Hi,
On 2023-03-21 15:26:38 -0400, Tom Lane wrote:
> Matthias van de Meent <boekewurm+postgres@gmail.com> writes:
> > ... with that patch we actually don't need the attcacheoff in the
> > pg_atttribute struct: it only needs to be present in the derived
> > "TupleAttrAlignData" structs which carry the
> > length/alignment/storage/byval info.
>
> Yeah, I was wondering about that too: keeping attcacheoff as local
> state in slots might get us all its win without so much conceptual
> dirtiness.
It's also the place where it's the least likely to help - afaict attcacheoff
is only really beneficial for fastgetattr(). Which conditions it's use more
strictly - not only can there not be any NULLs before the accessed column,
there may not be any NULLs in the tuple at all.
Greetings,
Andres Freund