[Debian-med-packaging] Bug#780659: Bug#780659: insighttoolkit only built on amd64 and i386
Matthias Klose
doko at debian.org
Wed Mar 25 08:34:24 UTC 2015
On 03/25/2015 08:28 AM, Andreas Tille wrote:
> Hi,
>
> On Tue, Mar 24, 2015 at 04:27:02PM -0400, Paul Novotny wrote:
>> On Tue, 2015-03-17 at 14:53 +0100, Matthias Klose wrote:
>>> Package: src:insighttoolkit4
>>> Version: 4.6.0-3
>>>
>>> insighttoolkit is only built on amd64 and i386, while insighttoolkit3 worked on
>>> some more architectures. any reason for not trying to build that? Sure, it
>>> needs 50g - 100g disk for the build, a 500mb log file, and such. given the
>>> alternative to remove binaries for other architectures before the next Ubuntu
>>> release, I gave it a try ...
>>
>> I looked at this a while back. From what I could tell, some of the tests
>> were failing because GDCM was reading files incorrectly on big-endian
>> systems. The GDCM package is built with its tests off, so I think that
>> is how GDCM can be built for big endian architectures. Upstream GDCM
>> lists big endian support as a "major" missing feature, and unlikely to
>> work [1].
>
> I personally creating packages with tests switch of because we know the
> tests will fail for certain architectures does not make any sense. We
> do not provide packages to fill in gaps in our package x architecture
> matrix but to serve users. We also should not trust on users on those
> architectures since these leaf packages are usually run on mainstream
> architectures and thus the packages with failed test would just fill
> space on our ftpserver with no use at all.
but who do you trust to build on i386 with warnings like:
warning: converting to non-pointer type
warning: iteration 3u invokes undefined behavior
More information about the Debian-med-packaging
mailing list