A library that builds on Babel and Jinja to make extracting strings easy and uniform
Python Makefile
Switch branches/tags
Nothing to show
Latest commit e2ac254 Feb 5, 2015 @clouserw Merge pull request #31 from Osmose/fix-test-babel-1.3
Fix test failures with Babel 1.3, and run tox tests with 1.3.



Build Status

Tower provides some additional functionality to Jinja and it's i18n extension, and the Babel library.


  • Pulls strings from a variety of sources: Python, JavaScript, and .lhtml files.
  • Collapses whitespace in all strings to prevent unwieldy msgids.
  • Supports Gettext context (msgctxt) in all gettext, and ngettext calls.
  • Supports merging PHP and Python .pot files. This is temporary. If you want to support that for an extended time look at phppo2pypo in the Translate Toolkit.


  • Django
  • Babel
  • Jinja2 and Jingo
  • translate-toolkit

See requirements.txt for details.


Install from pypi with:

pip install tower

Source code is at http://github.com/clouserw/tower.

Install from GitHub with:

pip install -e git://github.com/clouserw/tower.git#egg=tower


Add to INSTALLED_APPS in your Django settings file:

    # ...
    # ...

Then configure.


Default: None--you must set this.

The value is a dict of domain to file spec and extraction method tuples.

For example, this creates a domain "messages" and in that domain extracts all the l10n strings from .py and .html files:

    'messages': [
        ('fjord/**.py', 'tower.extract_tower_python'),
        ('fjord/**.html', 'tower.extract_tower_template'),

Use tower.extract_tower_python for Python files and tower.extract_tower_template for Jinja2 templates.

The l10n strings will be saved in a .pot file with the name of the domain. In the above example, it'd be messages.pot.


Default: ['messages']

By default, all domains specified in DOMAIN_METHODS get merged into one big .pot file. If you don't want that, you can specify which domains get their own .pot file with STANDALONE_DOMAINS.


Gettext keywords are the actual names of the functions your application uses to denote strings which should be extracted. Babel defines a sane default list in DEFAULT_KEYWORDS which includes the regulars like _(), gettext(), ngettext(), etc. If your app uses different functions you'll need to define them here. This setting is a dictionary and the value follows gettext's keywordspec.

If you set this, it is merged with Babel's DEFAULT_KEYWORDS setting.


# We lazy load strings in our classes so we can cache them and they are
# still localized in the right language for the end user.  This will let
# us use _lazy('Localized string').
    '_lazy': None,


Default: False

If you have trouble extracting strings with Tower, try setting this to True.


This points to the source code directory where you want your locale/ directory to be.


This is a function that takes arbitrary set of args and combines them with ROOT to form a new path.


import os

# The settings file is in fjord/fjord/settings/base.py. From
# base, up two directories is the initial fjord directory
# which is where all the source code and the fjord Django
# project are.
ROOT = os.path.dirname(os.path.dirname(__file__))

path = lambda *args: os.path.abspath(os.path.join(ROOT, *args))


Default: True

By default tower will ensure it's gettext and ngettext functions are installed into Jinja2 on every call to tower.activate(). You likely want this, but if you need to provide your own gettext and ngettext functions, set this to False, and in your project call jingo.env.install_gettext_translations or jingo.env.install_gettext_callables.



./manage.py extract


./manage.py merge

A note on whitespace

When tower extracts strings, it collapses whitespace. This makes it easier for localizers. It also means you need to use ugettext, ungettext, ugettext_lazy and ungettext_lazy from tower. Otherwise the msgids being passed in won't have their whitespace stripped and thus won't match anything in your .mo file.

A note on safe-ness

L10n strings are marked "safe" for Jinja2 automatically, so they will not be HTML-escaped:

{{ _('Hello <strong>World</strong>') }}

This works as expected. When interpolating into an L10n string, however, it will be marked as "unsafe" and escaped, unless you use jingo's |fe() helper (which will escape the arguments but not the string they are interpolated into). Like this:

{{ _('Hello <strong>{0}</strong>')|fe(user.nickname) }}

Run tests


python run_tests.py

To test on all supported versions of Python and Django:

$ pip install tox
$ tox