Re: Schema variables - new implementation for Postgres 15 (typo)

> It can be used only
> from extensions, where is ensured so a) the value is not changed, b) and in
> a lifetime of returned value is not called any query or any expression that
> can change the value of this variable. I fixed this code and I enhanced
> comments. I am not sure if this function should not be removed. It is not
> used by backend, but it can be handy for extensions - it reduces possible
> useless copy.

Hmm, how safe is it for third-party code to access the stored data directly
rather than a copy?  If it makes extension fragile if they're not careful
enough with cache invalidation, or even give them a way to mess up with the
data directly, it's probably not a good idea to provide such an API.

> > I'm also wondering if there should be additional tests based on the last
> > scenario reported by Dmitry? (I don't see any new or similar test, but I
> > may
> > have missed it).
> >
>
> The scenario reported by Dmitry is in tests already.

Oh, so I missed it sorry about that.  I did some testing using
debug_discard_cache in the past and didn't run into this issue, so it's
probably due to a more recent changes or before such a test was added.

> I am not sure if I
> have to repeat it with active debug_discard_cache. I expect this mode will
> be activated in some testing environments.

Yes, some buildfarm animal are configured to run with various
debug_discard_caches setting so it's not needed to override it in some tests
(it makes testing time really slow, which will be annoying for everyone
including old/slow buildfarm animals).

> I have no idea how to simply emulate this issue without
> debug_discard_caches on 1. It is necessary to raise the sinval message
> exactly when the variable is checked against system catalogue.

Manually testing while setting locally debug_discard_cache should be enough.

> updated patches attached

Thanks!