importing problems with new C functions?

[wkerzendorf]

Hello

@orbitfold @wkerzendorf

I'm having trouble with import (this is with latest master branch of code):

In [1]: from tardis.tests import montecarlo_test_wrappers
WARNING: ConfigurationDefaultMissingWarning: Requested default configuration file /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tardis.cfg is not a file. Cannot install default profile. If you are importing from source, this is expected. [tardis._astropy_init]
ERROR: ImportError: dlopen(/Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so, 2): Symbol not found: _compute_distance2electron
  Referenced from: /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so
  Expected in: flat namespace
 in /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so [IPython.core.interactiveshell]
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-1-155ec909d77a> in <module>()
----> 1 from tardis.tests import montecarlo_test_wrappers

ImportError: dlopen(/Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so, 2): Symbol not found: _compute_distance2electron
  Referenced from: /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so
  Expected in: flat namespace
 in /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so

This problem means that the test suite doesn't work for me just now. Since it's complaining about computedistance2electon, I wonder if it's something to do with having moved those functions to C? Any suggestions?

[orbitfold]

Hello. Delete build directory and try again? Seems to me that montecarlo_test_wrappers.pyx was not recompiled. I think this happens because montecarlo_test_wrappers.pyx does not know it depends on montecarlo.pyx and thus does not recompiles when montecarlo.pyx changes.

[ssim]

Hello - hmmm no, I tried "nuking" the build directory

rm -r /Users/stuartsim/Tardis/tardis/build

but then still get the same problem after first doing:

python setup.py build

then

from tardis.tests import montecarlo_test_wrappers
ERROR: ImportError: dlopen(/Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so, 2): Symbol not found: _compute_distance2electron
  Referenced from: /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so
  Expected in: flat namespace
 in /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so [IPython.core.interactiveshell]
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-2-155ec909d77a> in <module>()
----> 1 from tardis.tests import montecarlo_test_wrappers

ImportError: dlopen(/Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so, 2): Symbol not found: _compute_distance2electron
  Referenced from: /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so
  Expected in: flat namespace
 in /Users/stuartsim/Tardis/tardis/build/lib.macosx-10.9-x86_64-2.7/tardis/tests/montecarlo_test_wrappers.so

In [3]: 

Is there possibly something odd about paths?

[wkerzendorf]

@orbitfold I get the same problem on my mavericks 10.9 machine

[orbitfold]

Hard to say why this happens. I don't own a Mac to test on. I don't have a similar issue on Linux. Symbol not found means that montecarlo_test_wrappers.so was not linked against anything containing compute_distance2electron which in this case is cmontecarlo.c. So the thing to check is if montecarlo_test_wrappers.pyx is linked against cmontecarlo.c.

[orbitfold]

Yet actual Tardis simulation runs fine right?

[wkerzendorf]

Nope. It's not importable. It crashes when I do 'from tardis import montecarlo'

[wkerzendorf]

@ssim @orbitfold I've converted the issue into a PR and attached a fix that @orbitfold and I found.

[wkerzendorf]

@orbitfold @ssim - ready for merge?

[ssim]

I have not tried it out, but it passes ok so I'm happy for us to merge - I will then try it when I get some time? So @wkerzendorf do merge if you are happy.

[ssim]

Excellent - and I can now confirm that, at least on my local machine, the tests now pass. I will try a clean setup shortly.

On 5 Jun 2014, at 16:52, Wolfgang Kerzendorf notifications@github.com wrote:

Merged #152.

—
Reply to this email directly or view it on GitHub.

[orbitfold]

For the record - the problem was with how clang does inlining (well they claim it is the correct behavior). As can be read about here http://clang.llvm.org/compatibility.html#inline