Windows test failures #4866
Comments
|
Only four test failures on latest master However |
|
The dtype failures are easy fixes. However the two other plotting failures only happen when running the full test suite, looks like some weirdness with matplotlib. |
|
@Komnomnomnom ok...all non-plotting tests look ok @cpcloud shouldn't these figures be closed after each test? |
|
@Komnomnomnom can you update with I am guessing only the plot failures remaining? (after you update from master) |
|
@jreback @Komnomnomnom There's a bit of code rot in the plotting tests....I'm finding things like: self.assert_(l1.get_color(), l2.get_color())(this will always test There's a couple of other things I'll clean up as well. PR to arrive later today |
|
And yes...the |
|
I was recently trying to build Pandas from source on Windows and actually ran into compile errors with objToJSON.c. Win 7 X64, VS 2008 Compiler. Not sure if these were resolved already. If you guys have lots of experience making the builds work on windows a walk-through would be great. 😄 Or having the nightly builds produce binaries for windows more frequently would be great too. ✌️ I used http://blog.victorjabur.com/2011/06/05/compiling-python-2-7-modules-on-windows-32-and-64-using-msvc-2008-express/ tutorial to help me get started building on windows. Any other advice would be great. |
|
@dragoljub you should be able to update to master and these errors are fix, IOW it should build (as of yesterday master)...so need failry new |
|
@dragoljub the only advice I can give is to make sure to use the same compiler as that used to build your Python binary (and numpy, cython etc), otherwise even if your build succeeds you can run into unexpected runtime issues (crashes and segmentation faults). Usually this means working with VS2008 but if working with a different Python distribution (e.g. pythonxy) this can mean using MinGW to compile, or if using Python 3 then VS2010. I'm not at all seasoned in building on windows, but this has been my experience. As jreback mentioned the build error you're encountering has been fixed by #4778. There's some more detail there of some woes I encountered building on windows. |
|
@jreback, @cpcloud I can confirm that the non-dtype plotting test failures are still occurring: |
|
ok well now at least we can see what's different... @Komnomnomnom What is the output of the following code on your Windows box ioff()
fig, ax = subplots(1, 1, figsize=(16, 8))
t = linspace(0, 2 * pi, 1000)
x = sin(t)
ax.plot(t, x)
print(fig.get_size_inches()) |
|
@cpcloud I don't think the tests fail unless the whole testsuite is run. I'm gonna try increasing the ram on the VM, in case that is the issue. |
|
Which I will do once it shuts down .... "Installing update 17 of 70" sigh |
|
hm.....what's the backend you're using? |
|
Virtualbox? |
|
no i meant plotting backend 😄 |
|
No idea, whatever's set up by enthought and / or the tests I presume. Will check when I'm back in. |
|
Well, upping the RAM helped the tests run faster, the whole testsuite uses about a gig of RAM. Tests still fail BUT @cpcloud I think you're onto something: In [7]: paste
from matplotlib.pyplot import ioff, subplots
from numpy import linspace, sin, pi
ioff()
fig, ax = subplots(1, 1, figsize=(16, 8))
t = linspace(0, 2 * pi, 1000)
x = sin(t)
ax.plot(t, x)
print(fig.get_size_inches())
## -- End pasted text --
[ 12.85 8. ]FYI In [3]: matplotlib.get_backend()
Out[3]: 'WXAgg'
In [4]: matplotlib.__version__
Out[4]: '1.3.0'I'm not that familiar with matplotlib but, given that the tests pass when run in isolation, C:\Users\kieran\Documents\GitHub\pandas [master +5 ~0 -0 !]> nosetests pandas.tests.test_graphics
................................................................
----------------------------------------------------------------------
Ran 64 tests in 254.672s
OKmy best guess is that when running the full testsuite From the matplotlib docs: "If you use the |
|
@cpcloud FYI amending the import of matplotlib in try:
import matplotlib as mpl
mpl.use('Agg', warn=False)
import matplotlib.pyplot as plt
except:
raise nose.SkipTestfixes the test failures during the full test run. |
|
awesome i'll put up a PR, since these should all be using |
|
@cpcloud @Komnomnomnom are these still recurring? close? |
|
not sure over here ... think this is just a windows issue |
|
Looks like 7b7474e has fixed the remaing plotting issues (thanks @cpcloud!). BUT, got another one now: #4992 fixes. |
|
btw - the tests on my computer sometimes complain that using the pylab interface to generate figures means that they will not be garbage collected. I don't know if this is an internals thing or a test thing, but it might be useful to use the oop interface to generate figures and such. |
|
@jtratner I tried to get rid of those by closing figures on |
|
@cpcloud leaving this open then? or new issue for the plots? (I merged @Komnomnomnom fixes this morning) |
|
I'll open a new issue about the gc stuff |
Windows 7 - 64-bit
Enthought Python 2.7 - 32 bit
VS2008 compiled cython+pandas
The text was updated successfully, but these errors were encountered: