unfrozen repo somewhere?

[Horst H. von Brand]

Les Mikesell <lesmikesell at gmail.com> wrote:
> Horst H. von Brand wrote:

[...]

> >> You probably can't match RHEL's QA for free, but testing - and not
> >> shipping things that don't pass - is the only way things can get
> >> better.

> > Pray tell us, exactly how do we /do/ that? I think we all agree that
> > that is the ideal, but I strongly believe it is unatainable in pratice.

> The big problem for me is that it's a package deal.  I wouldn't mind
> beta testing a few apps at a time on my working system with a stable
> OS and libraries, but to run them you also have to take an
> experimental kernel and device drivers.  And my history with those on
> fedora is that I waste too much time getting the hardware to work with
> new versions. Maybe that's changed...   If you had a way to separate
> the apps from the OS, you might find people more willing to test the
> parts that interested them.

Many problems are integration problems, i.e., package version foo doesn't
work with library version bar. It also happens that to run the very
latest-and-greatest version of the package you need experimental(ish)
versions of other stuff.

> >> I've heard the term used as in "an instrument's tuning is 'good
> >> enough' for folk music" or "an approximation is 'good enough' for
> >> government work".  What's 'good enough' for an operating system?

> > Works mostly. Doesn't crash too often. Doesn't destroy valuable data
> > except under extremely unlikely circumstances.

> Kernels that don't boot on machines where the last version worked

Keep the next-to-last one around just in case. Have a LiveCD of the lastest
stable/beta at hand for the case it gets too screwed up. Yes, it's a
nuisance.

>                                                                   or
> lose the ability to access devices like firewire drives isn't 'mostly'
> enough for me.

Ditto.

> > This is engineering, not mathematics. And even in mathematics there have
> > been mistakes...

> But would you want to test a plane where the engineers said it was
> probably good enough and didn't crash too often?

So you have never heard of plane crashes? Or other failures? In the end, as
100% perfect is not realistically doable, you have to make do with "good
enough", and what /that/ means depends on the circumstances: The kernel of
a game console (a crash means a minor inconvenience or at most a lost game)
has /very/ different safety/security requirements than software controlling
a radiation therapy aparatus (which could very well be lethal to the
patient).
-- 
Dr. Horst H. von Brand                   User #22616 counter.li.org
Departamento de Informatica                    Fono: +56 32 2654431
Universidad Tecnica Federico Santa Maria             +56 32 2654239
Casilla 110-V, Valparaiso, Chile 2340000       Fax:  +56 32 2797513