Re: our checks for read-only queries are not great

Greetings,

* Robert Haas (robertmhaas(at)gmail(dot)com) wrote:
> On Tue, Jan 14, 2020 at 1:46 PM Stephen Frost <sfrost(at)snowman(dot)net> wrote:
> > * Tom Lane (tgl(at)sss(dot)pgh(dot)pa(dot)us) wrote:
> > > Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> > > > Speaking of sensible progress, I think we've drifted off on a tangent
> > > > here about ALTER SYSTEM.
> > >
> > > Agreed, that's not terribly relevant for the proposed patch.
> >
> > I agree that the proposed patch seems alright by itself, as the changes
> > it's making to existing behavior seem to all be bug-fixes and pretty
> > clear improvements not really related to 'read-only' transactions.
>
> There seems to be no disagreement on this point, so I have committed the patch.

Works for me.

> > It's unfortunate that we haven't been able to work through to some kind
> > of agreement around what "SET TRANSACTION READ ONLY" means, so that
> > users of it can know what to expect.
>
> I at least feel like we have a pretty good handle on what it was
> intended to mean; that is, "doesn't cause semantically significant
> changes to pg_dump output." I do hear some skepticism as to whether
> that's the best definition, but it has pretty good explanatory power
> relative to the current state of the code, which is something.

I think I agree with you regarding the original intent, though even
there, as discussed elsewhere, it seems like there's perhaps either a
bug or a disagreement about the specifics of what that means when it
relates to committing a 2-phase transaction. Still, setting that aside
for the moment, do we feel like this is enough to be able to update our
documentation with?

Thanks,

Stephen