[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Linux 2.0.36 in slink?

On Fri, Dec 18, 1998 at 03:47:57AM -0000, Robert Woodcock wrote:
> Oscar Levi <elf@buici.com> wrote:
> >We don't have guidelines for release that we can use to decide if this
> >is important or not.  In your opinion, is it worth a three-week delay
> >to switch kernels?  I am not saying it would take that long.  I am
> >saying that a change like the kernel deserves extensive testing which
> >we are not now doing.  I'd rather ship a kernel with alleged security
> >holes than a broken distribution.
> Do you really think we can get slink out the door in 3 weeks?

That is a complicated question--I assume you ask it with the intention
of continuing the discussion.  Moreover, I'm gonna assume you don't
ship software for a living, so I'll have to be verbose.  The only
thing I can say with certainty is that the slink we have today doesn't
install...for me.  That means an indefinite delay since testing
requires installing.

In an ideally developed distribution, we'd have confidence tests for
each package.  Any package uploaded to the archive would have to pass
its confidence tests before being stored in any of the archive
directories.  Practically speaking, this is most important for
packages in base and especially ones that we create specially for
Debian.  IIRC, we used to have boot floppies that worked, but those no
longer exist in slink.  I believe this is true because I installed
slink a couple of times from a local mirror.

Having these tests helps us measure convergence which in turn helps us
predict when the distribution will be ready to ship.  The thread about
whether or not we should include a newer kernel started because there
is a growing number of folks who believe we should switch.  No one was
willing to risk it when 2.0.36 was first released, but now that the
jury has returned it looks like a win over 2.0.35.  Fine.  But
switching kernels will shake the jello.  Imagine a software project
like a bowl of jello.  We work and work and shake it and shake it.
Finding bugs gets the jello moving.  Fixing them and finishing
features calms it.  After the freeze, we hope to see it gel quickly,
but the fixes often introduce more problems.  It's slowing down, now.
It's slowing down.  It stops.  Ship it!  Without predictors for
convergence, any code change at this point could delay us
indefinitely.  This does not mean forever; it means we don't know for
how long.

OK.  I'm not saying that Debian is doing a bad job.  I'm here because
I want to be.  A lot of the important things are adequately in place:
bug reporting, the .deb format, the archive and mirrors, there
certainly are others.  We all know the weak spots, too.  So, could we
ship in three weeks?  I'd say the chances are slim, but it depends on
how the cd creation and boot-floppies go.  Does it really matter?
Well, I'd like to be able to install slink so we can continue
improving the distribution.

Reply to: