Re: [PoC] Federated Authn/z with OAUTHBEARER

Hi,

On 2025-04-08 13:02:11 -0400, Bruce Momjian wrote:
> On Tue, Apr 8, 2025 at 06:57:18PM +0200, Wolfgang Walther wrote:
> > Jacob Champion:
> > > On Tue, Apr 8, 2025 at 9:32 AM Wolfgang Walther <walther(at)technowledgy(dot)de> wrote:
> > > > And that should also not be a problem for distributions - they could offer a libpq and a libpq_oauth package, where only one of them can be installed at the same time, I guess? *
> > > My outsider understanding is that maintaining this sort of thing
> > > becomes a major headache, because of combinatorics. You don't really
> > > want to ship a libpq and libpq-with-gss and libpq-with-oauth and
> > > libpq-with-oauth-and-gss and ...
> >
> > That would only be the case, if you were to consider those other
> > dependencies as "dangerous" as cURL. But we already depend on them. So if
> > it's really the case that cURL is that much worse, that we consider loading
> > it as a module... then the combinatorics should not be a problem either.
> >
> > However, if the other deps are considered problematic as well, then the ship
> > has already sailed, and there is not point for a special case here anymore.
>
> Yes, I think this is what I am asking too. For me it was curl's
> security reputation and whether that would taint the security reputation
> of libpq. For Tom, I think it was the dependency additions.

I'd say that curl's security reputation is higher than most of our other
dependencies. We have dependencies for libraries with regular security issues,
with those issues at times not getting addressed for prolonged amounts of
time.

I do agree that there's an issue increasing libpq's indirect footprint
substantially, but I don't think that's due to curl's reputation or
anything. It's just needing a significantly higher number of shared libraries,
which comes with runtime and distribution overhead.

Greetings,

Andres Freund