[Sugar-devel] Community XO software builds

[Peter Robinson]

On Thu, May 7, 2015 at 12:10 AM, James Cameron <quozl at laptop.org> wrote:
> On Wed, May 06, 2015 at 09:19:45AM -0300, Gonzalo Odiard wrote:
>> I think we should try make a build using CentOS. I don't know if
>> have all the packages we need, but the rate of change in Fedora was
>> difficult to follow when OLPC had a team dedicated and now is almost
>> impossible. The true is we didn't finished to solve the problems we
>> found in F20, and Fedora is working in F22.
>
> I do not think we should switch from Fedora to CentOS, because;
>
> 1.  our installed base express interest in Fedora or Ubuntu,

Daniel Drake, myself and others put in a lot of effort back in the
F-14/15 days to get everything upstream into Fedora. I continue to
maintain that and produce a Sugar on a Stick release with every Fedora
release.

In the last release Daniel and I was involved in the delta between
Fedora and the OLPC release was very minimal. Basically kernel,
firmware, and some minor changes to a couple of Sugar packages for XO
HW and patches that weren't yet upstream.

> 2.  there are missing desktop packages, which means we are taking on
> maintenance of those packages on CentOS,

Having tried and failed to do this back when EL6 was new I believe
this is a dead end. It turned out to be _WAY_ more effort than
actually keeping Fedora up to date. The upstream RHEL releases are
every 6 months but if you need a fix for a package in the core 2500
odd packages and it's not easy you might be waiting a lot longer for a
fix.

In Fedora if you know the right people (like me) you can get a fix
into update-testing in a day. Also there's a much much wider QA group
across the packages we use and care about.

I can go on and on about the details required for this but basically I
suspect eyes have glazed over already.

> 3.  we would delay necessary work until the next release of CentOS, or
> if the work is too large we may never upgrade.

I suspect it would be never.

> Let me explain that last point.
>
> There is a continuous flow of changes into Fedora.  These changes
> eventually flow into Red Hat Enterprise Linux, and thus into CentOS.
>
> The most cost effective way to handle this flow was for developers to
> test changes on our builds, every week.  This gaves us awareness of
> the change and kept us involved to resist changes that cause damage.
> We were there once.  It required a low but continuous engineering
> effort.

It use to take around an hour to cut a release from Fedora/Sugar
repos. Quite often the delta from a patch for a fix being created and
a new OS was in the hours timeframe. It's the usual story of a little
bit of effort regularly stops it from being a major issue.

Kernel and olpc-os-builder aside I think you could probably produce a
working image of Fedora 22 now. I think all the userspace bits are
likely there and working due to my SoaS work.

It's actually the thing that annoys me most about the sugar community.
IMO we have a great working Sugar release that works pretty much
everywhere plus is a great proven base for XO releases yet so many
core developers have told me "if only you'd focus on Ubuntu we'd use
it" yet Ubuntu for _YEARS_ have shown that they couldn't given a shit
and even actively remove core bits needed (remember the Browse on
Mozilla years anyone??) to make it even harder.

> The next most cost effective way is to do this work only when a new
> release of Fedora occurs.  This results in lots of head scratching and
> bug fixing, and new builds, until the bugs are mostly gone.  We are
> here now.  It requires bursts of engineering effort.

Actually it needs work _BEFORE_ a new release happens, any work now
IMO should be focused on Fedora 23. That way you have everything in
place in time for Fedora 23 GA in October and you get the longest
value out of the release.

> The least cost effective way is to hold off doing that work for three
> years until the next CentOS release.  This would be a lot more work in
> a much shorter burst.

And you'll likely end up in a very disparate stability across devices.
Both ARMv7 and i686 is community supported in CentOS which means you
get likely dubious quality of work and I suspect due to toolchain
config choices for i686 it won't even run on the XO-1. Has anyone
actually tried booting CentOS-7 on a XO1? From what I've seen of the
ARMv7 efforts I see it as half arsed at best.

People ask me if I can help with CentOS. The answer is no. I have no
personal interest in CentOS. I have enough to do with personal
projects on Fedora.

> Delaying effort until a future time hasn't worked, and I don't think
> it will.  Meanwhile, I'm trying as hard as I can with what I'm doing.

And I've been trying as hard with Fedora as possible. The core Sugar
stack is in pretty good shape. There's some work needed on some
Activies but most of the work it to update them to the latest upstream
bits.

Peter

https://alt.fedoraproject.org/pub/alt/stage/22_TC2/Images/armhfp/Fedora-SoaS-armhfp-22-TC2-sda.raw.xz
https://alt.fedoraproject.org/pub/alt/stage/22_TC2/Live/i386/Fedora-Live-SoaS-i686-22-TC2.iso
https://alt.fedoraproject.org/pub/alt/stage/22_TC2/Live/x86_64/Fedora-Live-SoaS-x86_64-22-TC2.iso