Technology and Software, Tips

How to ssh into Samsung’s Linux on Dex

Problem: you have a new Samsung tablet or phone with Linux on Dex and you want to ssh into it from your computer. Why? To work with a better keyboard, to use `rsync`, `scp`, `sftp`, move files over sftp from Ubuntu’s Nautilus, etc.


Step 1) Edit /etc/ssh/sshd_config in Linux on Dex and change:

Port 22 to something above 1024. Example: Port 9022
Uncomment ListenAddress and make it listen to the IP address of the tablet (use ifconfig to get it). Example: ListenAddress
At the very end of the file: UsePAM no

If you connect to another network you’ll have to change the address for ListenAddress

I suggest these other configurations:

PermitRootLogin No
PubkeyAuthentication yes
PermitEmptyPassword no
PasswordAuthentication no

Step 2) Every time you start the Linux on Dex container, run the command
sudo mkdir /var/run/sshd

Unfortunately this directory doesn’t survive the termination of the container.
It must be owned by root.

Step 3) Change the ownership of the site certificates
sudo chown dextop /etc/ssh/ssh_host_*

Step 4) Start the ssh daemon in a terminal with /usr/sbin/sshd -D

Unfortunately starting it with systemd won’t work (this is a LXD container, not a VM) and starting it with root fails too (same problem)

You can create a bash script for the steps from 2 to 4 and run it

Step 5) Create the directory ~dextop/.ssh in Linux of Dex
mkdir ~dextop/.ssh
chmod 700 ~dextop/.ssh

Step 6) Create a public/private key pair on your computer. I recommend
ssh-keygen -t ed25519 -f ~/.ssh/linux_on_dex

Step 7) Copy your computer’s ~/.ssh/ into the ~/.ssh/authorized_keys in Linux of Dex

Step 8) I recommend to create a ~/.ssh/config file on your computer with this content:

Host *
Protocol 2
ServerAliveInterval 60
Host linux_on_dex
HostName # the address in your network
User dextop
Port 9022 # the one you choose
PubkeyAuthentication yes
IdentitiesOnly yes
IdentityFile ~/.ssh/linux_on_dex

Step 9) Connect from your computer with ssh linux_on_dex


I noticed that ssh freezes when the tablet is locked so the screen must be on and unlocked. This is far from ideal. If you know a workaround please add a comment to this post.

Technology and Software, Tips

Named pipes, ports and Erlang code in an Elixir project

I needed to read from a named pipe in an Elixir program, I made a mistake and I learned a few thing. Follow me.

First create the named pipe:

$ mkfifo "pipe"

In Erlang you read from it like this

$ erl
1> Fifo = open_port("pipe", [eof]).
2> receive
2>   {Fifo, {data, Data}} ->
2>     io:format("Got some data: ~p~n", [Data])
2>   end.

You can check that it works by executing echo hi > pipe from another shell.

So I expected to able to write this in Elixir

$ iex
 iex(1)> fifo ="pipe", [:eof])

However this is what I get

** (ArgumentError) argument error
:erlang.open_port({"pipe"}, [:eof])

If you can see my noob mistake in the call to don’t tell anybody yet, read on.

After trying many permutations of the arguments and read twice both and I gave up. I resolved to write it as an Erlang module and call it from my Elixir project. It was pretty easy:

1) Create an erlang directory in the main directory of the project. The name is not magical, you can name it as you wish.

2) Add this line into the project options of mix.exs

erlc_paths: ["erlang"],

That’s the directory you created.

3) Create an erlang/namedpipe.erl file with this code


read(Pipe) ->
  Fifo = open_port(Pipe, [eof]),
    {Fifo, {data, Data}} ->

See how it can almost map 1 to 1 to Elixir. Variables are capitalized, symbols are lowercased and there are statement terminators (comma and full stop). All functions are private to the module except the explicitly exported ones.

4) Run mix and see that it compiles the Erlang file. Great!

