Hello from powerpc

Hello from powerpc

[Carlos Sánchez de La Lama]

Hi all,

success, finally:

--8<---------------cut here---------------start------------->8---
$ ./pre-inst-env guix build hello
[...]
/gnu/store/ynpd0qpppl0fdh252wns87d158pbdx8q-hello-2.10

$ /gnu/store/ynpd0qpppl0fdh252wns87d158pbdx8q-hello-2.10/bin/hello
Hello, world!

$ uname -a
Linux bilbo 3.16.0-4-powerpc #1 Debian 3.16.36-1+deb8u1 (2016-09-03) ppc
GNU/Linux
--8<---------------cut here---------------end--------------->8---

I am unsure on the best way to integrate my changes upstream. I have to
review some of the patches I sent those last weeks, some of which
required minor changes to be accepted, and send some new ones to the
list.

For starters, one question: I *know* the exact bootstrap binaries I have
used now work (because I relaunched the bootstrap from scratch on
friday). However, if I regenerate them they are going to be slightly
different (because I made a change to gcc after generating them, and
static gcc inherits it). Do you guys think is better to accept the
proven bootstrap binaries as the "good" ones, keep current hashes and
publish the tarballs, or merge everything first and then generate new
bootstrap binaries, updating the hashes?

BR

Carlos

Re: Hello from powerpc

[Ludovic Courtès]

Hi Carlos!

csanchezdll@gmail.com (Carlos Sánchez de La Lama) skribis:

> success, finally:
>
> $ ./pre-inst-env guix build hello
> [...]
> /gnu/store/ynpd0qpppl0fdh252wns87d158pbdx8q-hello-2.10
>
> $ /gnu/store/ynpd0qpppl0fdh252wns87d158pbdx8q-hello-2.10/bin/hello
> Hello, world!
>
> $ uname -a
> Linux bilbo 3.16.0-4-powerpc #1 Debian 3.16.36-1+deb8u1 (2016-09-03) ppc
> GNU/Linux

Woohoo!  Congrats!  \o/

> I am unsure on the best way to integrate my changes upstream. I have to
> review some of the patches I sent those last weeks, some of which
> required minor changes to be accepted, and send some new ones to the
> list.

Sure.  Just ask if in doubt.

There’s the question of what to do with that architecture from a
maintenance viewpoint.

Ideally we’d provide binaries for that architecture, but for that we’d
need at the very least two build machines.  Do you have an idea as how
we could get donations for that?  Perhaps we could discuss it with the
Talos folks, they may be interested in having more free software
developers working on PPC.

Also we’ll need actual users and developers so we can sustain
maintenance of that platform—fix packages that break, help out when a
major toolchain upgrade doesn’t work as smoothly as it should, etc.  For
that we’ll mostly need to promote the port in PowerPC circles, I
suppose.

If we fail to do that, I think we’ll can still have the patches in but
prominently mark the platform as “unofficially supported” or something
along these lines (like Debian does).  If after a couple of years the
situation hasn’t improved, we might want to discuss whether to drop it.

WDYT?

> For starters, one question: I *know* the exact bootstrap binaries I have
> used now work (because I relaunched the bootstrap from scratch on
> friday). However, if I regenerate them they are going to be slightly
> different (because I made a change to gcc after generating them, and
> static gcc inherits it). Do you guys think is better to accept the
> proven bootstrap binaries as the "good" ones, keep current hashes and
> publish the tarballs, or merge everything first and then generate new
> bootstrap binaries, updating the hashes?

Does cross-compilation to powerpc-linux-gnu work in current master?
If/when it does, we can ask Hydra to cross-build a few things, like we
already do for other targets:

  https://hydra.gnu.org/job/gnu/master/mips64el-linux-gnuabi64.bootstrap-tarballs-0.x86_64-linux
  https://hydra.gnu.org/job/gnu/master/arm-linux-gnueabihf.bootstrap-tarballs-0.x86_64-linux

That way we can maybe reuse the latest binaries without yourself having
to rebuild them by hand, and it will provide a clearer audit trail.

Thoughts?

Ludo’.

Re: Hello from powerpc

[Carlos Sánchez de La Lama]

Hi!,

>> $ /gnu/store/ynpd0qpppl0fdh252wns87d158pbdx8q-hello-2.10/bin/hello
>> Hello, world!
>>
>> $ uname -a
>> Linux bilbo 3.16.0-4-powerpc #1 Debian 3.16.36-1+deb8u1 (2016-09-03) ppc
>> GNU/Linux
>
> Woohoo!  Congrats!  \o/

Thanks :)

>> I am unsure on the best way to integrate my changes upstream. I have to
>> review some of the patches I sent those last weeks, some of which
>> required minor changes to be accepted, and send some new ones to the
>> list.
>
> Sure.  Just ask if in doubt.

