Installing more than one version of Erlang/OTP on a machine

Question

Is this possible to have different versions of Erlang/OTP installed simultaneously on the same platform?

Answer 1

Using the Nix package manager there is no need to globally install interpreters (especially because sometimes multiple versions are needed), and nix-shell will open up a sub-shell with the Erlang executable available in the path.

1. Getting only the Erlang (no shell.nix)

For the current version in the active channel:

nix-shell -p erlang

For other versions not in the current channel, a specific channel can be given:

nix-shell -I nixpkgs=channel:nixos-unstable -p erlangR22

Or add path to the Nix expression in your NixOS/nixpkgs clone:

$ nix-shell -I nixpkgs=~/clones/nixpkgs -p erlangR23

2. More elaborate project setup configurations: use a shell.nix file

A complex development environment can be spun up, and calling nix-shell shell.nix will take care of all - even automatically when entering a directory if set up with direnv(archived).

2.1 Examples

Erlang

This will drop you in a shell with erl and rebar3 available, along with the other programs specified in buildInputs.

{ pkgs ? import ~/clones/nixpkgs {} }:

pkgs.mkShell {

  buildInputs = with pkgs; [
    beam.packages.erlangR22.erlang
    beam.packages.erlangR22.rebar3
    curl
    ffmpeg
    git
    google-cloud-sdk
    jq
  ];

  # Where would be the best place for this?
  shellHook = ''
    export ERL_AFLAGS="-kernel shell_history enabled"
  '';

Elixir/Phoenix web project

This (archived) will set up an environment for an Elixir/Phoenix web app complete with a spun up PostgreSQL dev instance:

####################################################################
# Importing a cloned Nixpkgs repo  (from my home directory), because
# the latest channels don't have Elixir 1.9.
# See https://nixos.org/nix/manual/#idm140737317975776 for the meaning
# of `` and `~` in Nix expressions (towards the end of that
# section).
####################################################################

{ pkgs ? import ~/clones/nixpkgs {} }:

pkgs.mkShell {

  buildInputs = with pkgs; [
    beam.packages.erlangR22.elixir_1_9
    postgresql_11
    nodejs-12_x
    git
    inotify-tools
  ];

  shellHook = ''

    ####################################################################
    # Create a diretory for the generated artifacts
    ####################################################################

    mkdir .nix-shell
    export NIX_SHELL_DIR=$PWD/.nix-shell

    ####################################################################
    # Put the PostgreSQL databases in the project diretory.
    ####################################################################

    export PGDATA=$NIX_SHELL_DIR/db

    ####################################################################
    # Put any Mix-related data in the project directory
    ####################################################################

    export MIX_HOME="$NIX_SHELL_DIR/.mix"
    export MIX_ARCHIVES="$MIX_HOME/archives"

    ####################################################################
    # Clean up after exiting the Nix shell using `trap`.
    # ------------------------------------------------------------------
    # Idea taken from
    # https://unix.stackexchange.com/questions/464106/killing-background-processes-started-in-nix-shell
    # and the answer provides a way more sophisticated solution.
    #
    # The main syntax is `trap ARG SIGNAL` where ARG are the commands to
    # be executed when SIGNAL crops up. See `trap --help` for more.
    ####################################################################

    trap \
      "
        ######################################################
        # Stop PostgreSQL
        ######################################################

        pg_ctl -D $PGDATA stop

        ######################################################
        # Delete `.nix-shell` directory
        # ----------------------------------
        # The first  step is going  back to the  project root,
        # otherwise `.nix-shell`  won't get deleted.  At least
        # it didn't for me when exiting in a subdirectory.
        ######################################################

        cd $PWD
        rm -rf $NIX_SHELL_DIR
      " \
      EXIT

    ####################################################################
    # If database is  not initialized (i.e., $PGDATA  directory does not
    # exist), then set  it up. Seems superfulous given  the cleanup step
    # above, but handy when one gets to force reboot the iron.
    ####################################################################

    if ! test -d $PGDATA
    then

      ######################################################
      # Init PostgreSQL
      ######################################################

      pg_ctl initdb -D  $PGDATA

      ######################################################
      # PORT ALREADY IN USE
      ######################################################
      # If another `nix-shell` is  running with a PostgreSQL
      # instance,  the logs  will show  complaints that  the
      # default port 5432  is already in use.  Edit the line
      # below with  a different  port number,  uncomment it,
      # and try again.
      ######################################################

