django-ftl is a Django package for using for Fluent, a localization system for today's world.
This package builds upon the Python implementation of Fluent and provides:
- A structure for setting up and managing your
.ftl
files. - Methods for switching/setting the current language.
- Integration into Django templates.
The defacto standard in Django world is GNU Gettext. See this Fluent vs gettext page for a comparison. In brief, here are some advantages:
- Fluent makes concerns like plural rules the job of the translator.
- Fluent gives translators the power to obey language specific rules (gender, case, plurals) that the developer may not be aware of, and shouldn't have to build into the software.
- Fluent integrates number and date formatting, and gives both developer and translators control over these, instead of these having to be handled separately, and only controlled by the developer.
To give an example, in GNU Gettext there is support for plural rules. However, this is the only language specific feature Gettext supports, and it is kind of bolted on afterwards. The developer also has to partially hard code the English rules (that is, the fact that there are two variants in English), as per the Django docs:
msg = ngettext(
'there is %(count)d object.',
'there are %(count)d objects.',
count) % {
'count': count,
}
Finally, this still doesn't work very well, because often you want to special case zero anyway - "there are no objects" (or "your inbox is empty" etc.) instead of "there are 0 objects".
In Fluent, plural rules are one example of a more generic mechanism for
selecting variants, and the translator is in control. The equivalent with
fluent/django-ftl, with special handling of the zero case included, looks like
this in an English .ftl
file:
there-are-some-objects = { $count -> [0] There are no objects. [1] There is one object. [other] There are { $count } objects. }
The Python code referencing this will only need to use the ID
(there-are-some-objects
) and pass the $count
argument.
Another problem that comes up is gender - for example, in French adjectives must agree in gender with the person being described. This can be solved in Fluent by passing the gender of the person as an argument, and allowing the translator to use the variant mechanism to write the correct language. This contrasts with GNU Gettext where the developer would have to create separate message strings for each case, because the message format is not powerful enough to allow the translator to add variant selection. Also, these different message strings will be identical in languages which don't have that feature โ in other words, the grammatical features of all languages end up either having a disproportionate effect on the source code and on other translators, or being neglected entirely.
The documentation for how to use django-ftl is in the docs/folder and online at https://django-ftl.readthedocs.io.
This package should be considered a beta. While it has a good feature set, test suite and docs, it has not been used a huge amount in production. In addition, it currently relies on our fork of python-fluent, which has the following significant features/changes not yet merged to the official version:
Refined error handling for functions (see PR).
Compiler implementation (see compiler branch). This is an additional and much faster implementation of
FluentBundle
that compiles FTL messages to Python AST, running the result through compile and exec.Use of
exec
for high performance Python code is an established technique used by other projects (e.g. Jinja2 and Mako), and we only use exec on data derived from FTL files, which will normally be created by translators and not potential attackers. Nevertheless there are understandably some security concerns. Using AST objects rather than strings for creating Python code dynamically makes our use ofexec
intrinsically safer than some of these other projects, and we also use some defence-in-depth techniques.When using the compiler as opposed to the resolver, the additional up-front processing of the FTL messages could incur a noticeable startup cost (typically of the order of .5ms per message). For long running Django processes this is usually a very good trade-off given the performance benefits.
'Escaper' functionality (see escapers branch). This allows us to handle embedded HTML correctly. The mechanism for doing this has been implemented as part of python-fluent in a generic way, and then used in django-ftl to handle HTML escaping in templates with a minimum of work.
Tools used in rendering this package: