Rushabh Lathia <rushabh(dot)lathia(at)gmail(dot)com> writes:
> With the commit mentioned in the $subject, I am seeing the
> change in behaviour with the varlena header size.
Interesting. AFAICS, the new behavior is correct and the old is wrong.
SET STORAGE PLAIN is supposed to disable use of TOAST features, including
short varlena headers. So now that's being honored by the UPDATE, but
before it was not. I have no idea exactly why that changed though ---
I'd expect that to be implemented in low-level tuple-construction logic
that the planner rewrite wouldn't have changed.
regards, tom lane