Andrew Dunstan <andrew@dunslane.net> writes:
> On 03/28/2011 11:14 AM, Tom Lane wrote:
>>> I think the most straightforward and reliable fix for this would be to
>>> forbid recursive containment of a rowtype in itself --- ie, the first
>>> ALTER should have been rejected. Can anyone think of a situation where
>>> it would be sane to allow such a thing?
> I think we should forbid it for now. If someone comes up with a) a good
> way to make it works and b) a good use case, we can look at it then. I
> expect the PostgreSQL type system to be a good deal more constrained
> than a general in-memory programming language type system. If lack of
> working type recursion were a serious problem surely we'd have seen more
> squawks about this by now.
The immediate issue in CheckAttributeType() could be fixed by tracking
which types it was processing and not recursing into an already-open
type. Which, not at all coincidentally, is 90% the same code it'll need
to have to throw error. The issue for really "making it work" is how do
we know if there are any other places that need a recursion defense?
I'm pretty sure that find_composite_type_dependencies would, and I don't
know where else there might be a hidden assumption that column
references don't loop. So I think that it's mostly about testing rather
than anything else. If I were fairly confident that I knew where all
the risk spots were, I'd just fix them rather than trying to forbid the
construction.
regards, tom lane