MIME-Version: 1.0
References: <CAMmSBy8WR2FgKEMApDFTYqSeuA1=H107wWZ-UpE3ZbqGhqnXeQ@mail.gmail.com>
 <b0654142-781b-e527-32f2-551fd17ec5e0@dereferenced.org>
In-Reply-To: <b0654142-781b-e527-32f2-551fd17ec5e0@dereferenced.org>
From: Roman Shaposhnik <roman@zededa.com>
Date: Mon, 10 May 2021 18:45:12 -0700
Message-ID: <CAMmSBy_umr3OqThczKQPcb3_Ofd7=orreGVqr3+iuY+7zm8MMg@mail.gmail.com>
Subject: Re: Progress on Alpine for RISC-V (and BeagleV)
To: Ariadne Conill <ariadne@dereferenced.org>
Cc: ~alpine/devel@lists.alpinelinux.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Sun, May 9, 2021 at 8:35 AM Ariadne Conill <ariadne@dereferenced.org> wr=
ote:
>
> Hello,
>
> On Sat, 8 May 2021, Roman Shaposhnik wrote:
>
> > Hi!
> >
> > A few updates for anyone interested in Alpine bringup on the board:
> >
> > I=E2=80=99ve bootstrapped quite a bit of Alpine packages and published =
updated
> > aports here https://github.com/rvs/aports
> >
> > A growing collection of packages is now available under
> >   https://eve-alpine-packages.s3.amazonaws.com/edge/{main,community}
> >
> > It is enough to generate basic images, so for example, here=E2=80=99s a=
 minirootfs
> > published that can be used to bootstrap your own containers
> >   https://eve-alpine-packages.s3.amazonaws.com/edge/images/alpine-minir=
ootfs-210509-riscv64.tar.gz
>
> Very good!

There's also binary install images available now:
    https://eve-alpine-packages.s3.amazonaws.com/edge/images/alpine-riscv-2=
10511-riscv64.tar.gz
and
    https://eve-alpine-packages.s3.amazonaws.com/edge/images/alpine-netboot=
-210511-riscv64.tar.gz

> > I still need TONS of help to fix/bootstrap more packages =E2=80=94 and =
for that
> > I published a builder environment =E2=80=94 so if you (on a latest Dock=
er for Mac
> > or docker + qemu-user-riscv64) do:
> >         $ docker build -t alpine-riscv64
> > https://eve-alpine-packages.s3.amazonaws.com/edge/images/Dockerfile
> >         $ docker run -it --rm alpine-riscv64
> >         builder$ cd aports
> > You will end up in an environment that I=E2=80=99m using to build all o=
f these
> > packages and can help me with:
> >          cd <YOUR FAVORITE PACKAGE>
> >          hack APKBUILD (or not)
> >          abuild -r
> >
> > Now, for the most perplexing question of all: it is pretty clear that
> > it may be a bit overwhelming for a single person (like me) to make
> > sure that we do ALL of the packages at once to support RISC-V.
>
> When we first got mips64 bootstrapped, I just blocked packages on mips
> that were non-essential that didn't build.

Aha! Quick question: from that experience -- do you remember what were
the packages that were deemed a MUST for the port?

> Most maintainers want to see
> their packages be portable across architectures and will work to fix them
> when they notice arch=3D"all !riscv64" in their APKBUILDs.

For sure! But are you saying that it would be acceptable to first bulk upda=
te
all the APKBUILD files with arch=3D"all !riscv64" (sans those that actually
compile just fine)?

> > So I'm wondering if there's any way to do it steps -- where we bring
> > up progressively more and more packages into the release, but at
> > least the ones that are already supported are not allowed to bitrot
> > anymore on RISC-V. Is there any process around Alpine releases
> > that would allow us to do that?
>
> Yes, please open MRs for packages which need modifications for the RISC-V
> port.

I plan to start doing it next week -- once I get a chance of running with A=
lpine
as my default distribution on this board for a few days more.

> I'm willing to operate a qemu-user builder for riscv64 (I'm quite
> convinced this is the only way we're going to make progress with the
> current build system, but ncopa is planning a new one out).

That's what I've been doing so far and I must say -- the performance
was REALLY decent (I just used my beefy laptop). On top of that I've
been thinking of trying to see how it flies on the biggest boxes available
on Packet.net.

Thanks,
Roman.