Andres Freund <andres(at)2ndquadrant(dot)com> writes:
> I'd really not even be surprised if a committer backpatches a
> MemoryContextReset() addition, not realizing it behaves differently in
> the back branches.
As far as that goes, the only consequence would be a possible memory leak
in the back branches; it would not be a really fatal problem. I'd rather
live with that risk than with the sort of intentional API break (and
ensuing back-patch pain) that you're proposing.
regards, tom lane