Multilingual Templates and Modules

Languages:

The Synchronizer tool is used to keep modules updated across wikis.

This page explains how to create global, cross-wiki and multilingual modules and templates, and how to keep them synchronized across Wikimedia wikis.

Why is this needed? Because we do not have a single Wikipedia, we have 300+ separate Wikipedias and other wiki projects, and every time someone creates a good new template or Lua module, it gets copied and translated 300+ times. Every translator has to thoroughly understand MediaWiki markup, making copying a very tedious and error-prone process, partially because template authors often assume their templates will be used in just one language. Once copied, the original templates are often improved, and each copy has to be updated while maintaining all existing translations. The pure human expense of copying templates and modules is so high that most of them are either never copied or never updated, especially for the smaller wikis.

Is this the best approach? No, but it is the best approach with the current technology. A significant MediaWiki rewrite is required to make this possible on the system level. Multi-site templates has been requested from the start in 2001, but not much was done simply because this is a hard problem to solve. With this approach it is possible to create multilingual content now, and once MediaWiki supports it, we can easily migrate multilingual content to the new system without much work.

General method[edit]

A global module is a Lua module designed to be used with exactly the same code in every wiki.
To become global, a module must provide ways for wikis to localize everything they need to localize, without having to touch the code of the module itself. This page describes several techniques to accomplish this.
Once a module is globalized, the Synchronizer tool can be used to keep it updated across Wikimedia wikis.
To make a global template, turn it into a global module.

Best practices[edit]

This section describes some of the current best practices to develop global modules.

Naming[edit]

This section can be ignored for modules designed to be called from templates only.

Global modules that are meant to be used by other modules should be named the same in all wikis to avoid dependency breaks with other global modules. For example, if a module named A requires a module named B, but in some wiki, module B is named C, then module A will not work in that wiki, unless the source code of module A is changed locally to require C instead of B, which would defeat globalization (of module A).

If a local community does not accept the global name, or renaming is too much trouble, then a workaround is to create a "redirect module" with the global name, that simply requires and returns the module with the local name.

Fortunately, the fact that the Module namespace is named differently in each language doesn't break dependencies, because "Module" is an alias for the Module namespace in all languages.

Master module[edit]

Global modules should pick one wiki where to do the development. Generally this will be the home wiki of the module, but it may migrate for various reasons, for example to increase the chances of recruiting new developers by centralizing the development in a larger or more appropriate wiki.

Global modules should start with a comment that includes a link to the master module, to reduce the chances of forking and help recruit new developers (example).

Sandbox[edit]

Global modules should have a /sandbox subpage where to test out changes before deploying them to the master module and the other wikis (example).

Testcases[edit]

Global modules should have a /testcases subpage with good unit tests to ensure high quality and stability of the module (example). Test cases should:

Use Module:ScribuntoUnit
Run with both the main module and the sandbox versions, so that we can compare the results (example)
Use require('strict') to avoid accidentally using non-declared variables
Output results both in /testcases/doc and the main /doc page of the module, to catch errors as early as possible

Documentation[edit]

Global modules should have a /doc subpage with documentation of all public functions of the module (example) and a section with the testcase runs for both the primary and the sandbox versions of the module (example).

Configuration[edit]

Global modules that require configuration should have a separate /config submodule to prevent local wikis from modifying the source code to configure the module (example).

Synchronization[edit]

Once a module is able to be copied unchanged to other wikis, the Synchronizer tool can be used to copy it and keep it synced across all Wikimedia wikis.

Backwards compatibility[edit]

Global modules should generally keep development backwards-compatible, because changes that are not backwards-compatible will often require manual updates to each and every wiki, template and module that uses the module.

Localization of template parameters[edit]

Global modules can have their parameters localized by the template callers. For example, consider the following module that simply outputs the given text (or "Example" if none is given):

local p = {}

function p.main(frame)
 	 local args = frame.args
 	 local text = args['text'] or 'Example'
 	 return text
end

return p

Then a Spanish template would localize the module like so:

{{#invoke:Example|main
| text = {{{texto|Ejemplo}}}
}}

Notice that the template not only localizes the name of the "text" parameter ("texto" means "text" in Spanish), but also the default text ("Ejemplo" means "Example" in Spanish).

