Edgewall Software

Localization (L10N) of Trac

Trac uses Babel for localization. Starting with version 0.12, Trac contains the i18n framework and L10N message catalog files.

From the user point of view, the Preferences / Language panel can be used to select the language from the list of available translations, otherwise the web browsers' language preferences will be honored, if possible.

If you're only interested in using a localized version of Trac, then nothing more than a standard TracInstall is needed, provided you installed Babel before.

The more complete installation instructions below are aimed at developers and translators, who want to contribute translations for Trac 0.12.x and beyond.

Note that this whole page is dedicated to the technical details of translating the Trac web application. There's also a parallel effort to translate the TracGuide, see TracGuideTranslations. While there's much more "text" to translate there, there are also much less technical constraints, so if you're intimidated by what follows but nevertheless would like to contribute back to Trac by providing your language expertise, here's a nice alternative!

Installation

Getting the Tool Chain

Python Tools

  1. Install Babel version 0.9.6 or newer
    Read the detailed installation instructions for Babel if you want to build it from source, but you should also be able to install it with a simple easy_install or by using one of the packages in the Download page linked above.
  2. Likewise, install Genshi version 0.6, using easy_install or by using one of the packages available for download.
  3. Get a source code checkout for the Trac release line you want to translate:
    • trunk for 1.1.x,
      e.g. svn checkout http://svn.edgewall.org/repos/trac/trunk trac-1.0dev
    • branches/1.0-stable for 1.0.x,
      e.g. svn checkout http://svn.edgewall.org/repos/trac/branches/1.0-stable trac-1.0dev
    • branches/0.12-stable for 0.12.x,
      e.g. svn checkout http://svn.edgewall.org/repos/trac/branches/0.12-stable trac-0.12.x

All the discussion and examples below refer to trunk, but everything will work out exactly the same way for the 0.12-stable branch or any other more recent branch.

Translation Workflow

It is very important to not waste time translating an out-of-date catalog.

In the repository, only the catalog templates (trac/locale/messages.pot and trac/locale/messages-js.pot) get updated regularly by Trac developers. If we would do so for the all the .po catalogs, this would generate lots of big and noisy changesets. So we leave that task up to the translators, which is something they should do before starting the actual translation work or before integrating updates coming from contributors when they're acting as translation team coordinators.

There are several different procedures to get the catalogs in sync with the templates, choose the one you're more comfortable with.

Using Python tools

  1. Run python setup.py extract_messages.
    This will generate the catalog template file: source:trunk/trac/locale/messages.pot
    You should never have to edit that file manually.
    Note: this step is for committers only. If you're not a Trac svn committer, you have to use the available messages.pot, or bug us to update that file (though we do it on a regular basis).
  2. Run python setup.py update_catalog.
    This will regenerate the various string catalogs (trac/locale/*/LC_MESSAGES/messages.po), preserving the previously translated strings and merging the new translations found in the messages.pot file. Translations that are no longer needed are commented out (#~) and moved at the bottom of the .po file (be careful, they will be discared by the next update, so if they contain valuable content, save it somewhere).
    Usually, you will be interested in working on one locale only, so you can pass the -l option to specify which catalog should be updated. For example:
    python setup.py update_catalog -l nl.
  3. Now, open your favorite editor and add or modify translations for the extracted strings in the messages.po file for your language.
  4. Once you're done, you should validate your translations:
    1. Run python setup.py update_catalog -l again, for normalizing the manual edits. For example:
      python setup.py update_catalog -l nl.
      This step is also quite useful for spotting the possible mistakes, like accidental changes of msgid strings. Look for lines beginning with #~, they are indicative of such errors. Note that the comments corresponding to translations which are no longer present in the messages.pot file (as discussed in step 2. above) have been discarded by this second update_catalog run.
    2. Run python setup.py check_catalog -f.
      This will search the catalog for common mistakes, like placeholder mismatches, invalid embedding in Genshi's structure messages, etc.
    3. Run python setup.py compile_catalog -f.
      This will generate one compiled catalog (message.mo file) for each (or each specified) messages.po file. You don't need to do anything with those files, they'll get installed automatically.
      You will most probably need to use the force option (-f) in order to compile catalogs marked "fuzzy". It is fine to have fuzzy translations but try to never contribute a translation still containing those markers, as they mean that a given translation has not been reviewed by a human but were machine generated. Fuzzy translations should be either cleared when they are wrong, or validated when they are correct, by removing the fuzzy tag or even the whole line iff fuzzy was the only keyword on that line
      You can also use the -l option here to specify the locale for which the messages.po has to be compiled.
    4. Run python setup.py install for installing Trac and locale data. Alternatively you can use python setup.py develop once for all, if you want to be able to run Trac from your working copy. Beware that there's no on the fly compilation of the catalogs, so any change will have to go through step 4.3 (compile) before being visible in your Trac instance.

