fitsio requires glibc v2.14

[kadrlica]

The fitsio v0.9.8 dependency of easyaccess appears to have been built with glibc v2.14. Scientific Linux 6.7 (Fermilab cluster operating system), only has glibc v2.12 and I get the following error trying to import fitsio:

>>> import fitsio
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/s1/kadrlica/.conda/envs/fitsio/lib/python2.7/site-packages/fitsio/__init__.py", line 9, in <module>
    from . import fitslib
  File "/home/s1/kadrlica/.conda/envs/fitsio/lib/python2.7/site-packages/fitsio/fitslib.py", line 29, in <module>
    from . import _fitsio_wrap
ImportError: /lib64/libc.so.6: version `GLIBC_2.14' not found (required by /home/s1/kadrlica/.conda/envs/fitsio/lib/python2.7/site-packages/fitsio/_fitsio_wrap.so)

This issue cannot be solved as easily as a:

conda install glibc=2.19 -c asmeurer

because conda was built with the older version of glibc and installing a newer version leads to errors and seg faults like:

/home/s1/kadrlica/.conda/envs/fitsio/bin/.openssl-post-link.sh: line 7: 1460654 Segmentation fault      $PREFIX/bin/.openssl-libcrypto-fix $PREFIX
Error: Error: post-link failed for: openssl-1.0.2j-0 

The proposed solution would be to re-build fitsio with an older version of glibc.

[kadrlica]

The glibc error was triggering a fitsio import error, which in turn was triggering an import error in easyaccess. I've built fitsio with gclib v2.12 and this fixes these problems. The conda packages for fitsio can be found here.

[mgckind]

I'm looking at this right now, we have 2 options:

1. is to build all versions (py 2.6, 2.7, 3.4, 3.5 and 3.6 at least) for linux/os-x using glibc v.2.12 and use that as default, which might need to be you (then I can copy the packages) as I all glibc versions I have are newer.
2. Install glibc from source somewhere and build it there
3. Use a recent version of glibc when this happens build fitsio using pip install that comes with conda

I'm inclined for 3 but 1 is good too as it might not need to be done to often

[kadrlica]

I'm not sure if I understood options 2 & 3. There are several conda distributions of glibc, but I've been nervous that they might break other system components (I think I ran into an issue with curl?).

I'm happy to build and test.

[mgckind]

But, didn't you say that using conda to install glibc might not wok?

[kadrlica]

I didn't have success before, but I think it is the "right" way to do things. We would need to make sure everything was compatible with the same glibc version.