All the instructions below work on Linux, macOS and Windows.


The recommended way to install LFortran is using Conda. Install Conda for example by installing the Miniconda installation by following instructions there for your platform. Then create a new environment (you can choose any name, here we chose lf) and activate it:

conda create -n lf
conda activate lf

Then install LFortran by:

conda install lfortran -c conda-forge

Now the lf environment has the lfortran compiler available, you can start the interactive prompt by executing lfortran, or see the command line options using lfortran -h.

The Jupyter kernel is automatically installed by the above command, so after installing Jupyter itself:

conda install jupyter -c conda-forge

You can create a Fortran based Jupyter notebook by executing:

jupyter notebook

and selecting New->Fortran.

Build From a Source Tarball

This method is the recommended method if you just want to install LFortran, either yourself or in a package manager (Spack, Conda, Debian, etc.). The source tarball has all the generated files included and has minimal dependencies.

First we have to install dependencies, for example using Conda:

conda create -n lf python cmake llvmdev
conda activate lf

Then download a tarball from, e.g.:

tar xzf lfortran-0.9.0.tar.gz
cd lfortran-0.9.0

And build:

cmake -DWITH_LLVM=yes -DCMAKE_INSTALL_PREFIX=`pwd`/inst .
make -j8
make install

This will install the lfortran into the inst/bin.

Build From Git

We assume you have C++ compilers installed, as well as git and wget. In Ubuntu, you can also install binutils-dev for stacktraces.

If you do not have Conda installed, you can do so on Linux (and similarly on other platforms):

wget --no-check-certificate -O
bash -b -p $HOME/conda_root
export PATH="$HOME/conda_root/bin:$PATH"

Then prepare the environment:

conda create -n lf -c conda-forge llvmdev=11.0.1 bison=3.4 re2c python cmake make toml
conda activate lf

Clone the LFortran git repository:

git clone
cd lfortran

Generate files that are needed for the build (this step depends on re2c, bison and python):


Now the process is the same as installing from the source tarball. For example to build in Debug mode:

make -j8

Run tests:


Run an interactive prompt:


Enabling the Jupyter Kernel

To install the Jupyter kernel, install the following Conda packages also:

conda install xeus xtl nlohmann_json cppzmq

and enable the kernel by -DWITH_XEUS=yes and install into $CONDA_PREFIX. For example:

cmake \
    -DWITH_LLVM=yes \
    -DWITH_XEUS=yes \
cmake --build . -j4 --target install

To use it, install Jupyter (conda install jupyter) and test that the LFortran kernel was found:

jupyter kernelspec list --json

Then launch a Jupyter notebook as follows:

jupyter notebook

Click New->Fortran. To launch a terminal jupyter LFortran console:

jupyter console --kernel=fortran

Build From Git with Nix

One of the ways to ensure exact environment and dependencies is with nix. This will ensure that system dependencies do not interfere with the development environment. If you want, you can report bugs in a nix-shell environment to make it easier for others to reproduce.

With Root

We start by getting nix. The following multi-user installation will work on any machine with a Linux distribution, MacOS or Windows (via WSL):

sh <(curl -L --daemon

Without Root

If you would like to not provide nix with root access to your machine, on Linux distributions we can use nix-portable.


Now just prepend all nix-shell commands with NP_RUNTIME=bwrap ./nix-portable. So:

# Do not
nix-shell --run "bash"
# Do
NP_RUNTIME=bwrap ./nix-portable nix-shell --run "bash"


Now we can enter the development environment:

nix-shell --run "bash" --cores 4 -j4 --pure ci/shell.nix

The --pure flag ensures no system dependencies are used in the environment.

The build steps are the same as with the ci:


To change the compilation environment from gcc (default) to clang we can use --argstr:

nix-shell --run "bash" --cores 4 -j4 --pure ci/shell.nix --argstr clangOnly "yes"

Note About Dependencies

End users (and distributions) are encouraged to use the tarball from, which only depends on LLVM, CMake and a C++ compiler.

The tarball is generated automatically by our CI (continuous integration) and contains some autogenerated files: the parser, the AST and ASR nodes, which is generated by an ASDL translator (requires Python).

The instructions from git are to be used when developing LFortran itself.

Note for users who do not use Conda

Following are the dependencies necessary for installing this repository in development mode,


LFortran can print stacktraces when there is an unhandled exception, as well as on any compiler error with the --show-stacktrace option. This is very helpful for developing the compiler itself to see where in LFortran the problem is. The stacktrace support is turned off by default, to enable it, install binutils and compile LFortran with the -DWITH_STACKTRACE=yes cmake option.

In Ubuntu, apt install binutils-dev.

On macOS, you can install Spack, then:

spack install binutils
spack find -p binutils

The last command will show a full path to the installed binutils package. Add this path to your shell config file, e.g.:

export CMAKE_PREFIX_PATH_LFORTRAN=/Users/ondrej/repos/spack/opt/spack/darwin-catalina-broadwell/apple-clang-11.0.0/binutils-2.36.1-wy6osfm6bp2323g3jpv2sjuttthwx3gd

and compile LFortran with the -DCMAKE_PREFIX_PATH="$CMAKE_PREFIX_PATH_LFORTRAN;$CONDA_PREFIX" cmake option. The $CONDA_PREFIX is there if you install some other dependencies (such as llvm) using Conda, otherwise you can remove it.