Hi!

In a week or two I will start work on setting up the 3.12 builders.
Once those are up and running we have a feature freeze, which means we
should avoid add new features and focus on fixing the current issues.

Once we have the builder up, we should avoid do significant changes in
toolchains and dependencies that influences how packages are built. For example:

 - abuild
 - gcc
 - linux-headers
 - binutils
 - go
 - llvm/clang
 - rust

If those (or similar packages) needs to be bumped to new major version,
now is the time to do so.

I have already bumped gcc to 9.3 and binutils to 2.34, and I plan to
tag a new release of abuild before the freeze.

Other things that I'd like to have fixed for the 3.12 release:
 - fix aarch64 crystal with newer llvm[1]
 - go default linking[2]

I would also like to fix rust triples to be consistent[3].

Are there other things that should be prioritized before the feature freeze?


[1]: https://gitlab.alpinelinux.org/alpine/aports/issues/11017
[2]: https://gitlab.alpinelinux.org/alpine/aports/issues/11260
[3]: https://gitlab.alpinelinux.org/alpine/aports/issues/11349

Hello,

I don't think it's a good idea to freeze Rust completely for stable
releases. Rust upstream currently doesn't provide patch releases and
packages using Rust are usually quick to jump onto new rustc features.
As such I'd vote for keeping Rust up-to-date in stable releases. Since
it's in community (so 6 months of support) and Rust releases every 6
weeks this would mean 4~5 upgrades. I don't expect breakage from that
since Rust has a fairly sane deprecation scheme. This would also allow
us to keep firefox in community since that usually needs a recent Rust
compiler.

As for fixing the rust triplets: I started to doubt if that's actually
useful. I mean it _is_ nice to have a custom triplet with our name on
it for the options we set, but right now there are some really weird
issues during compilation with custom triplets that I can't seem to get
behind and we currently have to work around some rust crates not
working with custom triplets and needing patches (see e.g. FF).

Regards,

Rasmus

On Mon, 2020-03-30 at 12:14 +0200, Natanael Copa wrote:
> Hi!
> ...
> Once we have the builder up, we should avoid do significant changes
> in
> toolchains and dependencies that influences how packages are built.
> For example:
> 
> ...
>  - rust
> ...
> 
> I would also like to fix rust triples to be consistent[3].
> 
> 
> [1]: https://gitlab.alpinelinux.org/alpine/aports/issues/11017
> [2]: https://gitlab.alpinelinux.org/alpine/aports/issues/11260
> [3]: https://gitlab.alpinelinux.org/alpine/aports/issues/11349

On Mon, 30 Mar 2020 12:35:58 +0200
Rasmus Thomsen <oss@cogitri.dev> wrote:

> As for fixing the rust triplets: I started to doubt if that's actually
> useful. I mean it _is_ nice to have a custom triplet with our name on
> it for the options we set, but right now there are some really weird
> issues during compilation with custom triplets that I can't seem to get
> behind and we currently have to work around some rust crates not
> working with custom triplets and needing patches (see e.g. FF).

I believe those problems comes from inconsistent triples, at least in
the firefox case? I think the firefox patch is an ugly hack for a
different underlying problem.

-nc

> 
> Regards,
> 
> Rasmus
> 
> On Mon, 2020-03-30 at 12:14 +0200, Natanael Copa wrote:
> > Hi!
> > ...
> > Once we have the builder up, we should avoid do significant changes
> > in
> > toolchains and dependencies that influences how packages are built.
> > For example:
> > 
> > ...
> >  - rust
> > ...
> > 
> > I would also like to fix rust triples to be consistent[3].
> > 
> > 
> > [1]: https://gitlab.alpinelinux.org/alpine/aports/issues/11017
> > [2]: https://gitlab.alpinelinux.org/alpine/aports/issues/11260
> > [3]: https://gitlab.alpinelinux.org/alpine/aports/issues/11349  

Can you take a look at

at https://gitlab.alpinelinux.org/alpine/abuild/merge_requests/40

before you tag an abuild release.

