Internationalisation
Getting Started |
THIS IS CURRENTLY A DRAFT
Translations in Ubuntu (from a developer perspective)
This document tries to outline:
- How Ubuntu deals with translations (and i18n in general) from a developer's perspective
- Which tools are interesting for developers to make sure that translations and i18n work fine
General information
Be sure to read the TranslationLifecycle in the translations wiki pages for a general overview of how translations work in Ubuntu.
Packaging
Language Packs
General
Packages in main and restricted don't contain translations (*.mo) files themselves, they are stripped during the build on the Launchpad buildds and put into language-pack-* packages instead. This is done in an attempt to bundle the translations you're interested in and cut down the occupied space.
Note: once a package is promoted to main, it only needs to be rebuilt the first time after promotion so that pkgbinarymangler can strip the translations. After this, import to Launchpad Translations and export in language packs "will just happen".
Translations Templates
For pkgstriptranslations (in the pkgbinarymangler package) to do the job right, please make sure that your package in main or restricted builds a .pot during the build. It does not necessarily need to be shipped in the source or in the binary package. Generating it during the build is good enough. This can be achieved by running the following in debian/rules:
- For GNOME packages:
You should include the gnome.mk CDBS class, which in turn includes the langpack.mk one, which contains the rules necessary to build the templates.
Alternatively, if your package does not use CDBS, you should run intltool in debian/rules to extract strings and generate the template on build:
cd po; intltool-update -p
If there's more than one ./po/ directory (or it has a different name), please make sure to adjust the call above.
- For KDE packages:
- TODO
- For GNU packages:
- TODO: check
make update-po
Other translation files
Desktop Entries
Translations of desktop entries (.desktop files) are also stripped out and included in language packs.
Contrary to what other distros do (except currently OpenSUSE), the translations are not expected to be in the .desktop files themselves (static translations), but rather in the same .mo file of the application, from which they are then loaded at runtime. We do this in order to be able to translate them in Launchpad and to ship them in language packs, so that they can be edited and updated by Ubuntu translators.
The runtime loading is done by a patch in glib (or in kdelibs for Kubuntu), which recognises the the X-GNOME-Gettext-Domain additional key we add to all Ubuntu .desktop files and adds gettext support when loading them. Note that if the .desktop file already contains translations, these static translations will be preferred.
As an example for this key in action, here's how it is done in the Transmission source package in Jaunty (from the 20_add_X-Ubuntu-Gettext-Domain.diff patch):
Index: transmission-1.40-2ubuntu1/gtk/transmission.desktop.in =================================================================== --- transmission-1.40-2ubuntu1.orig/gtk/transmission.desktop.in 2008-12-09 18:57:13.000000000 +0000 +++ transmission-1.40-2ubuntu1/gtk/transmission.desktop.in 2008-12-09 18:58:54.000000000 +0000 @@ -9,3 +9,4 @@ Type=Application MimeType=application/x-bittorrent; Categories=Network;FileTransfer;P2P;GTK; +X-GNOME-Gettext-Domain=transmission
Note, however, that if the package is using the `langpack.mk` CDBS class (or the gnome.mk class, which includes it) it will generally not be necessary to use a patch for this, as langpack.mk will take care of it.
Also note that X-GNOME-Gettext-Domain used to be called X-Ubuntu-Gettext-Domain in the past.
In summary, for packages in main we strip the translations out of the .desktop files, and gnome-panel (TODO: is this the right package?, TODO: how is this done in KDE?) has a patch to load a translation from the .mo file. It basically uses that key to figure out where to get the translation from.
Additional information:
Schema files
Policy files
Interactive upgrade hooks
After certain upgrades (most notably upgrades between Ubuntu releases), there are follow-up actions which should be taken by the user. These are not logically part of the upgrade, but should be dealt with at some point after the upgrade is complete.
They take the form of helper scripts that can be called by the package's postinst script, and can be internationalized through language packs.
Here's an example showing how these can be internationalized:
_Name: The great UTF-8 Migration Priority: Medium Command: bc Terminal: True GettextDomain: hwdb-clinet DisplayIf: shell-command Title: Change the default title here DontShowAfterReboot: False _Description: This command will convert your stuff into UTF-8. Use this command if you want a working gnome desktop and you feel the world should be a better place.
This is a notification description file as it should be put into a source package. NB the '_' prefix for fields that should be translated. For getting these strings into the package's PO template, you should add a stanza like this to po/POTFILES.in:
[type: gettext/rfc822deb] debian/utf8-migration.installnote
The file should be shipped in /usr/share/packagename and the package's postinst script should install it if the notification should be displayed. At that time it needs to remove the '_' prefixes:
install -d /var/lib/update-notifier/user.d/ sed 's/^_//' /usr/share/utf8-migration-tool/utf8-migration.installnote > /var/lib/update-notifier/user.d/utf8-migration
Verifying translation uploads
This information is intended to help package maintainers to optimize their packages, so that translation related files get automatically imported into Launchpad.