      # sed -i "s|^#port.*$|port = 5433|" $PGDATA/postgresql.conf

    fi

    ####################################################################
    # Start PostgreSQL
    # ==================================================================
    # Setting all  necessary configuration  options via  `pg_ctl` (which
    # is  basically  a wrapper  around  `postgres`)  instead of  editing
    # `postgresql.conf` directly with `sed`. See docs:
    #
    # + https://www.postgresql.org/docs/current/app-pg-ctl.html
    # + https://www.postgresql.org/docs/current/app-postgres.html
    #
    # See more on the caveats at
    # https://discourse.nixos.org/t/how-to-configure-postgresql-declaratively-nixos-and-non-nixos/4063/1
    # but recapping out of paranoia:
    #
    # > use `SHOW`  commands to  check the  options because  `postgres -C`
    # > "_returns values  from postgresql.conf_" (which is  not changed by
    # > supplying  the  configuration options  on  the  command line)  and
    # > "_it does  not reflect  parameters supplied  when the  cluster was
    # > started._"
    #
    # OPTION SUMMARY
    # --------------------------------------------------------------------
    #
    #  + `unix_socket_directories`
    #
    #    > PostgreSQL  will  attempt  to create  a  pidfile  in
    #    > `/run/postgresql` by default, but it will fail as it
    #    > doesn't exist. By  changing the configuration option
    #    > below, it will get created in $PGDATA.
    #
    #   + `listen_addresses`
    #
    #     > In   tandem  with   edits   in  `pg_hba.conf`   (see
    #     > `HOST_COMMON`  below), it  configures PostgreSQL  to
    #     > allow remote connections (otherwise only `localhost`
    #     > will get  authenticated and the rest  of the traffic
    #     > discarded).
    #     >
    #     > NOTE: the  edit  to  `pga_hba.conf`  needs  to  come
    #     >       **before**  `pg_ctl  start`  (or  the  service
    #     >       needs to be restarted otherwise), because then
    #     >       the changes are not being reloaded.
    #     >
    #     > More info  on setting up and  troubleshooting remote
    #     > PosgreSQL connections (these are  all mirrors of the
    #     > same text; again, paranoia):
    #     >
    #     >   + https://stackoverflow.com/questions/24504680/connect-to-postgres-server-on-google-compute-engine
    #     >   + https://stackoverflow.com/questions/47794979/connecting-to-postgres-server-on-google-compute-engine
    #     >   + https://medium.com/scientific-breakthrough-of-the-afternoon/configure-postgresql-to-allow-remote-connections-af5a1a392a38
    #     >   + https://gist.github.com/toraritte/f8c7fe001365c50294adfe8509080201#file-configure-postgres-to-allow-remote-connection-md

    HOST_COMMON="host\s\+all\s\+all"
    sed -i "s|^$HOST_COMMON.*127.*$|host all all 0.0.0.0/0 trust|" $PGDATA/pg_hba.conf
    sed -i "s|^$HOST_COMMON.*::1.*$|host all all ::/0 trust|"      $PGDATA/pg_hba.conf

     #  + `log*`
     #
     #    > Setting up basic logging,  to see remote connections
     #    > for example.
     #    >
     #    > See the docs for more:
     #    > https://www.postgresql.org/docs/current/runtime-config-logging.html

    pg_ctl                                                  \
      -D $PGDATA                                            \
      -l $PGDATA/postgres.log                               \
      -o "-c unix_socket_directories='$PGDATA'"             \
      -o "-c listen_addresses='*'"                          \
      -o "-c log_destination='stderr'"                      \
      -o "-c logging_collector=on"                          \
      -o "-c log_directory='log'"                           \
      -o "-c log_filename='postgresql-%Y-%m-%d_%H%M%S.log'" \
      -o "-c log_min_messages=info"                         \
      -o "-c log_min_error_statement=info"                  \
      -o "-c log_connections=on"                            \
      start

    ####################################################################
    # Install Node.js dependencies if not done yet.
    ####################################################################

    if test -d "$PWD/assets/" && ! test -d "$PWD/assets/node_modules/"
    then
      (cd assets && npm install)
    fi

    ####################################################################
    # If $MIX_HOME doesn't exist, set it up.
    ####################################################################

    if ! test -d $MIX_HOME
    then

      ######################################################
      # ...  but first,  test whether  there is  a `_backup`
      # directory. Had issues with  installing Hex on NixOS,
      # and Hex and  Phoenix can be copied  from there, just
      # in case.
      ######################################################

      if test -d "$PWD/_backup"
      then
        cp -r _backup/.mix .nix-shell/
      else
        ######################################################
        # Install Hex and Phoenix via the network
        ######################################################

        yes | mix local.hex
        yes | mix archive.install hex phx_new
      fi
    fi

    if test -f "mix.exs"
    then
      # These are not in the  `if` section above, because of
      # the `hex` install glitch, it  could be that there is
      # already a `$MIX_HOME` folder. See 2019-08-05_0553

      mix deps.get

      ######################################################
      # `ecto.setup` is defined in `mix.exs` by default when
      # Phoenix  project  is  generated via  `mix  phx.new`.
      # It  does  `ecto.create`,   `ecto.migrate`,  and  run
      # `priv/seeds`.
      ######################################################
      mix ecto.setup
    fi
  '';

  ####################################################################
  # Without  this, almost  everything  fails with  locale issues  when
  # using `nix-shell --pure` (at least on NixOS).
  # See
  # + https://github.com/NixOS/nix/issues/318#issuecomment-52986702
  # + http://lists.linuxfromscratch.org/pipermail/lfs-support/2004-June/023900.html
  ####################################################################

  LOCALE_ARCHIVE = if pkgs.stdenv.isLinux then "${pkgs.glibcLocales}/lib/locale/locale-archive" else "";
}

---

How to find packages with attributes path on the console

$ nix-env -qaP 'erlang*'
# ...
nixos.erlangR20            erlang-20.3.8.9
nixos.erlangR21            erlang-21.3.8.3
nixos.erlang               erlang-22.1.7
# ...

$ nix-env -f ~/clones/nixpkgs/ -qaP 'erlang*'
# ...
nixos.erlangR20            erlang-20.3.8.9
nixos.erlangR21            erlang-21.3.8.3
nixos.erlang               erlang-22.1.7
# ...
=== >>> erlangR23            erlang-23.0.2  <<<====