and have a similar need to Kohei for varlena objects greater than 1GB, in my case, also vector/matrix objects stored by the pggraphblas extension.
Thinking through my options, I was wondering if it would be a good idea to use large object support to store matrix backing data, perhaps first as a varlena object up to the limit, and then switch (in the flatten_into function for my types) to an object storing the oid of the large object holding the matrix data?
I'm going to experiment with this approach, but first wanted to raise the issue to shoot some holes in first before trying it.
