Implementation of Voronoi RT

[astrofrog]

@bluescarni - just for info, this is the PR with all the functionality we've worked on so far (makes it easier to review the whole thing).

[astrofrog]

Note to self: need to disable the PDA for this type of grid. Need to not define cell_width for example.

[astrofrog]

@bluescarni - let's see if the tests pass here!

[bluescarni]

@astrofrog mhmm looks like the c++ compiler is not found?

[astrofrog]

@bluescarni - I've tried to add a line to .travis.yml to get it to install the C++ compiler

[bluescarni]

@astrofrog seems like it is not enough yet. How about adding the g++ compiler explicitly a-la sudo apt-get install g++? Otherwise it would seem some kind of problem on the Python side - to be pedantic it should not try to call gcc, but g++ when compiling c++ code.

[astrofrog]

@bluescarni - ok, trying that :)

[bluescarni]

@astrofrog thanks, this looks better. It fails now in the tests when it tries to unpickle the neighbours tables computed with qhull and used for sanity checks.

My understanding is that the class OrderedDict, used by pickle, was introduced in python 2.7 but the travis instance has python 2.6. Maybe I could store the tables for comparison in another format readable by older Python versions?

[astrofrog]

@bluescarni - it might be good to store it in a different format, since pickles don't really have any guarantees on long-term storage. Can you see if there is a different way of storing it as e.g. an ASCII table or FITS table?

[astrofrog]

Alternatively, you could store it as an HDF5 table (astropy's Table.write can write to hdf5)

[astrofrog]

Note to self - we need to add a volumes property to VoronoiGrid

[astrofrog]

@bluescarni - I've been testing this out extensively this week, and it's working great! I just made a number of small changes above that you can take a look at if you like :)

[bluescarni]

@astrofrog glad to hear it works :)

I went through the latest changes and they look good to me. I think we still need to figure out how to use the extra walls support (sphere, cylinder, etc.), but for now I guess it will do.

[bluescarni]

@astrofrog It looks like the tables now work ok, but a test fails because the hyperion_vor executable is not found. Indeed, I need to build it explicitly on my machine as it is not included in the default build configuration.

[astrofrog]

@bluescarni - ok, I tried to add that to see if it works!

[bluescarni]

@astrofrog cheers! was about to open another PR, you beat me to it.

[bluescarni]

@astrofrog tests are passing on 3 setups out of 5.

There is a strange out-of-memory failure on Python 3.4, which seems unrelated to the Voronoi part.

And there is a problem when deepcopying the neighbours table with Numpy 1.5 on Python 2.7. It complains that AttributeError: 'list' object has no attribute 'name'. I suppose we could avoid deepcopying (and return a reference to the internally-stored table instead), but it does not feel like a good solution.

[astrofrog]

@bluescarni - maybe we can just drop support for numpy 1.5, most installations now come with something more recent.

[astrofrog]

Is the deepcopying of tables something you can reproduce outside of Hyperion? If so, can you report it on the astropy issue tracker?

[bluescarni]

@astrofrog I am going to check about the deepcopying of the tables, I need to install locally a Numpy 1.5 first. If that turns out to be the problem we can get it fixed upstream then.

[astrofrog]

@bluescarni - ok, sounds good!

[astrofrog]

@bluescarni - it seems the table issue is the only one that remains after I restarted the jobs (the memory issue was intermittent)

[astrofrog]

@bluescarni - I removed some duplicate code in 941ad11e0a72ea59948c452eeee8655820dd9bde (this from what I had originally added) but other than that, this looks good to go!

[astrofrog]

All right, let's merge!

[astrofrog]

Thanks @bluescarni!