control: tags -1 + wontfix > The tests are run during build like this: > > python$$v -c "import sys ; sys.path.insert(0, > './tmp/usr/lib/python3/dist-packages/') ; import numpy; numpy.test(verbose=5)" > > but numpy.test returns a boolean True/False to indicate whether it > passed or failed, not raising an exception or anything else that would > cause python to exit with a non-zero code. Simple, untested, patch > attached.
sadly it is not that simple. With just the recent upload of numpy/1.8.3 there are 4 release architectures where tests are failing: https://github.com/numpy/numpy/issues/16046 https://github.com/numpy/numpy/issues/16047 https://github.com/numpy/numpy/issues/16048 https://github.com/numpy/numpy/issues/16049 If i enable build failures for test failures, it means the new version of numpy will never transition to testing until those tests failures get fixes; given the relative small footprint of those archs, it's unlikely this will be a timely process, so it could pass weeks, if not months before they get fixed, if ever. This is not an acceptable process. We could skip the failing tests, but then at every new upload we're have to remove the skip instructions and re-assess what tests are failing: it will just be extra work and obtain the same result as of now. For all that's said above, i'm tagging this but as wontfix, but leave it open for others to express their opinion Regards, -- Sandro "morph" Tosi My website: http://sandrotosi.me/ Me at Debian: http://wiki.debian.org/SandroTosi Twitter: https://twitter.com/sandrotosi