
should be be enough to recognize e.g. additional top-level
statements. Obviously not enough to intercept parsing deeper into a statement,
but at least something.
And parse-analysis for some types of things will be doable with the current
infrastructure, by e.g. handling the new top-level statement in the hook, and
then passing the buck to the normal parse analysis for e.g. expressions in
that.
Obviously not going to get you that far...
> (As a comparison point, we do have a top-level hook for replacing
> the planner; but I have never heard of anyone actually doing so.
> There are people using that hook to *wrap* the planner with some
> before-and-after processing, which is quite a different thing.)
Citus IIRC has some paths that do not end up calling into the standard
planner, but only for a few simplistic cases.
> I don't have any better ideas to offer :-( ... but I very much fear
> that the approach proposed here is a dead end.
I unfortunately don't see a good way forward without changing the way we do
parsing on a more fundamental level :(.
Greetings,
Andres Freund