[pypy-dev] pypy tensorflow compatibility

[Christoph Lassner]

Dear pypy devs,

I'm a big fan of pypy's ideas and was very happy to see the great
improvements introduced with the cpyext library in the latest release.
I could out-of-the box compile numpy, scipy and scikit learn, which
allows to use pypy easily for machine learning.

Tensorflow would be another great package to use for deep learning and
other tensor computations. It builds out-of-the box, however it is not
possible to import the SWIG generated python module. Looking at it with
`ldd -r`, several undefined symbols become apparent, hinting at a
missing linked file.

The missing symbols are, for example:

PyPyInt_FromLong
PyPyList_SET_ITEM
PyPyErr_Occurred
_PyPyInstance_Lookup
PyPyDict_SetItem
...

so it looks like the pypy standard library it `missing'. Which file
would have to be linked against to resolve them?

For reference, I also commented on this github issue:
https://github.com/tensorflow/tensorflow/issues/252 .

Thank you for the infos!

Best,
Christoph