But now

$ iex
** (ArgumentError) argument error
:erlang.open_port("pipe", [:eof])

Oh oh, what’s going on? I finally realized that it’s because of the wrong quote character! Single quotes are needed when passing strings to Erlang. Single quotes in Elixir are character lists, which is what Erlang needs. Double quotes are UTF-8 encoded binary data, which Erlang doesn’t understand.

So this works:

$ iex

Now go to another shell and run echo hi > pipe and confirm that it works for you too.

But wait, I did use double quotes in my first failed Elixir attempt. So did I do all of this for nothing? Embarrassingly, yes. This works:

$ iex
iex(1)> fifo ='pipe', [:eof])
iex(2)> receive do
...(2)>   {fifo, {:data, data}} ->
...(2)>     IO.puts("Got some data: #{data}")
...(2)> end

At least I learned how to embed Erlang code in an Elixir project and to care about single and double quotes.

Finally, you can embed that code in an Elixir module and call it both with a single quoted char list or a double quoted binary. Write this into lib/namedpipe.ex

defmodule NamedPipe do
  def read(pipe) when is_binary(pipe) do

  def read(pipe) do
    fifo =, [:eof])
      receive do
        {fifo, {:data, data}} ->

It uses guards to decide which version of the read function to call. Now

$ iex -S mix


Technology and Software

What’s the [5.0] in Rails 5’s ActiveRecord::Migration

The migration examples in the Rails Edge Guide are all like this

class MyMigration < ActiveRecord::Migration[5.0]

What’s that [5.0]? It can’t possibly be Ruby. Or is it?

With a little experimenting in rails c it’s easy to understand that [ is a class method of ActiveRecord::Migration. It’s defined in activerecord-5.0.0.beta1/lib/active_record/migration.rb and it accepts the 4.2 and 5.0 arguments. It allows us to select which version of the migrations we want to use. Production ready versions of ActiveRecord don’t have that method so it should go away as soon as Rails 5 goes out of beta.


Technology and Software

Configuring nginx for letsencrypt

Letsencrypt comes with a plugin for Apache. The one for nginx is still experimental. The manual configuration is pretty easy. On the server to protect with SSL:

git clone
cd letsencrypt
letsencrypt-auto certonly -a manual --rsa-key-size 4096 \
--email -d -d

This creates a directory /etc/letsencrypt with your account data and your certificates in the live/ subdirectory.

Edit you nginx configuration file and add

ssl_certificate /etc/letsencrypt/live/;
ssl_certificate_key /etc/letsencrypt/live/;

to the SSL configuration of your server. It’s important that you use fullchain.pem for the certificate, and not the cert.pem in the letsencrypt directory. Programs like curl and wget won’t work if you use cert.pem. The reason is explained in the first answer to an issue I wrongly opened to letsencrypt. A more detailed explanation is here.

Restart nginx to test your new certificate.

Remember to setup a cron job to renew the certificate before it expires in 90 days. You should also check Mozilla’s SSL Configuration Generator to improve the security of your https server.

Technology and Software

Ruby 2.3.0 InstructionSequence

Ruby 2.3 has been released on Christmas day as every Ruby version. It comes with a nice present: the RubyVM::InstructionSequence class with methods to compile scripts, save them and load them later. A quick example:

rvm install ruby-2.3.0
rvm use ruby-2.3.0
cat > test.rb
class Christmas

cat > compile.rb
instruction_sequence = 
  RubyVM::InstructionSequence.compile_file("test.rb")"test.iseq", "wb") do |file|

cat > instruction_sequence = nil"test.iseq", "rb") do |file|
  instruction_sequence =

ruby compile.rb
compile.rb  execute.rb  test.iseq  test.rb
ruby execute.rb

It works!

InstructionSequence comes with a caveat:

The goal of this project is to provide “machine dependent” binary file to achieve:

  • fast bootstrap time for big applications
  • reduce memory consumption with several techniques

“Machine dependent” means you can’t migrate compiled binaries to other machines.

Does it means that the compiled code won’t work on another machine? I generated the .iseq file on a Ubuntu 12.04 machine and uploaded it to a Ubuntu 14.04 one, both 64 bit. It keeps working and the directory structure of the two machines can be different, despite the presence in the compiled code of metadata about the source file.

