RE: Postgres

[Brent Fulgham <brent.fulgham@xpsystems.com>]

> >    base/libc0.2_2.1.2-10.deb            and
> >    devel/libc0.2-dev_2.1.2-10.deb
> > 
> > The relevant fix however, was to the pfinet server, which is part of
> > the hurd package.  The relevant ChangeLog entry is:
> > 
> >    1999-10-22  Roland McGrath  <roland@baalperazim.frob.com>
> > 
> > 	   * socket-ops.c (S_socket_getopt): Implement the call.
> > 	   All options supported by the code in linux-inet are in
> > 	   fact of type int, so we can support just that one size.
> > 	   (This is still a bogus untyped interface!)
> > 
> > So you need a hurd package dated well beyond that.
> > 
> 
> But i have a current hurd package. But, maybe it isn't active, I don't
> know. So, please explain it to me. I installed this hurd - system
> with the big tarball ( this one is outdated, sure ). After that
> i set up the system so taht it was running. But after that i installed
> hurd_19991209 via dpkg. typing dpkg -l hurd shows the following :
> 
> ii  hurd            19991209       The GNU Hurd
> 
> the same results with hurd-dev, gnumach and gnumach-dev.
> the libc0.2's are also the newest versions.
> So did I make a mistake ? Can't I update the system via dpkg ?
> 
Everything seems current.  You can certainly update a system via dpkg --
in fact, that's the prefered way :-)

One possibility is that the build of Hurd doesn't include the fix.  Maybe
Marcus can tell us (I'm not sure) -- but I would guess by the date
on the package he's saying it's a build from CVS as of December 9, 1999,
which certainly seems like it should have the pfinet fix.

Mark, can you verify if the fix is sufficient for the relevant sockopt
setting?  Maybe this particular socket option is not yet supported.

SVR4 indicates:
SO_REUSEADDR -- Indicates that the rules used in validating addresses
provided with calls to `bind` should be modified to allow re-use of
local addresses.

This seems pretty critical -- most likely postgres won't work if it can't
reuse local addresses.

I notice that getopt/setopt are available under pfinet, but that they are
not
implemented under pfilocal.  Am I correct in assuming that pfilocal might be
used for socket handling under loopback conditions?  If so, that might
explain why we are seeing the option unsupported error message.

Mark, can you shed any light on this?

-Brent