Configuring your compilers

We use the MinGW-w64 compiler, which is a free 64-bit compiler for Windows. These instructions are basically a direct copy of those created by Michael Hirsch, Ph.D., with a few extra steps added thanks to advice from mingwandroid on GitHub on help with final configuration issues. We very gratefully acknowledge these people who helped us figure out how to repeatably configure our Windows compilers.

We recommend creating a new environment for these instructions to prevent conflicting dependencies. Once you’ve verified that the procedure works in a clean environment, then you can repeat the process in another environment where things are already installed. Lastly, please note that any pre-existing fortran compilers and PATH additions may cause these instructions to not work.

To set up your compilers, follow these steps:

  1. Download and install Anaconda.

  2. Download the MinGW-w64 installer from Source Forge.

  3. Install the MinGW compiler with the following options:

    • Architecture: x86_64
    • Threads: posix
    • Exception: seh
    • Build revision: 0
    • Destination folder: c:\mingw

  4. Insert the path to the mingw bin folder to your system path. This will differ depending on whether you’re running Windows 7 or 10, so we recommend googling how to add something to your path variable for your architecture. The path that you want to add to your system variable should look something like C:\mingw\mingw64\bin.

  5. [Optional but recommended] In an Anaconda prompt, create the environment in which you want to have f2py and cython working, and then activate it:

    conda create -n py36_test python=3.6
activate py36_test

  6. Configure your Anaconda environment to use MinGW when it compiles by executing the command below from the Anaconda prompt.

    IMPORTANT!!! You must execute this command in every environment you want to compile from!

    (echo [build] & echo compiler = mingw32) > %CONDA_PREFIX%\Lib\distutils\distutils.cfg

  7. Install several necessary packages to your environment:

    conda install numpy libpython m2w64-toolchain cython

  8. Make sure you have the correct C++ redistributable installed on your computer. The type you need will depend on what version was used to build Anaconda. Detailed instructions on this can be found here.

Your system should now be configured to compile Fortran code and to cythonize Python code. You can test your configuration by going through the instructions here: Testing your compilers.