Conda for C++ packages

What makes conda strong

Conda is different from your average package manager. It’s different in the following ways:

• It’s truly cross-platform: conda works on Windows (!), OS X and Linux. That makes it very different from apt, brew, dnf, yum, chocolatey …
• Conda supports environments natively. You can have multiple environments with different versions of some library side-by-side – and switch between them conveniently using conda activate my_environment.
• Conda supports distributing binaries. A lot of C++ libraries are difficult to build, or take a long time to build. Therefore shipping binaries (on all platforms) is a huge feature. You can find big libs such as boost on conda.
• Conda is language-agnostic: you have cargo for Rust, pip for Python and Pkg for Julia, Conan for C++ … where will it stop?
• Contrary to the naive attempt, conda does true dependency solving – conda uses a SAT solver to find a solution to the dependency problem if it exists.

And last but not least, conda has already found widespread use in many communities. The Python data science landscape and the R community are some of the heavy conda users out there. One of the largest community engineering efforts out there is probably the maintenance of conda-forge, a big repository of Open Source software that’s built and maintained completely in the open.

That’s why we (at QuantStack) think that conda is the package manager of the future – not just for Python or R, but also for C++. In this blog post we’ll show how convenient it is to use conda environments as a C++ user.

Installing conda

The best way to get a minimal conda installation is using miniconda: https://docs.conda.io/en/latest/miniconda.html. Once installed you can install new packages using

$ conda install xtensor xsimd -c conda-forge which will download the package and install it in the currently activated environemnt (the default environment is called “base”). The environment (on Linux) mirrors a standard unix root-environment (with all the /etc, /include, /lib… folders).

Installing the conda C++ toolchain

For compatibility reasons, it’s advisable to install the conda toolchain for compiling (instead of using the system compilers, for example).

To use the conda compilers you need to install

1
2
3
4
5
6
# Linux
conda install gcc_linux-64 gxx_linux-64 # (optionally) gfortran_linux-64
# OS X
conda install clang_osx-64 clangxx_osx-64 # (optionally) gfortran_osx-64
# Windows
conda install vs2015_win-64 # or the newer vs2017_win-64

Note that on OS X you also need to have a SDK available.

With those packages installed, environment variables such as $CXX should automatically point to the correct C++ compiler, the same one used to compile the libraries downloaded from conda/conda-forge. The compiler activation script also sets the CXXFLAGS variable to come with the same set of standard C++ flags.¹

Using the C++ library installed with conda

Using the installed C++ libraries is actually surprisingly easy – given one uses CMake:

1
2
3
4
5
6
7
cd your_project 
mkdir build && cd build
# make sure your environemnt is activated!
conda activate {base / myenv}
cmake .. -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX
# now build ...
make

Now CMake is going to prioritize the libraries and CMake modules found in the conda environment.

If you are a Windows user, then the command will look slightly different:

1
2
3
4
5
6
7
8
cd your_project
mkdir build
cd build
# make sure your environemnt is activated!
conda activate {base / myenv}
cmake -G "NMake Makefiles" -DCMAKE_INSTALL_PREFIX=%MINICONDA%\\Library -DBUILD_TESTS=ON ..
# now build ...
nmake

And that’s it, this should show roughly how conda can be used with CMake for C or C++ projects.