[Pkg-rust-maintainers] Efficiently supporting (or not) other arches in the long term

[Angus Lees]

(First: An enormous thanks to inifinity0 and others who have been tracking
down the build failures each release.)

I think somehow we need to "better align upstream+our incentives" - I
assume all parties want to see Rust well supported on lots of
architectures, we just have to remove whatever is preventing that.

I think to do that we have to somehow make the Debian buildd farm available
for upstream, and encourage upstream to test/check releases against our
architectures.

I suspect this means building some sort of pre-release (either beta or
nightly) - presumably via experimental on all Debian archs.  I suspect it
will need to be nightly in order to be swift enough for upstream to get
into a useful break-fix-check feedback loop.  If we get this right, we
should be able to step out of the critical path and just let upstream deal
directly with our build reports.

Then we need to make build breakages prominent enough for upstream to
notice.  Automatically filing bugs (perhaps - only useful if we can keep
the false positives very low), prominent build status reports, or
something.  Not sure what other options we have here...

Re testing migration: I think there's something to be said for just
allowing the occasional version to be held back from testing migration.  If
we fall several versions behind in testing because it's broken on mips
(eg), then I suspect this will fix itself one way or another once mozilla
et al start shipping higher priority things that demand newer rustc's.
Again, if we "hide" failures by whitelisting architectures, or skipping
entire testsuites then we're also just hiding the failures from upstream.
I'd like it if we could come up with some way that left build failures
prominent but also removed the pressure infinity0 is feeling from having to
chase every failure on every release.

 - Gus

On Thu, 29 Dec 2016 at 09:12 Sylvestre Ledru <Sylvestre at debian.org> wrote:

> From the issues I saw in the past, test failures seemed minimal to me,
> corner cases without impact on the overall quality of the compiler.
> In most of the rust releases, we are already ignoring some tests.
>
> If we are talking about a few tests failing, I would not set on the team
> unreasonable goals.
>
> Le 28 décembre 2016 23:00:22 GMT+01:00, Josh Triplett <
> josh at joshtriplett.org> a écrit :
>
> On Wed, Dec 28, 2016 at 03:40:00PM +0000, Ximin Luo wrote:
>
>  Sylvestre Ledru:
>
>  Le 26/12/2016 à 15:43, Ximin Luo a écrit :
>
>  Rust upstream does not regularly run tests for anything other than amd64 or i386:
>
>  https://forge.rust-lang.org/platform-support.html
>
>  This has resulted in the past few releases of Debian's rustc being held back from testing migration, due to arm64 breakages that we had to file upstream:
>
>  1.10.0 https://github.com/rust-lang/rust/pull/34889
>  1.11.0 https://github.com/rust-lang/rust/issues/36348
>  1.12.0 there probably was one but I can't remember it
>  1.13.0 https://github.com/rust-lang/rust/issues/37225
>  1.14.0 https://github.com/rust-lang/rust/issues/38612
>
>  We can't keep doing this long-term, I don't have time to chase these things every 6 weeks. And soon we might start supporting 7 new architectures, which will have their own test failures.
>
>  What shall we do about this? I think it's nice to support as many architectures as possible, and that is strongly encouraged by Debian - but upstream don't provide the same level of support. I'll suggest two options:
>
>  Or we could just fail at
> build time on tier-1 and continue if the
>  testsuite fails on the other archs.
>  To be honest, this is what I am doing with LLVM & Clang...
>
>
>  I'm reluctant to do this, because I don't know if these failures would cause buggy behaviour or security problems. If we do this, at least we should somehow make it very clear to users that the other platforms have these problems, and perhaps even include the test failure logs in the binary package.
>
>
> I don't think we should ship builds in unstable/testing/stable that
> don't pass Rust's extensive testsuite.  Perhaps in experimental.
>
> - Josh Triplett
>
> ------------------------------
>
> Pkg-rust-maintainers mailing list
> Pkg-rust-maintainers at lists.alioth.debian.org
> https://lists.alioth.debian.org/mailman/listinfo/pkg-rust-maintainers
>
>
> --
> Envoyé de mon appareil Android avec K-9 Mail. Veuillez excuser ma brièveté.
> _______________________________________________
> Pkg-rust-maintainers mailing list
> Pkg-rust-maintainers at lists.alioth.debian.org
> https://lists.alioth.debian.org/mailman/listinfo/pkg-rust-maintainers
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-rust-maintainers/attachments/20161229/71b6f54f/attachment-0001.html>