On Thu, Aug 18, 2016 at 10:02 PM, Yaroslav Halchenko <deb...@onerussian.com>
wrote:

>
> On Thu, 18 Aug 2016, Aaron M. Ucko wrote:
>
> > Source: singularity-container
> > Version: 2.1.2-1
> > Severity: important
> > Justification: fails to build from source
>
> > Builds of singularity-container for architectures other than x86 Linux
> > have been failing at the configuration stage.  The non-Linux builds
> > complain that various CLONE_* flags, most crucially CLONE_NEWNS, are
> > unavailable; I presume these platforms are a lost cause.  As for Linux
> > on non-x86, the build system complains that the architecture "is not
> > supported"; I haven't checked whether that's simply a case of
> > excessive conservatism.
>
> > Could you please take a look and restrict the package's official
> > Architecture accordingly?
>
> I thought that the idea generally is to not restrict by default and
> possibly to let interested in porting to look at it. Policy 5.6.8
> states "Specifying a list of architectures or architecture wildcards
> other than any is for the minority of cases where a program is not
> portable or is not useful on some architectures".
>
> Theoretically I think singularity could be ported for other
> architectures (CCing upstream for clarification), not that me or
> upstream is going to embark on such a voyage ATM ;)
>
> If there is somewhere policy/recommendation that I should better drop
> from any to a restricted list of architectures, please let me know, and
> I will do so for the next release.
>

That is a good point regarding non-Linux builds. At present I am not
checking for the clone() or unshare() system calls, as they have been on
Linux for quite some time so I didn't think of them not being available,
but they are Linux specific system calls (as are the CLONE_* flags which I
am checking for). So it sounds like I need to add some checks for clone()
and unshare() and bomb out early with a reasonable error message about an
unsupported platform before checking for the flags.

As far as architecture support, Singularity *should* build on non-x86
architectures just fine, but my ability to port and test are limited to
x86. I am aware that IBM has successfully ported and used Singularity on
Power, but I've never gotten any patches so I assumed it just worked. Are
the failed build logs for non-x86 available?

 Thank you!


-- 
Gregory M. Kurtzer
High Performance Computing Services (HPCS)
University of California
Lawrence Berkeley National Laboratory
One Cyclotron Road, Berkeley, CA 94720

Reply via email to