On Wed, Jul 15, 2026 at 11:53:40AM -0400, Tom Lane wrote:
> I do not think this is a great idea. We don't use the tupdesc today
> perhaps, but it's plausible that it'd be needed in the future.
> Different table AMs might have different sets of system columns.
>
> Unless you can show a measurable performance gain from this change,
> we should leave it alone.
One thing that may make this removal worth it would be to look at if
heap_getsysattr() is used in some extension outside of the core code
and see if these removals lead to some simplifications with the
TupleDescs involved (say for example less locks required). The
in-core callers of heap_getsysattr() don't make that perspective
really attractive.
--
Michael
