Build failure on CentOS using cmake: return-statement with a value, in function returning 'void'

[classfire]

Hello,
I'm trying to install boost.numpy on my CentOS 6.6 platform using cmake and python 3.5.0 and numpy 1.10.0:

cmake -DPYTHON_LIBRARY=/usr/local/lib/libpython3.5m.so -D Boost_NO_BOOST_CMAKE=ON ../

-- Configuring project "BoostNumpy"
:
:
ImportError: No module named 'sphinx'
-- - Sphinx not found, "make html" will do nothing at all.
-- Project "BoostNumpy" configured.
-- + Now, type-in "cd /root/BoostNumpy-master/build && make && make test && make html && make install"!
-- + "BoostNumpy" will be installed in /usr/local.
-- Configuring done
-- Generating done
-- Build files have been written to: /root/BoostNumpy-master/build

make

[ 8%] Building CXX object CMakeFiles/boost_numpy.dir/src/boost/numpy/numpy.cpp.o
/root/BoostNumpy-master/src/boost/numpy/numpy.cpp: In function ‘void boost::numpy::initialize()’:
/root/BoostNumpy-master/src/boost/numpy/numpy.cpp:50: error: return-statement with a value, in function returning 'void'
make[2]: *** [CMakeFiles/boost_numpy.dir/src/boost/numpy/numpy.cpp.o] Error 1
make[1]: *** [CMakeFiles/boost_numpy.dir/all] Error 2
make: *** [all] Error 2

Suggestions and advice appreciated!

Thank you

[TallJimbo]

I think I know what's going on.

The return statement with a value is coming from the NumPy C API's import_array and/or import_ufunc macros (without those expanding, there is no line 50 in numpy.cc), and they're assuming they're being called directly within a Python 3 module initialization function. In Python 3, module initialization functions return PyObject* while Python 2's do not return anything. This is apparently contrary to the NumPy docs; this should probably be reported upstream.

Anyhow, you might be able to get it working if you just change the signature of boost::numpy::initialize to return PyObject*. That ought to imply some further changes in how you call it and check for errors, and it's not clear what the return value ought to be when there isn't a failure. But it might take me a while to get to a better fix for that, as I don't have Python 3 development environment setup at the moment, so I'm hoping this will tide you over.

[nbecker]

In my code I have:

#if PY_MAJOR_VERSION == 2
static void wrap_import_array() {
import_array();
}
#else
static void * wrap_import_array() {
import_array();
}
#endif

On Mon, Jun 6, 2016 at 1:11 PM, Jim Bosch [email protected] wrote:

I think I know what's going on.

The return statement with a value is coming from the NumPy C API's
import_array and/or import_ufunc macros (without those expanding, there
is no line 50 in numpy.cc), and they're assuming they're being called
directly within a Python 3 module initialization function. In Python 3,
module initialization functions return PyObject* while Python 2's do not
return anything). This is apparently contrary to the NumPy docs
http://docs.scipy.org/doc/numpy-1.10.1/reference/c-api.array.html#c.import_array;
this should probably be reported upstream.

Anyhow, you might be able to get it working if you just change the
signature of boost::numpy::initialize to return PyObject*. That ought to
imply some further changes in how you call it and check for errors, and
it's not clear what the return value ought to be when there isn't a
failure. But it might take me a while to get to a better fix for that, as I
don't have Python 3 development environment setup at the moment, so I'm
hoping this will tide you over.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#60 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/AAHK0J9I9VvfdaP625RZJ_6Hx-B-X2Ynks5qJFSogaJpZM4IvF5z
.

Those who don't understand recursion are doomed to repeat it

[classfire]

Hi,
I modified the return type of the initialize() function to PyObject_, in numpy.hpp and numpy.cpp. make and make install then proceeded satisfactorily. Running make test, however, has a similar error pattern to another open issue (#26):
`Running tests...
Test project /root/BoostNumpy-master/build
Start 1: dtype_test
1/4 Test #1: dtype_test ......................._**Failed Error regular expression found in output. Regex=[ERROR:] 1.05 sec
Start 2: ndarray_test
2/4 Test #2: ndarray_test ..................... Passed 0.16 sec
Start 3: indexing_test
3/4 Test #3: indexing_test .................... Passed 0.13 sec
Start 4: dstream_test
4/4 Test #4: dstream_test ..................... Passed 0.21 sec

75% tests passed, 1 tests failed out of 4

Total Test time (real) = 1.58 sec

The following tests FAILED:
1 - dtype_test (Failed)
Errors while running CTest
make: *** [test] Error 8
`
I see the "long()" function is used in dtype_test.py, something Python3 will not recognize?

Thank you.

[classfire]

Yes, that fixed it: I commented out all the 'assertEquivalent()' calls that used long() in dtype_test.py and the test suite now passes.

[TallJimbo]

Thanks for the testing, and I'm glad you've gotten it working. I'll try to work these into the master branch when I get a chance. If you have a patch (or a git fork) you'd be willing to contribute, that'd be even better.

[TallJimbo]

Fixed merged to master at f1ffaac.