Re: [Numpy-discussion] missing FloatingPointError for numpy on cygwin64
Le 01/02/2015 08:48, David Cournapeau a écrit : > The first > place I would look for cygwin-related FPU issues is there: > https://github.com/numpy/numpy/blob/master/numpy/core/setup.py#L638 The pointer is a good one. Thanks to you, I have found the problem (a wrong numpy-specific fenv.h was included in the math routines instead of the system-wide version) and submitted a patch to the bug tracker. With the patch, almost all tests pass on cygwin64! ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
Re: [Numpy-discussion] missing FloatingPointError for numpy on cygwin64
Le 01/02/2015 08:48, David Cournapeau wrote: > Could you report the value of `sys.platform` on cygwin64 ? The first > place I would look for cygwin-related FPU issues is there: > https://github.com/numpy/numpy/blob/master/numpy/core/setup.py#L638 sys.platform is simply cygwin, just as on cygwin32. It should not be too hard to support cygwin64: most tests pass in the testsuite, almost all failing ones are related to exceptions or warnings that are not raised as in my first observation. However, debugging this seems to be beyond my skills! ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
Re: [Numpy-discussion] missing FloatingPointError for numpy on cygwin64
Hi Sebastian, I think you may be one of the first person to report using cygwin 64. I think it makes sense to support that platform as it is becoming more common. Could you report the value of `sys.platform` on cygwin64 ? The first place I would look for cygwin-related FPU issues is there: https://github.com/numpy/numpy/blob/master/numpy/core/setup.py#L638 David On Sat, Jan 31, 2015 at 9:53 PM, Sebastien Gouezel < sebastien.goue...@univ-rennes1.fr> wrote: > Dear all, > > I tried to use numpy (version 1.9.1, installed by `pip install numpy`) > on cygwin64. I encountered the following weird bug: > > >>> import numpy > >>> with numpy.errstate(all='raise'): > ...print 1/float64(0.0) > inf > > I was expecting a FloatingPointError, but it didn't show up. Curiously, > with different numerical types (all intxx, or float128), I indeed get > the FloatingPointError. > > Same thing with the most recent git version, or with 1.7.1 provided as a > precompiled package by cygwin. This behavior does not happen on cygwin32 > (I always get the FloatingPointError there). > > I wonder if there is something weird with my config, or if this is a > genuine reproducible bug. If so, where should I start looking if I want > to fix it? (I don't know anything about numpy's code) > > Sebastien > > ___ > NumPy-Discussion mailing list > NumPy-Discussion@scipy.org > http://mail.scipy.org/mailman/listinfo/numpy-discussion > ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
[Numpy-discussion] missing FloatingPointError for numpy on cygwin64
Dear all, I tried to use numpy (version 1.9.1, installed by `pip install numpy`) on cygwin64. I encountered the following weird bug: >>> import numpy >>> with numpy.errstate(all='raise'): ...print 1/float64(0.0) inf I was expecting a FloatingPointError, but it didn't show up. Curiously, with different numerical types (all intxx, or float128), I indeed get the FloatingPointError. Same thing with the most recent git version, or with 1.7.1 provided as a precompiled package by cygwin. This behavior does not happen on cygwin32 (I always get the FloatingPointError there). I wonder if there is something weird with my config, or if this is a genuine reproducible bug. If so, where should I start looking if I want to fix it? (I don't know anything about numpy's code) Sebastien ___ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion