On 2004-06-23, John Summerfield penned:
Yes, but there's no way to test those backports thoroughly enough to
match the amount of testing that went into stable in the first place.
Do you believe that?
The point of stable is not just that each package has been tested to the nth
degree, it's also that the system as a whole has been tested -- the
complex web of interactions among packages. In order to accept these
backports into stable, we either have to perform the same amount of
testing *on the whole system* as we did to release stable in the first
place, or the system can't be certified to be truly stable.
When we change one line on the flight software where I work, we can't
just test the unit that was changed and move on. We have to perform the
complete system test all over again to make sure nothing unexpected
happens. The same concept applies to servers.
I've worked in environments where we didn't test the system after making
"small" changes. It's not pretty.
I switched over from OS/2 to 5.0. I was surprised later to discover
people regarded it as buggy. I don't recall how much I used 6.0, but
where I work we still have a 7.0 box in place: I chose 7.0 over 7.1 so
as to have a 2.2 kernel as standard (required for a sat card).
It seems odd to me to choose a release based on the kernel, but okay.
It seems *very* odd that you're telling us that RedHat switched major
kernel numbers for a minor release.
The most troublesome system I have is one running Woody, the video
regularly gets stuffed up and it's prone to losing its keyboard.
Changing the graphics card made no difference.
Is it possible it's the motherboard? Those are some weird problems.
It sounds like a lot more work for the developers. RedHat had
commercial customers to support their developers. How would you
suggest Debian manage this?
I thnk Red Hat didn't have commercial customers when it started on
this model.
No, but they were always a company looking to make money off of their
product (not that there's anything wrong with that). Debian has no such
plans, and that's one of the reasons why I trust them to do what's right
rather than what's profitable. It's also one of the reasons it's been
suggested as a reference implementation a number of times.
As I already said, there are enough developers doing enough work - the
packages are out there. What is missing official adoption by Debian,
and the coordination that would follow its adoption.
No, what's missing is the testing infrastructure. *System* testing, not
just the individual package.
If there was an official line of "built for Stable" packages
comprising packages people felt were needed, linked to from the
dowload page, be sure a lot of people would try them out.
And now a lot of people who aren't motivated enough to do a google
search or ask on d-u are installing packages that haven't been fully
tested with the system. The status quo at least ensures that the people
who are using backports have at a minimum the ability to research
questions.