I invite the readers to check the other methods of the class. They let allow for the compilation of strings of text and procs, setting compilation options, disassembling iseq code plus several instance methods that operate on an instruction sequence.

Technology and Software

Ruby performances with PostgreSQL and MySQL

(original post in the Italian Ruby Forum)

I had to convert a database seeding script from PostgreSQL 9.4 to MariaDB 10 (customer’s choice and with little enthusiasm I had to comply). This lead to a number of interesting discoveries about the pg and mysql2 Ruby drivers. Apart a few minor issues [1] [2] [3] [4] I immediately noticed that the script with MariaDB run 20 times (twenty) more slowly than the  PostgreSQL one: 21 minutes vs 1 minute and 3 seconds. Unusable and inexplicable.

Such a big difference can not be due to the database, so I started to investigate the configuration. Even the MySQL coming with Ubuntu 12.04 was too slow and I can expect that the distributors set it up reasonably well. At this point the suspect becomes the driver. I opened this issue and they gave me two valuable tips: use a profiler ( ) and the gem-import activerecord ( ). I knew both of them but sometimes you have to be reminded about tools you don’t use often. Ops.

The profiler show that the driver uses pg prepared statements that give obvious benefits with the number of records created by my script (a little over 32,000). The version of mysql2 I had to use (0.3.x) does not have prepared statement (but the newer version does) and that seems to make the difference. I rewrote the script to use activerecord-import, which  inserts a whole array of objects at once. The script looks a bit unnatural, because I repeatedly needed the ids of the record I created to pass them along the associations, but the execution times for mysql2 dropped from 21 minutes to 1 minute and 33″. It was worth it. There are only 1,045 calls to the db and yet is always slower than 32k calls made by the original script with pg. The script with pg and activerecord-import dropped to 47 seconds.

Despite all the enhancements introduced in the import-activerecord calls my script’s calls to PostgreSQL add up to 9.4 seconds. The calls to MariaDB are  49.8 seconds. Ruby accounts for 40 seconds, regardless of the database used.


1) Work on PostgreSQL has performance advantages with Ruby due to drivers.

2) mysql2 0.4.0+ has prepared statements but if you’re working with Rails you must be careful. There are issues [A] [B] and it seems you need Rails 4.2.5+ to use it. I didn’t test the combination yet.

3) For details of my profiling research (tables, times, calls) read 


Finally, the issues I run into:

[1] https: // …

[2] For MariaDB install the gem mysql2 with

bundle config \
build.mysql2 --with-mysql-config=/path/to/mariadb/bin/mysql_config

Careful: this is globals so use –with-mysql-config=/usr/bin/mysql_config when you need to connect to MySQL.

[3] My script would clear the db before seeding using TRUNCATE CASCADE, but MySQL and MariaDB don’t have it. This  is the workaround

 connection = ActiveRecord :: Base.connection
 Connection.Execute ("SET foreign_key_checks = 0;")
 [all models] .each do | model |
   Connection.Execute ("TRUNCATE model.table_name # {}")
 Connection.Execute ("SET foreign_key_checks = 1;")

[4] But neither ActiveRecord has TRUNCATE, so either you use some gems that add it to AR or even for PostgreSQL you need a loop like that, but you don’t need the SET foreign_ley_checks statements.

Technology and Software

LibreOffice, Python3 and AttributeError: ‘NoneType’ object has no attribute ‘supportsService’

I had to run a modified version of the famous script on Ubuntu 14.04. It run on 12.04 well but Ubuntu 14.04 comes with Python3 and the Uno library to interface Python with LibreOffice or OpenOffice doesn’t work.

Solution: convert the script to Python3 (exceptions, print, has_key have been changed) then install these libraries:

sudo apt-get install libreoffice-dev libreoffice-script-provider-python python3-uno

and the program will work. If you fail to install them you’ll get the AttributeError: ‘NoneType’ object has no attribute ‘supportsService’ error because loadComponentFromURL won’t be able to read the input file.