Testing the package locally
To test if your package conforms to the requirements of the import script in Launchpad, please do the following:
In /etc/pkgbinarymangler/striptranslations.conf set enable: to true
Optionally, install any build dependencies: sudo apt-get build-dep $PKGNAME
Build the package with debuild -uc -us -b
You will find a <pkgname>-translations.tar.gz tarball along with the .deb package(s)
- Untar that tarball into a temporary directory
- Compare if the directory structure and the .pot and .po files conform to the requirements listed below
Guidelines
If the application uses gettext, like most GTK and GNU applications do:
- The filename of the template (.pot file) should be the same as what the expected translation domain is.
The actual translations (.po files) must have a filename of the format $LANGUAGECODE.po or $LOCALECODE.po (e.g. de.po, pt_BR.po, zh_TW.po)
- All translation-related files (the .pot template and the corresponding .po templates) must end up in the same (sub-)directory.
- If the application has multiple translation domains (i.e. multiple .pot files), each .pot file and their corresponding .po files should end up in a separate subdirectory (e.g. po/nautilus/nautilus.pot and po/libnautilus/libnautilus.pot)
Templates and translations for the main application or library should go into po/.
Templates and translations for help files should go into help/po/. They will be ignored by Launchpad, since we currently have no infrastructure to deal with them properly.
- Templates and translations for documentation files should go into doc/po/. They will also be ignored by Launchpad for the same reason as the help translations.
- If there are .po files present but no associated .pot template, please remove them from the package. They are useless without a .pot template. There might be a bug in the package or in the source code whereby the .pot file doesn't get generated, so make sure to double-check that as well.
- Please check if the .pot file has any meaningful content. Empty .pot files should be removed and the package build rules be fixed.
- If your package contains patches and those patches result into extra templates, like 'patches.pot', please merge those changes into the main template, as we have no use for patch .pot files. Such .pot files should not exist in the final tarball.
- If the package generates multiple .pot files and later merges them into one, only the final .pot file should be present in the final tarball. The other ones are useless.
- If the package creates the same .pot file in architecture-dependent subdirectories, please consider to merge them, as we can only deal with one single .pot per translation domain.
- Test routines and test files should not generate any .pot template or ship any .po file. If such files are present, please remove them.
If you have multiple .po files for a language, each in a different encoding, please consider renaming the ones encoded in UTF-8 to be the main .po file for that laguage and discard the other ones. They might be interesting for Windows users, but not for Linux users, since translations get converted with iconv into the encoding the user is runnning anyway. And nowadays UTF-8 is the default encoding on Linux. Example: if you have zh_TW.po (Big5 encoding), zh_TW.UTF-8.po and zh_TW.cp960.po, please remove zh.po and zh.cp960.po and rename zh_TW.UTF-8.po to zh.po.
The no locale is obsolete. If you find no.po files, please look into it and find out if it's Norwegian Bokmål or Nynorsk. Usually it's Bokmål. Please rename the file to the proper language code: nb.po for Bokmål, nn.po for Nynorsk. If nb.po and nn.po are already present, remove no.po.
If you have an en_US.po file, please investigate if it really contains translations (i.e. if the source code uses en_GB or another language) and if not, you can safely discard it.
Glossary of i18n-related terms
Translation domain (domain)
The translation domain is a unique string assigned by the programmer in the code (usually in the build system) and used by the gettext functions to locate the message catalog where translations will be loaded from. The general form to compute the catalog's location is:
dir_name/locale/LC_category/domain_name.mo
which in Ubuntu expands to either /usr/share/locale/$LOCALE/LC_MESSAGES/domain_name.mo for translations not included in language packs (not in main) or /usr/share/locale-langpack/$LOCALE/LC_MESSAGES/domain_name.mo for those shipped in language packs. $LANGUAGE is generally the ISO 639-1 2-letter code (e.g. 'ca' or 'de') or ISO 639-2 3-letter code (e.g. 'ast') for the particular language. As an example, when using Nautilus in a Catalan locale with Ubuntu, the gettext functions will look for the message catalogue at:
/usr/share/locale-langpack/ca/LC_MESSAGES/nautilus.mo
The corresponding translation template should have the same translation domain in its filename, e.g. 'nautilus.pot'. The translation domain must be unique across all applications and packages. I.e. something generic like 'messages.pot' won't work.
Translation template (POT file, PO template file)
Translation (PO file)
Message catalog (MO file)
Documentation
TODO:
Converting xml to PO with xml2po
- Launchpad import
Packages not using the Gettext format
Mozilla
OpenOffice.org
Coding
Individual programming languages
Additional documentation
Translation process
TODO
- String freeze
- Translation
- Translation deadline
- Language pack updates
Launchpad
The import queue
All translation templates (.pot files) and translations (.po) end up in the Launchpad Translations import queue before being imported into Launchpad.
Some of the Ubuntu import queues:
Templates generally get approved automatically, but sometimes there is some manual work involved in approving them. The Ubuntu Translations Coordinators Team manages the import queue. We've got a mailing list where issues can be discussed and we welcome the help of any community members experienced with translations.
If your project is hosted in Launchpad
General Advice
If you want to make sure that your packages translations are set up correctly, try running:
- For GNOME packages:
cd po; intltool-update -m
and be sure to either fix the issues or report them upstream.
Note: this is only worth trying if the package uses intltool for i18n (i.e. look for intltool.m4 in the top-level source directory)
- For GNU packages:
- For KDE packages:
Sometimes empty .po files caused packages to build from source.