2

Let's say I'm trying to wrap vector.

module/__init__.pxd:

from libcpp.vector cimport vector from libc.stdint cimport uint32_t cdef class MyVector: cdef vector[uint32_t]* thisptr

module/__init__.pyx:

from libc.stdint cimport uint32_t from libcpp.vector cimport vector from cython.operator cimport dereference as deref cdef class MyVector: # the field declaration is in *.pxd def __cinit__(self): self.thisptr = new vector[uint32_t]() self.thisptr.push_back(42) def __dealloc__(self): del self.thisptr self.thisptr = <vector[uint32_t]*> NULL def mysize(self): return self.thisptr.size() def myget(self): return deref(self.thisptr)[0]

module/__init__.pyxbld and run.pyxbld:

def make_ext(modname, pyxfilename): from distutils.extension import Extension return Extension( name=modname, sources=[pyxfilename], language='c++' )

run.pyx:

from module cimport MyVector cdef main(): obj = MyVector() print(obj.thisptr.size()) # 1 print(obj.mysize()) # 1 print(obj.myget()) # 42 main()

test.py:

import pyximport pyximport.install() import run

When I run test.py, it crashes with the following traceback:

Traceback (most recent call last): File "/usr/lib64/python3.4/site-packages/pyximport/pyximport.py", line 210, in load_module mod = imp.load_dynamic(name, so_path) File "module/__init__.pxd", line 5, in init run (/home/pastafarianist/.pyxbld/temp.linux-x86_64-3.4/pyrex/run.cpp:918) cdef class MyVector: AttributeError: 'module' object has no attribute 'MyVector'

The same code works if I move module/__init__.pyx to module.pyx and module/__init__.pxd to module.pxd. What am I missing and how do I fix it?

A few other related questions.

  1. Is there any way to expose a templated wrapper to Cython code, so that I can have MyVector[uint16_t] without needing to write another wrapper?
  2. Am I correct in adding a pyxbld file for each source file which interacts with C++ code? Is this redundant? (I like the convenience of pyximport and I don't want to recompile the code every time manually while I'm still trying to get it to work.)
  3. How do I compile module to a standalone extension? What should setup.py look like?
  4. In the above code I never used deref before calling C++ methods. How does Cython understand where I mean -> and where I mean .?

I would appreciate help with any of these questions.

UPD: actually, I was wrapping Google's sparsehash, and I have figured out a way to do what I wanted, but it looks like black magic. I would still appreciate a clarification of what was going on with this error and how to write Cython wrappers properly.

Improve this question

2 Answers 2

Reset to default
2

Your main issue is that the code in __init__.so (generated from __init__.pyx) does not get executed unless you also have __init__.py (which does not get executed in this case). Therefore, you can solve the problem by making an empty __init__.py file. This all feels like a messy edge-case. In your original code, the MyVector class would never be added to the module and so it fails to find it with the error you describe.

Short answers to your other questions are:

  1. no - unfortunately there is no good way to build a templated wrapper class. You don't need to build wrapper classes for anything you only use in Cython though.

  2. and

  3. I think these would end up being pretty significant answers in their own right and so I'm skipping them. setup.py seems to be preferred to pyximport in general, and good examples of how to use it are in the documentation.

  4. Cython keeps track of the types of objects internally. It's fairly straightforward for it to substitute -> whenever it knows it's using a pointer and . otherwise, so as you say you don't have to do so yourself.

Improve this answer
Sign up to request clarification or add additional context in comments.

2 Comments

Thank you. I'd like to note that I wanted a wrapper class for a more pythonic interface, even inside a Cython module. Could you provide an example of a simple Cython wrapper? I've modeled the solution that I referenced in the UPD after GSL (that's also how I figured out about __init__.py), but it's too complex, and I've probably overengineered something.
It looks OK - you've fitted it to work broadly like a dict which I think is as Pythonic as you can get. Unfortunately, Cython does involve wrapping every function independently, which gives you quite long interfaces, so I'd say next time just cut it down to the bits you need. I can't see a whole lot else.
-1

You should consider using 'swig'. This tool will create all the code you need for interacting between C++ and Python. Link to swig website.

I used swig in a complex project and it worked very nicely.

Anyway, you can use the swig created code as a reference if you don't want to use this framework in your code.

Improve this answer

2 Comments

I need to use the wrapped class from Cython. Is it possible to integrate SWIG-wrapped classes with it?
I don't see why not. Those class will be available to regular Python code. Plenty of examples online.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.