Re: Bidirectional hard joins (fwd)

Oleg Bartunov <oleg@sai.msu.su> writes:
> Could you, please, comment the proposal.

Okay: "ugly and unimplementable".

Where are you going to put these back-references that the description
glosses over so quickly?  They can't be in the row itself; that doesn't
scale to large numbers of references to the same row.  I think you'd end
up building an external datastructure that would in the final analysis
offer no better performance than standard indexes.

I'd also want to see an analysis of how this interacts with MVCC before
we could consider whether it makes any sense in Postgres.  In
particular, which version of a row does the reference point at, and how
will concurrent updates (possibly aborted) be handled?
        regards, tom lane