arrayfire-python conflict with pycuda on anaconda with Linux

The failure is reported through arrayfire/tests/simple/array_test.py and arrayfire/tests/simple/_util.py, with formatting in arrayfire/array.py and error handling in arrayfire/util.py. Reproduce the Anaconda/Linux setup with pycuda installed, then compare it with the separate-PYTHONPATH workaround. Done means the simple tests pass without requiring that workaround, with the dependency interaction documented or fixed.

Hi,

I am using anaconda on linux, I have been previously using arrayfire on windows fine. I installed latest arrayfire on the Linux OS.. The error was

ERROR:root:Traceback (most recent call last):
  File "/home/sabih/anaconda3/lib/python3.6/site-packages/arrayfire/tests/simple/_util.py", line 32, in run
    test(verbose)
  File "/home/sabih/anaconda3/lib/python3.6/site-packages/arrayfire/tests/simple/array_test.py", line 26, in simple_array
    display_func(b)
  File "/home/sabih/anaconda3/lib/python3.6/site-packages/arrayfire/tests/simple/_util.py", line 50, in print_func_impl
    _print_log += str(arg) + '\n'
  File "/home/sabih/anaconda3/lib/python3.6/site-packages/arrayfire/array.py", line 1281, in __str__
    return self._get_metadata_str(dims=False) + self._as_str()
  File "/home/sabih/anaconda3/lib/python3.6/site-packages/arrayfire/array.py", line 1301, in _as_str
    safe_call(be.af_array_to_string(c_pointer(arr_str), "", self.arr, 4, True))
  File "/home/sabih/anaconda3/lib/python3.6/site-packages/arrayfire/util.py", line 79, in safe_call
    raise RuntimeError(to_str(err_str))
RuntimeError: In function void print(const char*, af_array, int, std::ostream&, bool) [with T = int; af_array = void*; std::ostream = std::basic_ostream<char>]
In file src/api/c/print.cpp:

After spending a day trying to think as to what the problem could be, I discovered that the problem is due to conflict with the install of pycuda, as soon as I removed pycuda, the arrayfire tests passed. The different version of pycuda give the same error.

Now I solved the issue by installing pycuda in a separate folder and adding that to the PYTHONPATH instead of the default in anaconda.