See Plantilla:Extracto for a real case of a template that localizes a global module with this technique. Also, see Template:Excerpt for a case where a global module is localized to the English Wikipedia, demonstrating that localization is not always the same as translation.

Localization of user-readable strings[edit]

Many modules need to output user-readable strings, such as error messages and interface elements (like buttons). Hard-coding the text of these strings forces other wikis to modify the code in order to localize them, preventing globalization. To avoid this, developers should provide ways to localize user-readable strings without having to modify the code itself. This section explains several ways to accomplish this.

Template parameters[edit]

User-readable strings can be localized through template parameters when calling the module. This approach is convenient when:

The text is likely to vary with each template call
The text is likely to be changed by users when calling the template
The text is likely to contain a magic word, a template call, a parser function or some other wiki element

An example of a module using this approach would be:

local p = {}

function p.main(frame)
 	 local args = frame.args
 	 local text = args['text'] or 'Example'
 	 return text
end

return p

This way, every template may modify the text when calling the module, like so:

{{#invoke:Example|main
| text = Ejemplo
}}

Notice that in this example, if a template calls the module without specifying the text parameter, then the hard-coded English text 'Example' would be used. This is not necessary. Modules may require template callers to set the text parameter by throwing an error if they don't. However, it's often friendlier to have a fallback.

Config file[edit]

Another way to localize user-readable strings is through a separate /config subpage. This approach is convenient when:

The module is meant to be called by many templates per wiki, thus allowing localization to be done only once and then reused
There're many messages to localize, so it's easier to have them all together in their own place
There's already a need for a /config file for other reasons, so we might as well use it for localization too

An example of a module using this approach would be:

local config = require('Module:Example/config')

local p = {}

function p.main(frame)
 	 local text = config.text or 'Example'
 	 return text
end

return p

Then wikis would be able to create /config files like the following:

return {
 	 text = 'Ejemplo'
}

Translation tables[edit]

Another way to localize user-readable strings is through a central translation table at Commons. This approach is convenient when:

The strings should vary with the preferred language of the user, rather than the language of the wiki or page.
We want to centralize localization efforts on a single page.

The Module:TNT was created specifically to get strings from translation tables. An example module using TNT could look like this:

local TNT = require('Module:TNT')

local p = {}

function p.main(frame)
 	 local text = TNT.format('I18n/Example', 'text')
 	 return text
end

return p

See Data:I18n/Template:Graphs.tab for a simple but real example of a translation table with two messages, each having a single parameter. It's important to store parameters as parts of the strings because in many languages the parameter would have to be placed at a different position in the string according to the norms of the language.

Translation tables should start with the "Data:I18n/..." prefix to separate them from other types of tabular data. If a message has not yet been localized, TNT will fall back to English (or other fallback language as defined by the language's fallback sequence ). TNT also supports all standard localization conventions such as {{PLURAL|...}} and other parameters .

One downside of this approach is that it requires installing and setting up Extension:JsonConfig , which may not have been done on non-Wikimedia wikis, limiting the ability to reuse these modules on third-party wikis.

MediaWiki messages[edit]

In some cases, MediaWiki itself (or some extension) may have the messages we need already localized. For example, if we need the string "New page" we may use MediaWiki:Newpage, like so:

local p = {}

function p.main(frame)
 	 local msg = mw.message.new('newpage')
 	 local text = msg:plain()
 	 return text
end

return p

See Special:AllMessages for a list of all available messages.

All of the above[edit]

Depending on the case, all of the above methods may be combined. For example, MediaWiki messages may be used when available, and when not, a translation table or config file is queried, and if no localization is found there, then a hard-coded English text is used, unless a template parameter overrides it.

Combining several methods can be effective, but the benefits should be weighted against the downsides of the increased complexity, which may cause performance loss and bugs, as well as more difficulty in maintaining the code and recruiting new developers.

Template data[edit]

Template parameters are usually stored as a JSON templatedata block inside the template's /doc subpage. This makes it convenient to translate, but when a new parameter is added to a global template, all /doc pages need to be updated in every language. Module:TNT helps with this by automatically generating the templatedata block from a table stored on Commons. Placing the following line into every /doc subpage will use Data:Templatedata/Graph:Lines.tab table to generate all the needed templatedata information in every language. Even if the local community has not translated the full template documentation, they will be able to see all template parameters, centrally updated.

{{#invoke:TNT|doc|Graph:Lines}}