Re: PgStat_HashKey padding issue when passed by reference

Em qua., 10 de set. de 2025 às 23:53, Michael Paquier <michael@paquier.xyz>
escreveu:

> On Mon, Sep 08, 2025 at 09:36:52PM -0500, Sami Imseih wrote:
> > But my concern is the flexibility of this approach. If someone is to add
> an
> > OID field next, they will not be able to as that will be introducing
> > padding.  On the other hand, passing the key by reference and
> > documenting the reason in pgstat_shmem.c will not lose this
> > flexibility.
>
> I don't mind discarding the static assertion idea, but at the end what
> counts for me here is that I don't want to sacrifice future changes in
> the pgstats code that would always require passing around the hash key
> by reference.



> So I would just do like attached, documenting at the
> top of PgStat_HashKey that we should not have padding in it, removing
> three memset(0) calls that expected it.
>
Currently no compiler guarantees that static initialization will fill
possible holes,
something that *memset* guarantees.
I think this change is unsafe.

best regards,
Ranier Vilela