Repository navigation
Closed
Description
Hello,
i've just uploaded numpy/1.21.5 and the test suite produced some failures on a few architectures we support. Please note you're going to find 2 sets of test results, because we're building for both Python 3.9.9 and Python 3.10.1.
armel
failures summary:
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_half.py::TestHalf::test_half_fpe
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_numeric.py::TestBoolCmp::test_float
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_numeric.py::TestSeterr::test_divide_err
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_numeric.py::TestSeterr::test_errobj
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_numeric.py::TestFloatExceptions::test_floating_exceptions
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_numeric.py::TestFloatExceptions::test_warnings
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_ufunc.py::TestUfuncGenericLoops::test_unary_PyUFunc_O_O_method_full[arccosh]
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestRemainder::test_float_remainder_overflow
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestLog2::test_log2_special
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestSpecialFloats::test_exp_exceptions
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestSpecialFloats::test_log_values
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestSpecialFloats::test_sincos_values
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestSpecialFloats::test_square_values
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestSpecialFloats::test_reciprocal_values
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestFPClass::test_fpclass[-4]
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestFPClass::test_fpclass[-2]
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestFPClass::test_fpclass[-1]
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestFPClass::test_fpclass[1]
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath_complex.py::TestClog::test_special_values
FAILED tmp/usr/lib/python3/dist-packages/numpy/lib/tests/test_function_base.py::TestMedian::test_empty
i386
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestRemainder::test_float_remainder_overflow
FAILED tmp/usr/lib/python3/dist-packages/numpy/random/tests/test_generator_mt19937.py::TestRandomDist::test_pareto
mipsel
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_numeric.py::TestBoolCmp::test_float
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestFPClass::test_fpclass[-4]
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestFPClass::test_fpclass[-2]
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestFPClass::test_fpclass[-1]
FAILED tmp/usr/lib/python3/dist-packages/numpy/core/tests/test_umath.py::TestFPClass::test_fpclass[1]
Activity
seberg commentedon Dec 21, 2021
Hmmm, quick scan, most failures seem to be related to floating point errors and I am not sure how much trouble we should go to try and get those right...
This one worries me a bit more:
It happens in the line:
But I wonder if it is something weird to do with
-NaNbeing incorrectly handles as well? I would guess at vectorization, but I doubt that makes sense for mipsel or armel.charris commentedon Dec 21, 2021
Armel doesn't have an FPU, correct? Are the instructions trapped and emulated, or is it library based? I'm trying to see how much can be attributed to the library (if used) and the reporting of exceptions (if available).
sandrotosi commentedon Dec 21, 2021
i must admit i'm not really knowledgable on all the specs and quirks of these "uncommon" archs. there's a main debian archs list here which contains references to each of them, here's the one for armel
WarrenWeckesser commentedon Dec 22, 2021
FYI (cross-reference): The i386 failure of
test_paretolooks like the same issue as #18387sandrotosi commentedon Mar 19, 2024
closing outdated issues i reported long ago