Slowly progressing, quite busy with lately. I will try to address
pending comments on patches I already sent (and which are required for
powerpc support) before sending the new ones. There is nothing really
complex, the patches are rather simple (as usual, once you know what to
patch ;) )

> Ideally we’d provide binaries for that architecture, but for that we’d
> need at the very least two build machines.  Do you have an idea as how
> we could get donations for that?  Perhaps we could discuss it with the
> Talos folks, they may be interested in having more free software
> developers working on PPC.

I don't really have any ideas. I work in a tech startup which has
nothing to do with GUIX, so no luck here. My interest in GUIX is purely
personal. About asking the Talos people, I wonder how much sense it does
as their product is powerpc64 and what I have bootstrapped is
powerpc32. I feel powerpc32 is almost dead now (I even read today Debian
is dropping support on next Debian 9) so it only has interest for those
having an old machine around. It might have some users on the embedded
market (there are still some FPGAs with powerpc cores inside AFAIK), but
I wont expect getting much industrial support/donations.

> If we fail to do that, I think we’ll can still have the patches in but
> prominently mark the platform as “unofficially supported” or something
> along these lines (like Debian does).  If after a couple of years the
> situation hasn’t improved, we might want to discuss whether to drop
> it.

That would be enough, for starters, I think. Making it "unofficially
supported" and see how much interest it brings. Maybe Debian dropping
powerpc support makes all those powerpc users around look at GUIX to
keep their systems up-to-date.

> Does cross-compilation to powerpc-linux-gnu work in current master?
> If/when it does, we can ask Hydra to cross-build a few things, like we
> already do for other targets:

Not in current master, I think. At least some minor patches are
required. Anyways, rebuilding the bootstrap binaries is not so hard
(they are cross-build, so I do it in a bigger x86 machine), but
everytime the bootstrap binaries change, everything must be rebuilt on
the target machine as they are root to all dependency graphs (am I
correct on this?). That's why I would like to "fix" the bootstrap
binaries (like on current supported targets, where bootstrap binaries
were generated some time ago and actual binaries have not changed
since).

In any case, I think first step is have the patches needed for bootstrap
generation in master, then generate the binaries, and at that point we
can fix them.

BR

Carlos

-- 
'La vie est une longue chute, Marcus. Le plus important est de savoir tomber.'

Joël Dicker, "La Vérité sur l'affaire Harry Quebert" (2012)

Re: Hello from powerpc

[Ludovic Courtès]

Hello Carlos!

csanchezdll@gmail.com (Carlos Sánchez de La Lama) skribis:

> Slowly progressing, quite busy with lately. I will try to address
> pending comments on patches I already sent (and which are required for
> powerpc support) before sending the new ones. There is nothing really
> complex, the patches are rather simple (as usual, once you know what to
> patch ;) )

:-)

>> Ideally we’d provide binaries for that architecture, but for that we’d
>> need at the very least two build machines.  Do you have an idea as how
>> we could get donations for that?  Perhaps we could discuss it with the
>> Talos folks, they may be interested in having more free software
>> developers working on PPC.
>
> I don't really have any ideas. I work in a tech startup which has
> nothing to do with GUIX, so no luck here. My interest in GUIX is purely
> personal. About asking the Talos people, I wonder how much sense it does
> as their product is powerpc64 and what I have bootstrapped is
> powerpc32. I feel powerpc32 is almost dead now (I even read today Debian
> is dropping support on next Debian 9) so it only has interest for those
> having an old machine around. It might have some users on the embedded
> market (there are still some FPGAs with powerpc cores inside AFAIK), but
> I wont expect getting much industrial support/donations.

OK, fine.  So maybe we’ll keep it as an unofficial port, until/unless we
find out that there’s more demand that we thought out there.

>> If we fail to do that, I think we’ll can still have the patches in but
>> prominently mark the platform as “unofficially supported” or something
>> along these lines (like Debian does).  If after a couple of years the
>> situation hasn’t improved, we might want to discuss whether to drop
>> it.
>
> That would be enough, for starters, I think. Making it "unofficially
> supported" and see how much interest it brings. Maybe Debian dropping
> powerpc support makes all those powerpc users around look at GUIX to
> keep their systems up-to-date.

OK!

>> Does cross-compilation to powerpc-linux-gnu work in current master?
>> If/when it does, we can ask Hydra to cross-build a few things, like we
>> already do for other targets:
>
> Not in current master, I think. At least some minor patches are
> required. Anyways, rebuilding the bootstrap binaries is not so hard
> (they are cross-build, so I do it in a bigger x86 machine), but
> everytime the bootstrap binaries change, everything must be rebuilt on
> the target machine as they are root to all dependency graphs (am I
> correct on this?). That's why I would like to "fix" the bootstrap
> binaries (like on current supported targets, where bootstrap binaries
> were generated some time ago and actual binaries have not changed
> since).
>
> In any case, I think first step is have the patches needed for bootstrap
> generation in master, then generate the binaries, and at that point we
> can fix them.

Sure, I’ll wait for the patches then.  :-)

When we’re done, I think it’d be great if you could write a blog entry
for the web site to report on your experience porting Guix, and to
encourage interested hackers into using the port and contributing to it.

Thanks!

Ludo’.