On Thu, May 05, 2016 at 06:22:17PM -0500, Vince Mulhollon wrote:
> It'll hit the embedded world pretty hard.

Sane embedded doesn't run x86. Heck, even m68k is better than x86 there...

> Today you can buy a brand new soekris box that only runs i586. Brand
> new off the shelf, today. My 6 or 7 year old one is running right now
> as an asterisk server at home. Draws about 5 watts. Its not exactly
> the newest piece of hardware they sell but its still available as an
> embedded / embeddable device. It draws about twice the power of a
> rasp-pi, a good solid one amp at 12 volts.
>
> Its kinda sad because two years ago there were threads on the soekris
> boards about having to move from Ubuntu to Debian when i586 was
> dropped by Ubuntu and now they'll have to move AGAIN just two years
> later to ... something.

Soekris uses Geode, which is 686 save for NOPL. There's no significant gain
from NOPL, thus Debian is not planning dropping support for Geode at this
time.

There's an open issue that some stuff produces NOPL when configured for 686,
but that's being dealt with (I don't know the exact details).

> The root cause of the problem is the Debian/Ubuntu ecosystem is moving
> away from the "universal OS" mantra

Please take a look at second-class architectures: alpha hppa hurd-i386
kfreebsd-{amd64,i386} m68k mips64el powerpcspe ppc64 sh4 sparc64 x32.

This is where stuff like 586 belongs. This bunch includes outgoing
architectures that stopped being relevant at 586's time (m68k), at
Pentium 4's time (alpha hppa), possibly incoming but lacking manpower
(mips64el sparc64 x32), flip-flopping because of low manpower (kfreebsd),
niche (ppc64 -- it has a newer mainstream version in first-class),
something ancient but with 100% free hardware coming once patents expire
(sh4) and a joke arch (hurd).

All of these but powerpcspe are currently in mostly working order -- there
are smaller or bigger problems but they generally work. An architecture
with any users doesn't get dropped unless supporting it becomes hopeless
(ia64).

If you want arch/feature/etc X working... contribute! This includes not
only coding but also pushing stuff upstream, for many meanings of
"upstream": sometimes kernel, sometimes glibc, sometimes authors of a given
piece of software, sometimes Debian. Carrying a patchset yourself is not
a viable idea in the long run.

For example, I'm one of porters of x32, which is currently on the list of
second-class architectures. I'm fighting an uphill battle with some
maintainers who claim that x32 is "useless"; this is why I recently got
access to the NMU stick. I can't exactly NMU the kernel (vetoed in #778212)
or d-i, but the kernel part requires "just" work, while persuading the d-i
team is a matter of getting the rest of the port in a good enough shape.
All of this is doable, as long as you contribute instead of talking.

Of course, a single person can't possibly get all the needed parts done him-
or herself. I'll let you a secret: I'm not a Devuan user, nor I really
believe it will gain lots of popularity. So why am I on this mailing list?
Because I hope that you'll produce by some means (whatever
taking+integrating or coding from scratch) some parts that got broken by the
systemd invasion (mainly policykit/upower/etc), so they can be upstreamed
into Debian. When that is done, more ambitious projects like vdev would be
nice, too. But you can't say I'm a freeloader: my nosystemd-jessie/stretch
repository has working packages that you can use now. That's not a viable
solution long-term as required diffs will grow beyond meager tuits I can put
into their maintenance, though. That's why I need you to come with a
replacement, be it consolekit2 or whatever else. Yet having someone like me
waiting to "steal your hard work" is actually important for Devuan: if the
shape of sane inits in Debian degenerates enough to let the systemd crowd
drop them, the resultant droppage of support across 24509 source packages
will be a disaster that'd be hard to recover with your current manpower.

> the war is the move toward turning the OS into a GNOME bootloader for
> tablets.

GNOME doesn't even run on anything but i386 and amd64 (it compiles but fails
on startup). Since gnome-fallback got dropped, it absolutely needs either
certain OpenGL features (implemented only by nvidia/nouveau (not on arm* or
ppc*), radeon and intel) or llvmpipe (i386 and amd64 only). GNOME folks
will rush to claim that someone ported gnome-shell to freedreno, but that's
only a hacked version of Fedora, as far as I know not upstreamed, and
certainly not in Debian. I've repeatedly asked on debian-devel for reports
of anyone who managed to get gnome-shell running on !i386 !amd64, not a
single person stepped up.

Thus, the presence of other architectures as release archs suggests there's
more than just a "Gnome bootloader"...


Meow!
--
A tit a day keeps the vet away.