Localization
Localization at Mozilla
At Mozilla localizations are managed by locale communities around the world, who are responsible for maintaining high quality linguistic and cultural adaptation of Mozilla software into over 100 locales.
The exact process of localization management differs from project to project, but in the case of Gecko applications, the localization is primarily done via a web localization system called Pontoon and stored in the firefox-l10n GitHub repository.
Developers are expected to keep their code localizable using localization and internationalization systems, and also serve as localizers into the en-US locale which is used as the source locale.
In between the developers and localizers, there’s a sophisticated ecosystem of tools, tests, automation, validators and other checks on one hand, and management, release, community and quality processes facilitated by the Localization Team, on the other.
Content vs. UI
The two main categories in localization are content localization vs UI localization.
The former is usually involved when dealing with large blocks of text such as documentation, help articles, marketing material and legal documents.
The latter is the primary type when handling user interfaces for applications such as Firefox.
This article will focus on UI localization.
Lifecycle & Workflow
1) New feature
The typical life cycle of a localizable UI starts with a UX/UI or new feature need which should be accompanied by the UX mockups involving so called copy - the original content to be used in the new piece of UI.
2) UX mockup + copy review
The UX mockup with copy is the first step that should be reviewed by the Localization Team. Their aim is to identify potential cultural and localization challenges that may arise later and ensure that the UI is ready for localization on a linguistic, cultural, and technical level.
3) Patch l10n review
Once that is completed, the next stage is for front-end engineers to create patches which implement the new UI. Those patches should already contain the copy and place the strings in the localization resources for the source locale (en-US).
The developer uses the localization API by selecting a special identifier we call L10n ID and optionally a list of variables that will be passed to the translation.
We call this “a social contract” which binds the l10n-id/args combination to a particular source translation to use in the UI.
The localizer expects the developer to maintain the contract by ensuring that the translation will be used in the given location, and will correspond to the source translation. If that contract is to be changed, the developer will be expected to update it. More on that in part 6) String Updates.
The next review comes from either the Localization Team, or experienced front end engineers familiar with the internationalization and localization systems, making sure that the patches properly use the right APIs and the code is ready to be landed into mozilla-central.
4) Exposure in firefox-l10n-source
Once the patch lands in mozilla-central, strings won’t be exposed to localization until they are merged into firefox-l10n-source, a unified repository that includes strings for all shipping versions of Firefox (nightly, beta, release, ESR, etc.).
Twice a day, automation in this repository extracts new strings from gecko-dev into a separate update branch. This branch acts as a buffer to avoid exposing potential issues to over 100 locales. The Localization Team will take a final look at the localizability of the introduced strings. In case of issues, developers might be asked to land a follow up, or the patch could be backed out with the help of sheriffs.
As a last step, strings are merged into the main branch which is exposed to our web localization tool Pontoon.
5) Localization
Once strings are exposed to Pontoon, localizers will work on providing translations for the new feature either while the new strings are only in Nightly or after they are merged to Beta. The goal is to have as much of the UI ready in as many locales as early as possible, but the process is continuous and we’re capable of releasing Firefox with incomplete translations falling back on a backup locale in case of a missing string.
When translation completes in Pontoon, the final localized strings are stored in firefox-l10n.
6) String updates
Later in the software life cycle some strings might need to be changed or removed. As a general rule, once the strings lands in mozilla-central, any further update to existing strings will need to follow these guidelines, independently from how much time has passed since previous changes.
If it’s just a string removal, all the engineer has to do is to remove it from the UI and from the localization resource file in mozilla-central.
If it’s an update, we currently have two “levels” of change severity:
1) If the change is minor, like fixing a spelling error or case, the developer should update the en-US translation without changing the l10n-id.
2) If the change is anyhow affecting the meaning or tone of the message, the developer is requested to update the l10n string ID.
The latter is considered a change in the social contract between the developer and the localizer and an update to the ID is expected.
In case of Fluent, any changes to the structure of the message such as adding/removing attributes also requires an update of the ID.
The new ID will be recognized by the l10n tooling as untranslated, and the old one as obsolete. This will give the localizers an opportunity to find and update the translation, while the old string will be removed from the build process.
There is a gray area between the two severity levels. In case of doubt, don’t hesitate to request feedback of review from the Localization Team to avoid issues once the strings land.
Selecting L10n Identifier
Choosing an identifier for a localization message is tricky. It may seem similar to picking a variable name, but in reality, it’s much closer to designing a public API.
An l10n identifier, once defined, gets associated to a translated message in every one of 100+ locales and it becomes very costly to attempt to migrate that string in all locales to a different identifier.
Additionally, in Fluent an identifier is used as a last resort string to be displayed in an error scenario when formatting the message fails, which makes selecting meaningful identifiers particularly valuable.
Lastly, l10n resources get mixed and matched into localization contexts where it becomes important to avoid identifier collision from two strings coming from two different files.
For all those reasons, a longer identifier such as privacy-exceptions-button-ok is
preferred over short identifiers like ok or ok-button.
Localization Systems
Gecko has two main localization systems: Fluent and StringBundle, a legacy system.
Fluent
Fluent is a modern localization system designed by Mozilla to address the challenges and limitations of older systems.
It’s well suited for the modern web development cycle, providing a number of localization features including a good internationalization model and strong bidirectionality support.
To learn more about Fluent, follow the Fluent for Firefox Developers guide.
StringBundle
StringBundle is a runtime API used primarily for localization of C++ code. The messages are stored in .properties files and loaded using the StringBundle API and then retrieved from there via imperative calls.
The system provides external positional arguments which can be placed into the string. Adding new StringBundle messages should only be done after serious consideration.