Python 3.6：在 Windows 上安装 tensorflow 失败，出现 TypeError：stat：path should be string, bytes, os.PathLike or integer, not NoneType

Question

I am new to Python and Tensorflow. I tried to install tensorflow using the command pip install --upgrade tensorflow. However, the installation fails at two points:

1. Building wheel for wrapt (setup.py)
2. Running setup.py install for wrapt

During both these instances, the error is the same

TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType

Can anybody suggest some solution for this error?

Answer 1

I had same failure, tried older Python, older TensorFlow but no luck. Then fell upon this and it works for me (I'm on Windows). Make sure you do the following:

set WRAPT_INSTALL_EXTENSIONS=false

before running having pip install tensorflow .

Hope this helps!

Answer 2

set WRAPT_INSTALL_EXTENSIONS=false

Worked for me, but I had to run it from a command window, not powershell.

Answer 3

I wasted a lot of time troubleshooting this, but for me the root cause was that Microsoft Visual C++ Redistributable wasn't installed. (or the latest version)

The install script built the path of a dll in this package based on where it's supposed to be and one util function returns the path if it exists and None if it doesn't. The next util function will break with the above mentioned error if the dll doesn't exist and the previous function returned None. ( genericpath.py and msvc.py )

This error seems to be a symptom of another problem, which is the root cause. (missing C++ redistributable in my case). If you are feeling adventurous you can take a look at the stack trace and see who is calling the function with None. You might be able to add a few print s and see what is the root cause for you.

Answer 4

Took me hours to find this page finally as a solution. I tried everything, reinstall different versions of python, cleaning up my pc, etc. @AdamColton, you're a genius. Installation worked for me with tensorflow 2.0.0 and python 3.6.8 like a charm with this line of code before.