version dependency/tests on numexpr -- truediv etc are NA in numexpr 1.4.2

[yarikoptic]

while building for Ubuntu 12.04, bulk of tests fail due e.g.
ValueError: Unknown keyword argument 'truediv' from numexpr and AssertionError: Did not use numexpr as expected in test_expressions

Full log http://neuro.debian.net/_files/_buildlogs/pandas/0.13.0/pandas_0.13.0-2~nd12.04+1_i386.build

[jreback]

I don't think ever run on < Numexpr 2
which has been out since 2011

shouldn't this error have shown up before?

[jtratner]

want me to put up a PR that:

1. Checks >= 2.0 for numexpr
2. Sets can use numexpr setting to False if it's < 2.0
3. Skips tests (probably in setup) in test_expressions

?

[jreback]

would be gr8 @jtratner

[yarikoptic]

might be worthwhile so we could provide builds for elderly ubuntus, may be without the benefits of recent numexpr. Debian wheezy has numexpr 2.0.1

[jreback]

maybe also change one of the Travis builds (maybe 2.6 one

[jreback]

to use Numexpr 1.4.2

[jtratner]

@yarikoptic can you try building this branch and seeing if it works for you - I just added the division future import to all core files and removed the truediv kwarg from expressions. I believe this will actually make numexpr 1.4.2 work:

https://github.com/jtratner/pandas/tree/handle-old-numexpr

[yarikoptic]

On Fri, 03 Jan 2014, Jeff Tratner wrote:

[1]@yarikoptic can you try building this branch and seeing if it works for
you - I just added the division future import to all core files and
removed the truediv kwarg from expressions. I believe this will actually
make numexpr 1.4.2 work:

[2]https://github.com/jtratner/pandas/tree/handle-old-numexpr

Hi Jeff,

Either I screwed up somewhere (I always leave that possibility even to obvious
actions), or they had no complete healing effect (I think one of the
failles is gone), I still observe the same failures after cherry-picking

0001-Add-division-future-import-everywhere.patch
0002-remove-explicit-truediv-kwarg.patch

e.g. building on ubuntu 12.04 with numexpr 1.4.2-1.2build2

ERROR: pandas.computation.tests.test_eval.TestAlignment.test_align_nested_unary_op('numexpr', 'python')

Traceback (most recent call last):
File "/usr/lib/python2.7/dist-packages/nose/case.py", line 197, in runTest
self.test(*self.arg)
File "/tmp/buildd/pandas-0.13.0/debian/tmp/usr/lib/python2.7/dist-packages/pandas/computation/tests/test_eval.py", line 794, in check_align_nested_unary_op
res = pd.eval(s, engine=engine, parser=parser)
File "/tmp/buildd/pandas-0.13.0/debian/tmp/usr/lib/python2.7/dist-packages/pandas/computation/eval.py", line 209, in eval
ret = eng_inst.evaluate()
File "/tmp/buildd/pandas-0.13.0/debian/tmp/usr/lib/python2.7/dist-packages/pandas/computation/engines.py", line 52, in evaluate
res = self._evaluate()
File "/tmp/buildd/pandas-0.13.0/debian/tmp/usr/lib/python2.7/dist-packages/pandas/computation/engines.py", line 100, in _evaluate
truediv=self.expr.truediv)
File "/usr/lib/pymodules/python2.7/numexpr/necompiler.py", line 624, in evaluate
context = getContext(kwargs)
File "/usr/lib/pymodules/python2.7/numexpr/necompiler.py", line 436, in getContext
raise ValueError("Unknown keyword argument '%s'" % map.popitem()[0])
ValueError: Unknown keyword argument 'truediv'

Yaroslav O. Halchenko, Ph.D.
http://neuro.debian.net http://www.pymvpa.org http://www.fail2ban.org
Senior Research Associate, Psychological and Brain Sciences Dept.
Dartmouth College, 419 Moore Hall, Hinman Box 6207, Hanover, NH 03755
Phone: +1 (603) 646-9834 Fax: +1 (603) 646-1419
WWW: http://www.linkedin.com/in/yarik

[jtratner]

That's very strange. It shouldn't be adding truediv kwarg anywhere and I
ran the suite locally against numexpr 1.4.2. I will run again and try to
figure out what's up.

That said, I realize we may not be ablr support numexpr 1.4.2 because it
will segfault on certain division edge cases. This was only fixed a few
months ago...

I need to check this and the resolution may be to force no acceleration
under a specific version number of numexpr