+Title: The LANGUAGE variable is broken for English as main language
+Tags: en, linux, language, gettext, lang
+
+The ``LANGUAGE`` environment variable can `accept multiple fallback
+languages`__ (at least if your commands are using ``gettext``), so if your main
+``LANG`` is, say, ``es``, but you also speak ``fr``, then you can use
+``LANGUAGE=es:fr``.
+
+__ https://www.gnu.org/software/gettext/manual/gettext.html#The-LANGUAGE-variable
+
+But what happens when you main ``LANG`` is ``en``, so for example your
+``LANGUAGE`` looks like ``en:es:de``? You'll notice some message that used to
+be in perfect English before using the multi-language fallback now seem to be
+shown randomly in ``es`` or ``de``.
+
+Well, it is not random. The thing is, since English tends to be the de-facto
+language for the original strings in a program, it looks like almost **nobody**
+provides an ``en`` *translation*, so when fallback is active, almost no
+programs will show messages in English.
+
+For example, this is my Debian testing system with roughly 3.5K packages
+installed:
+
+.. code:: console
+
+ $ dpkg -l |wc -l
+ 3522
+ $ ls /usr/share/locale/en/LC_MESSAGES/ | wc -l
+ 12
+
+Only 12 packages have a plain English locale. ``en_GB`` does a bit better:
+
+.. code:: console
+
+ $ ls /usr/share/locale/en_GB/LC_MESSAGES/ | wc -l
+ 732
+
+732 packages. This is still lower than both ``en`` and ``de``:
+
+.. code:: console
+
+ $ ls /usr/share/locale/es/LC_MESSAGES/ | wc -l
+ 821
+ $ ls /usr/share/locale/de/LC_MESSAGES/ | wc -l
+ 820
+
+The weird thing is packages as basic as ``psmisc`` (providing, for example,
+``killall``) and ``coreutils`` (providing, for example, ``ls``) don't have an
+``en`` locale, and ``psmisc`` doesn't provide ``es``. This is why at some point
+it seemed like a random locale was being used. I had something like
+``LANGUAGE=en_GB:en_US:en:es:de`` and I use KDE as my desktop environment. KDE
+seems to be correctly *translated* to ``en_GB``, so I was seeing most of my
+desktop in English as expected, but when using ``killall``, I got errors in
+German, and when using ``ls``, I got errors in Spanish.
+
+If you don't provide other fallback languages, gettext will automatically fall
+back to the ``C`` locale, which is the original strings embedded in the source
+code, which are usually in English, and this is why if you don't provide
+fallback languages (other than English at least), all will work in English as
+expected. Of course if you use ``C`` in your fallback languages, before any
+non-English language, then they will be ignored as the ``C`` locale should
+always be present, so that's not an option.
+
+I find it very curious that this issue has almost zero visibility. At least my
+searches for the issue didn't throw any useful results. I had to figure it all
+out by myself like in the good old pre-stackoverflow times...
+
+.. note:: I know is not a typical use case, as since almost all software use
+ English for the ``C`` locale it hardly makes any sense to use fallback
+ languages in practice if your main language is English. But theoretically it
+ could happen, and providing an ``en`` translation is trivial.
projects / personal / website.git / commitdiff