Repository navigation
2 Errors , 10 failures with current master on my ubuntu 12.04 #2429
Description
I run sudo python setup.py test on my ubuntu laptop, I get 2 errors and 10 test failures. For none of them it is obvious to me why is fails for me but passes in Travis.
The errors are from astropy.coordinates
The failures are from:
- astropy.io.fits
- sphinx/ext/tests/test_automodapi.py
- doctest in astropy.table
Activity
bsipocz commentedon May 4, 2014
The test_automodapi.py one is harmless (the code was changed, but the test wasn't), I'll upload a fix (edited: #2430).
Edit2: and fixed the doctest, too in the same PR.mdboom commentedon May 5, 2014
@hamogu: This link seems to be broken: http://bpaste.net/show/258397/ Maybe put it in a gist?
hamogu commentedon May 5, 2014
It worked yesterday and I don't have a local copy of the output (I used
the
setup.py test -boption). I will try to regenerate the errors, butI am not sure if I can today.
hamogu commentedon May 5, 2014
Due to pastebin being broken and by popular demand:
https://gist.github.com/hamogu/13cba11fd487007da64e
astrofrog commentedon May 5, 2014
Actually it looks like pastebin was working again - is it still broken for you?
hamogu commentedon May 5, 2014
It put the stuff on pastebin alright and I could look at it for about 24
h, but if you click on the link right now, than pastebin only displays
and internal server error and not the stuff I saw last night.
astrofrog commentedon May 6, 2014
Ah I see. Are you still seeing failures in the latest master?
hamogu commentedon May 6, 2014
With the current (aobut 1 hour ago) master, I am down to 2 errors, 5 failures.
Let's try pastebin again: http://bpaste.net/show/260765/
astrofrog commentedon May 6, 2014
Some of the failures, for instance the one related to
small_angle_sphere_dist, seems to be due to residual files in your clone, since I don't see a mention of that function anywhere in the code. Can you try testing from a clean clone to see if that helps? (It might not)bsipocz commentedon May 6, 2014
and I don't see any
Gaussian1DModelleftover in my master tree (which I believe, has changed name toGaussian1D)hamogu commentedon May 6, 2014
OK, that was good. I am down to 4 failed:
http://bpaste.net/show/260968
Those are rounding errors or numerical problems now.
astrofrog commentedon May 6, 2014
@embray isn't going to like this - there are a couple of checksum failures in the FITS module. The checksums are almost identical, except for a few characters...
embray commentedon May 7, 2014
Yeah, that makes sense. Right now they're set up just to skip on Windows, but really it should be skipping them on 32-bit platforms (regardless of OS) (the one exception being Python2.6 where on Windows there were actually differences in floating point string formatting that I think resulted in different checksums).
hamogu commentedon May 8, 2014
Is that a loss of precision or are the checksums actually different depending on weather I have a 32 bis or 64 bis system? Should we have a test that is run only on 32 bis systems?
embray commentedon May 8, 2014
It's just slightly different data and hence difference checksums. It's not worth having two sets of tests--it's not important.
hamogu commentedon May 27, 2014
OK, so there is only one test left unexplained and that is just numerical precision. Seems to me that I can close this. Very few people have computers as old as mine that run on 32 bit systems, so I don't think that we need to fix this to the last digit at this point.