* Robert Haas (robertmhaas@gmail.com) wrote:
> On Thu, Jan 30, 2014 at 5:05 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> > Robert Haas <robertmhaas@gmail.com> writes:
> >> On Thu, Jan 30, 2014 at 11:04 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >>> I think this is totally misguided. Who's to say that some weird FDW
> >>> might not pay attention to attstorage? I could imagine a file-based
> >>> FDW using that to decide whether to compress columns, for instance.
> >>> Admittedly, the chances of that aren't large, but it's pretty hard
> >>> to argue that going out of our way to prevent it is a useful activity.
> >
> >> I think that's a pretty tenuous position. There are already
> >> FDW-specific options sufficient to let a particular FDW store whatever
> >> kinds of options it likes; letting the user set options that were only
> >> ever intended to be applied to tables just because we can seems sort
> >> of dubious. I'm tempted by the idea of continuing to disallow SET
> >> STORAGE on an unvarnished foreign table, but allowing it on an
> >> inheritance hierarchy that contains at least one real table, with the
> >> semantics that we quietly ignore the foreign tables and apply the
> >> operation to the plain tables.
> >
> > [ shrug... ] By far the easiest implementation of that is just to apply
> > the catalog change to all of them. According to your assumptions, it'll
> > be a no-op on the foreign tables anyway.
>
> Well, there's some point to that, too, I suppose. What do others think?
I agree that using the FDW-specific options is the right approach and
disallowing those to be set on foreign tables makes sense. I don't
particularly like the idea of applying changes during inheiritance
which we wouldn't allow the user to do directly. While it might be a
no-op and no FDW might use it, it'd still be confusing.
Thanks,
Stephen