Building NEST on macOS¶

Building NEST on macOS requires some developer tools. There are several sources from which you can install them, e.g., Conda, Homebrew, or MacPorts. The most important recommendation for an easy and stable build is not to mix tools from different sources. This includes your Python installation: Taking Python from Conda and all else from Homebrew may work, but can also lead to various complications.

This guide shows how to build NEST with a development environment created with Conda. The main advantage of Conda is that you can fully insulate the entire environment in a Conda environment. If you want to base your setup on Homebrew or MacPorts, you can still use the environment.yml file as a guide to necessary packages.

Preparations¶

Install the Xcode command line tools by executing the following line in the terminal and following the instructions in the windows that will pop up:
```
xcode-select --install
```
Create a conda environment with necessary tools (see also Tips for installing NEST with conda)
```
cd <nest_source_dir>
conda env create -p conda/
```
Note

To build NEST natively on a Mac with Apple’s M1 chip, you need to use Miniforge as described in Tips for installing NEST with conda.
Activate the environment with
```
conda activate conda/
```
This assumes that you have created the environment in the folder conda/ as given above. Note that the trailing slash is necessary for conda to distinguish it from a named environment.
If you want to build NEST with MPI, you must digitally sign the orterun and orted binaries
1. If you do not yet have a self-signed code-signing certificate, create one as described here: https://gcc.gnu.org/onlinedocs/gcc-4.8.1/gnat_ugn_unw/Codesigning-the-Debugger.html (restarting does not appear to be necessary any more).
2. Sign your binaries
```
codesign -f -s "gdb-cert" `which orted`
codesign -f -s "gdb-cert" `which orterun`
```
  Instead of the which command you can also give the full path to the binary inside your conda environment.
  
  Note
  
  You will need to sign the binaries every time you update the OpenMPI package in your environment.

Building NEST¶

Download or clone the NEST sources from https://github.com/nest/nest-simulator.
Create a build directory outside the NEST sources and change into it.
Double check that you have activated a virtual environment.
Configure NEST by running
```
cmake <nest_source_dir>
```
If you have libraries required by NEST such as GSL installed with Homebrew and Conda, this can lead to library conflicts (error messages like Initializing libomp.dylib, but found libomp.dylib already initialized.). To ensure that libraries are found first in your conda environment, invoke cmake like this
```
CMAKE_PREFIX_PATH=<conda_env_dir> cmake <nest_source_dir>
```
You can find the <conda_env_dir> for the currently active conda environment by running conda info and looking for the “active env location” entry in the output.

To compile NEST with MPI support, add -Dwith-mpi=ON as cmake option. For further CMake options, see CMake Options for NEST.

Compile, install, and verify NEST with

make -j4         # -j4 builds in parallel using 4 processes
make install
make installcheck

Install NEST outside of a virtual environment¶

By default NEST will be installed into the active virtual Python environment. If you wish to install it elsewhere, you can specify an install prefix. Follow the above instructions, but use cmake -DCMAKE_INSTALL_PREFIX:PATH=<nest_install_dir> <nest_source_dir> instead. Note that when NEST is installed in a non-standard location, automatic discovery of the Python module is impossible, and environment variables must be set before NEST can be used:

source <nest_install_dir>/bin/nest_vars.sh

Troubleshooting¶

Conda with Intel MKL¶

A default installation of Anaconda or Miniconda will install a version of NumPy built on the Intel Math Kernel Library (MKL). This library uses a different OpenMP library to support threading than what’s included with Apple Clang or GCC. This will lead to conflicts if NEST is built with support for threading, which is the default and usually desirable. One way to avoid this is to follow the instructions above. An alternative is to create a conda environment in which you install nomkl as the very first package. This will tell conda to install MKL-free versions of NumPy and other linear-algebra intensive packages.