如何在PyPy中导入.so模块？

Question

I am green to PyPy, and I use PyPy1.9 to interpret my Python code.

As you know, the CPython interpreter can import a module packaged in a .so file, which is compiled from C/C++ codes.

But PyPy1.9 interpreter tell me he can not find the module in a .so file, my code is a little slow and I want to optimize the performance, help!!

Answer 1

First, CPyExt has been in PyPy since… I forget the exact version, but a couple years now, well before 1.9.

Basically, all you have to do is build and install the module, just as you do with CPython.

Make sure you do the install with PyPy, not with CPython. Besides the fact that each Python installation on your system has a separate site-packages directory, CPyExt requires modules to be built against the PyPy headers and libs, not the CPython ones.

If you're installing manually, just do pypy setup.py install instead of python setup.py install . If you're using pip or easy_install , the PyPy versions of those tools are usually named something like pip_pypy and easy_install_pypy .

So, if you've installed pip , it should be as easy as:

pip_pypy install foo

(Possibly with a sudo , of course.)

If you haven't installed pip , install it for PyPy the same way you would for CPython. For example:

curl -O https://raw.github.com/pypa/pip/master/contrib/get-pip.py
pypy get-pip.py

If you're using a binary installer, then obviously you have to get the binary installer for PyPy-1.9. And, if there isn't one, you can't use a binary installer. (Just like if you wanted to install foo for CPython 2.7 on 32-bit Windows and they only had a 2.6 package or a 64-bit package.)

---

One last note: There are some packages that do not yet work with CPyExt. Most notably, numpy and everything that depends on it, and anything else that requires binding Fortran code. Some of these packages will check for CPython (or check for non-PyPy?) and give you an early error. Others may try to build and fail. Others may successfully build but not work in some cases (eg, some of them fail in any program with more than 1 thread, because they try to get fancy with the GIL). But the easiest thing to do is try it and see.

Answer 2

For PyPy you would need to utilize something like CFFI , which uses Foreign Function Interface to access library code (see the section on Loading Libraries for more information).

Edit : I stand corrected by the above answer. I'll leave this answer here though to have an alternative for accessing libraries and simply accept justified downvotes.