We need to define a workflow for the translations, as it leads to conflicts really easily.
For instance, I'm changing the code and some sentences in English, then I generate the translation strings with
bash --dump-po-strings neutrinet_cube_install.sh | msguniq -o locale/neutrinet.pot
And from that template, I start updating the French translation.
The problem is that the template (locale/neutrinet.pot) and the translation file (locale/fr/LC_MESSAGES/neutrinet.po) are generated by a program... The whole files might change even if we add just one line of code...
So, I propose to have a separate branch for translations (called locale-fr, locale-nl, etc.). Each time a new feature is merged, we pull the new changes from unstable branch into the locale branch. Then we push back the translated work into unstable again.