Tim

On Mon., Mar. 30, 2020, 7:14 a.m. Natanael Copa, <ncopa@alpinelinux.org>
wrote:

> Hi!
>
> In a week or two I will start work on setting up the 3.12 builders.
> Once those are up and running we have a feature freeze, which means we
> should avoid add new features and focus on fixing the current issues.
>
> Once we have the builder up, we should avoid do significant changes in
> toolchains and dependencies that influences how packages are built. For
> example:
>
>  - abuild
>  - gcc
>  - linux-headers
>  - binutils
>  - go
>  - llvm/clang
>  - rust
>
> If those (or similar packages) needs to be bumped to new major version,
> now is the time to do so.
>
> I have already bumped gcc to 9.3 and binutils to 2.34, and I plan to
> tag a new release of abuild before the freeze.
>
> Other things that I'd like to have fixed for the 3.12 release:
>  - fix aarch64 crystal with newer llvm[1]
>  - go default linking[2]
>
> I would also like to fix rust triples to be consistent[3].
>
> Are there other things that should be prioritized before the feature
> freeze?
>
>
> [1]: https://gitlab.alpinelinux.org/alpine/aports/issues/11017
> [2]: https://gitlab.alpinelinux.org/alpine/aports/issues/11260
> [3]: https://gitlab.alpinelinux.org/alpine/aports/issues/11349
>

Hi

Can you take a look at

at https://gitlab.alpinelinux.org/alpine/abuild/merge_requests/40

before you tag an abuild release.

Tim

On Mon., Mar. 30, 2020, 7:14 a.m. Natanael Copa, <ncopa@alpinelinux.org>
wrote:

> Hi!
>
> In a week or two I will start work on setting up the 3.12 builders.
> Once those are up and running we have a feature freeze, which means we
> should avoid add new features and focus on fixing the current issues.
>
> Once we have the builder up, we should avoid do significant changes in
> toolchains and dependencies that influences how packages are built. For
> example:
>
>  - abuild
>  - gcc
>  - linux-headers
>  - binutils
>  - go
>  - llvm/clang
>  - rust
>
> If those (or similar packages) needs to be bumped to new major version,
> now is the time to do so.
>
> I have already bumped gcc to 9.3 and binutils to 2.34, and I plan to
> tag a new release of abuild before the freeze.
>
> Other things that I'd like to have fixed for the 3.12 release:
>  - fix aarch64 crystal with newer llvm[1]
>  - go default linking[2]
>
> I would also like to fix rust triples to be consistent[3].
>
> Are there other things that should be prioritized before the feature
> freeze?
>
>
> [1]: https://gitlab.alpinelinux.org/alpine/aports/issues/11017
> [2]: https://gitlab.alpinelinux.org/alpine/aports/issues/11260
> [3]: https://gitlab.alpinelinux.org/alpine/aports/issues/11349
>

On Mon, 2020-03-30 at 12:46 +0200, Natanael Copa wrote:
> On Mon, 30 Mar 2020 12:35:58 +0200
> Rasmus Thomsen <oss@cogitri.dev> wrote:
> 
> > As for fixing the rust triplets: I started to doubt if that's
> > actually
> > useful. I mean it _is_ nice to have a custom triplet with our name
> > on
> > it for the options we set, but right now there are some really
> > weird
> > issues during compilation with custom triplets that I can't seem to
> > get
> > behind and we currently have to work around some rust crates not
> > working with custom triplets and needing patches (see e.g. FF).
> I believe those problems comes from inconsistent triples, at least in
> the firefox case? I think the firefox patch is an ugly hack for a
> different underlying problem.
> 
> -nc

No, in the Firefox case it happens because FF has a hardcoded list of
triplets and we're not on that list. There have been some improvements
in a third-party crate that FF uses for triplet detection (so
compilation itself works if FF were to use the latest version...), but
FF's python configure thingie still needs patches either way.

Regards,

Rasmus

On Mon, 30 Mar 2020 12:51:20 +0200
Rasmus Thomsen <oss@cogitri.dev> wrote:

> On Mon, 2020-03-30 at 12:46 +0200, Natanael Copa wrote:
> > On Mon, 30 Mar 2020 12:35:58 +0200
> > Rasmus Thomsen <oss@cogitri.dev> wrote:
> >   
> > > As for fixing the rust triplets: I started to doubt if that's
> > > actually
> > > useful. I mean it _is_ nice to have a custom triplet with our name
> > > on
> > > it for the options we set, but right now there are some really
> > > weird
> > > issues during compilation with custom triplets that I can't seem to
> > > get
> > > behind and we currently have to work around some rust crates not
> > > working with custom triplets and needing patches (see e.g. FF).  
> > I believe those problems comes from inconsistent triples, at least in
> > the firefox case? I think the firefox patch is an ugly hack for a
> > different underlying problem.
> > 
> > -nc  
> 
> No, in the Firefox case it happens because FF has a hardcoded list of
> triplets and we're not on that list. There have been some improvements
> in a third-party crate that FF uses for triplet detection (so
> compilation itself works if FF were to use the latest version...), but
> FF's python configure thingie still needs patches either way.

Ok. I still think we should either use `*-unknown-linux` or `*-alpine-linux`
consistenly only all arches.

I don't know which is to prefer though.

https://gitlab.alpinelinux.org/alpine/aports/issues/11349

-nc