WIP: PyPy support #521
WIP: PyPy support #521
Conversation
This commit includes a few minor modifications to pybind11 that are needed to get simple hello-world style functions to compile and run on the latest PyPy. Sadly, more complex things are still broken: for instance, creating new types fails with the error message 'TypeError: can't set attributes on type object <..>' when pybind11 tries to set the __module__ attribute. Digging into the pip codebase indicates that it believes that the underlying type is not a heap type, which is incorrect. So this is likely a PyPy bug.
|
Very interesting! How feasible is full PyPy support? Given PyPy's existing incompatibilities with the CPython API and general lean into the other direction (PyPy docs actively discourage C extension modules) I had always imagined this to be an uphill battle. |
|
It was surprisingly easy to get it to compile, and simple functions already work. So let's see -- assuming that the issue with type creation can be fixed, it will already be good enough for many things. I'll file a ticket on the PyPy bug tracker. |
|
PyPy bug report on bitbucket: https://bitbucket.org/pypy/pypy/issues/2434/support-pybind11-in-conjunction-with-pypys |
|
Whoops, I accidentally merged this in a git interactive rebase SNAFU.. Since there is no way to un-merge a PR, I've created a new one at #524 to track progress/discussion. |
This commit includes a few minor modifications to pybind11 that are
needed to get simple hello-world style functions to compile and run on
the latest PyPy.
Sadly, more complex things are still broken: for instance, creating new
types fails with the error message 'TypeError: can't set attributes on
type object <..>' when pybind11 tries to set the
__module__attribute.Digging into the pip codebase indicates that it believes that the
underlying type is not a heap type, which is incorrect. So this is
likely a PyPy bug.