Ingo Müller

Just to keep track of this somewhere, and because it may be of interest for someone finding this issue: I learned about two related resources: * https://substrait-function-compatibility.streamlit.app/: A table with...

> For most plans there is only one PlanRel which contains a root. That represents an entire plan. > > The reason you can have multiple PlanRels I believe is...

> I think that it is safe to inline `Rel` into a `RelRoot` or to discard a `Rel` that is not referenced by some `RelRoot` (not even transitively). We can...

Borderline off-topic: I am thinking of an analogy to [symbol visibility in C/C++](https://gcc.gnu.org/wiki/Visibility): symbols can be `public`, in which case the compiler must put the symbol into the resulting binary,...

Let me add two data points to the discussion: * DuckDB only accepts plans whose first relation is a `RelRoot` (and doesn't even properly handle the case where the first...

OK, I see how recursive CTEs are a reason to support cyclic references. If/when that should be supported, the documentation has to be changed, though: it currently says that the...

OK, sounds good! For me, knowing that cycles can exist is good for. (#612 has a higher priority for me, personally.)

Small addendum: I have checked several SQL dialects and all of them only allow self-references, not general cycles. The structure of recursive CTEs is usually a non-recursive base case and...

My question is answered: there can be cycles (at least self-references) and there is no restriction on `subtree_ordinal`. Unless people feel that we might want to change that, which I...