[pjotrp]

Docker for reproducible Science is an intermediate solution. While a Docker image can be moved and rerun (with some luck) the content of a Docker image is actually not transparent.

Reproducibility implies being able to regenerate the full container including software version control and visibility of the full dependency chain all the way down to BLAS and glibc! You can't do that by using apt, rpm, Perl CPAN, rubygems, Python pip and the like. None of these package managers have been designed for true isolation of packages and full reproducibility. That is why today people go with Docker. The shortcomings of these package managers drive people to Docker.

The technology for regenerating exact Docker containers exists in the form of GNU Guix and/or Nix packages. The fun fact is that when using GNU Guix, Docker itself is no longer required.

Watch GNU Guix.