On December 22, 2015 01:18:46 PM Sebastiaan Couwenberg wrote:
> > #724711 insighttoolkit4: Drops architecture support
> > - upstream, quite a few tests fail on e.g. powerpc and armhf
>
> This is relevant for OTB, because ITK4 limits the architectures to amd64
> & i386. It should be available on all the modern release architectures
> at least (specifically arm*). s390x and powerpc should also be more than
> powerful enough for insighttoolkit4, but I suspect insighttoolkit4
> doesn't support big endian architectures.
Upstream ITK does aim to support big-endian; that's not the issue.
> When upstream developers are unwilling or unstable to fix architecture
> specific test failures, I choose to ignore the test failures instead of
> excluding the architecture entirely.
I think we need to take the view that a test failure *IS* a build failure.
If you provide a package that only passes the weak test of syntactical
correctness (i.e. it builds) but may fail to produce a correct result (i.e. it
fails a test): that is dangerously misleading. I think Debian should adhere
to a higher standard that the resulting package is correct (to the best of our
knowledge).
In the specific case of ITK, it comes down to lack of people power. I spent
literally years trying to get it to build correctly on all architectures at
the same time long enough to transition to testing. I failed. Upstream had
no visibility on non-Intel architectures (recently it changed with interest in
the Rasberry Pi), so I tried recruiting people to help out troubleshooting
the issues or to run nightly dashboard builds on non-Intel architectures. I
got, I think, two builds running for a bit; but that effort fizzled too.
Ultimately, I took the hard decision that (a) I couldn't get it building
everywhere by myself and (b) having ITK v4 in Debian for the mainstream
architectures is better than not having ITK v4 in Debian at all. I'm not
happy with being forced to make that decision. But I don't think it was the
wrong one. In all the time since then I've had exactly one person complain
that his usage of ITK (on Sparc) was not supported in Debian. All the other
complaints about the decision were on philosophical grounds ("Debian should
release everywhere"). I think the philosophy is a good goal, but reality got
in the way this time.
I don't know what OTB is. If arm is important to you, then I second Gert's
suggestion that folks volunteer their arm machines for the ITK dashboard and
help upstream get arm in shape. As I mentioned, there is interest in the
RasPi upstream so you may get traction here. I have set up ITK nightly builds
for the x86 and amd64 architectures, and would be happy to assist anyone who
wants to do the same.
The dashboard is the first step, and largely automatic once set up. The bigger
issue is that folks are needed to address the issues that crop up. That means
triaging and providing patches, etc. That is a big time commitment, but a
necessary one, IMHO, before we enlarge the set of architectures for ITK.
Best,
-Steve