Re: 01/01: gnu: address@hidden: Remove aarch64-linux from supported-systems.

[Efraim Flashner]

On Sun, Dec 02, 2018 at 03:51:58PM -0500, Mark H Weaver wrote:
> address@hidden (Ludovic Courtès) writes:
> 
> > Efraim Flashner <address@hidden> skribis:
> >
> >> On Sat, Dec 01, 2018 at 05:23:32PM -0500, Mark H Weaver wrote:
> >>> Hi Efraim,
> >>> 
> >>> address@hidden writes:
> >>> 
> >>> > efraim pushed a commit to branch master
> >>> > in repository guix.
> >>> >
> >>> > commit 454e7132d6fffb5c9a5ce086ffd1b687416feb83
> >>> > Author: Efraim Flashner <address@hidden>
> >>> > Date:   Sat Dec 1 22:41:19 2018 +0200
> >>> >
> >>> >     gnu: address@hidden: Remove aarch64-linux from supported-systems.
> >>> >     
> >>> >     * gnu/packages/ocaml.scm (address@hidden)[supported-systems]: New 
> >>> > field.
> >>> 
> >>> What's the rationale for this change?
> >>> Debian includes OCaml 4.01 in its arm64 port.
> >>> 
> >>>   https://packages.debian.org/search?arch=arm64&keywords=ocaml
> >>>   
> >>> http://http.us.debian.org/debian/pool/main/o/ocaml/ocaml_4.01.0-5_arm64.deb
> >>> 
> >>>       Mark
> >>
> >> starting phase `configure'
> >> ../gnu/config.guess: unable to guess system type
> >
> > Would it be enough to add Automake as a native input and copy
> > ‘config.guess’ from there?
> 
> Ideally, we shouldn't need 'config.guess' at all.  Normally, it is only
> used if the GNU triplet is not explicitly passed to ./configure.  A few
> years ago, I fixed most instances of this problem by unconditionally
> passing --build=<triplet> to ./configure in the default 'configure'
> phase of gnu-build-system.
> 
>   
> https://git.savannah.gnu.org/cgit/guix.git/commit/?id=3c7d023d6458669c6bfa23bc85e098c91f699892
> 
> However, our OCaml package has a custom 'configure' phase that does not
> pass --build.  I'm not sure if that's because OCaml's configure phase
> doesn't support --build, or if it was omitted because it's not typically
> needed on x86_64.
> 
> * * *
> 
> Anyway, more generally, I hope that we will not get in the habit of
> simply removing systems from 'supported-systems' when builds fail on
> those systems, without investigating and concluding that it would be
> prohibitively difficult to support the software on that system.
> 
> To my mind, it's *good* to see failed builds on other architectures, to
> be reminded of bugs on non-x86_64 systems that should be fixed.  When we
> remove systems from 'supported-systems' without good reason, this is
> somewhat analogous to deleting unfixed bug reports.
> 
> What do you think?
> 
>       Mark

I looked back at the "state of aarch64 in guix¹" blog post, which stated:
OCaml 4.01.0: Doesn't build on aarch64, haven't investigated.

I believe this is the first one that I've marked unsupported in quite a
while, but adding a note as to why it's labeled unsupported would be
best; it gives us the best chance possible at revisiting the decision
later.

¹http://guix.info/blog/2017/state-of-aarch64-on-guix/

-- 
Efraim Flashner   <address@hidden>   אפרים פלשנר
GPG key = A28B F40C 3E55 1372 662D  14F7 41AA E7DC CA3D 8351
Confidentiality cannot be guaranteed on emails sent or received unencrypted

signature.asc
Description: PGP signature