If you skip the 4th step (validation), you risk to submit incomplete or, worse, buggy messages.po files. Going through the 4th step will ensure a better quality standard. Finally, seeing your translations "live" in Trac is not only a great way to realize if the translation still makes sense once put in context, but also quite gratifying when you see that you picked the correct wording ;-)

Note that the above was written at a time there was only one catalog file, messages.pot. So in order to be complete, you'd have to repeat this for all the catalogs:

  • messages-js
    • extract_messages_js
    • update_catalog_js
    • check_catalog_js
    • compile_catalog_js
    • generate_messages_js (the last step specific to Javascript i18n)
  • tracini
    • extract_messages_tracini
    • update_catalog_tracini
    • check_catalog_tracini
    • compile_catalog_tracini

If this look tedious and you're thinking about scripting all of this… don't worry, that has already been done, see #using-make below.

Using msgfmt tools

Although Babel is Python specific and is the tool which is advised to use, the traditional msgfmt toolset (part of GNU gettext) can also be used, but preferably for related tasks like checking the catalogs and getting statistics, rather than updating or the compiling the catalogs.

In particular, the following command is quite useful:

$ msgfmt --statistics --check trac/locale/pl/LC_MESSAGES/messages.po
884 translated messages, 166 fuzzy translations, 80 untranslated messages.

All is well, except for some fuzzy and missing translations ;-)

$ msgfmt --statistics --check trac/locale/zh_CN/LC_MESSAGES/messages.po
trac/locale/zh_CN/LC_MESSAGES/messages.po:2155: a format specification for argument 'unit', as in 'msgstr[0]', doesn't exist in 'msgid'
trac/locale/zh_CN/LC_MESSAGES/messages.po:2162: a format specification for argument 'unit', as in 'msgstr[0]', doesn't exist in 'msgid'
trac/locale/zh_CN/LC_MESSAGES/messages.po:2169: a format specification for argument 'unit', as in 'msgstr[0]', doesn't exist in 'msgid'
msgfmt: found 3 fatal errors
1058 translated messages.

Some errors were detected with format specifications.

This is a useful additional check, as neither python setup.py compile_catalog -l zh_CN nor python setup.py check_catalog -l zh_CN wouldn't have detected that issue. FIXME we detect mismatches, but not missing placeholders in msgstr

Using Transifex

The web-based collaborative site for translations Transifex has a project page for Trac and its convenient user interface makes it very attractive for translators who don't want to mess directly with the source code.

There, a number of "resources" (.pot messages catalog files in our case) are grouped in several "releases". Currently the "releases" corresponds to our maintenance branches.

So here's an overview of the Transifex "Releases" and their corresponding "Resources":

Once you're member of a translation team, you can add translations directly online. But in the end, the changes will have to be committed back to the source tree, so you better have to check beforehand with the translator coordinator of your language to see if is comfortable with getting updates from Transifex.

If you are a translation team coordinator, please refer to the Transifex sub page for more details.

The easy way using make and l10n targets

You'll find a Makefile in the toplevel directory of the Trac source. If you're lucky enough to have the GNU make tool, you can benefit from some automation for these translation tasks.

$ make help
                              ...
 ---------------- L10N tasks

  init-xy             create catalogs for given xy locale

  extraction          regenerate the catalog templates

  update              update all the catalog files from the templates
  update-xy           update the catalogs for the xy locale only

  compile             compile all the catalog files
  compile-xy          compile the catalogs for the xy locale only

  check               verify all the catalog files
  check-xy            verify the catalogs for the xy locale only

  stats               detailed translation statistics for all catalogs
  stats-pot           total messages in the catalog templates
  stats-xy            translated, fuzzy, untranslated for the xy locale only

  summary             display percent translated for all catalogs
  summary-xy          display percent translated for the xy locale only
                      (suitable for a commit message)

  diff                show relevant changes after an update for all catalogs
  diff-xy             show relevant changes after an update for the xy locale

  [locale=...]        variable for selecting a set of locales

Example usage:

$ make stats-pot update check compile stats locale='fr de'
translation statistics for messages.pot:
stats-pot: 0 translated messages, 1096 untranslated messages.
python setup.py update_catalog -l fr
running update_catalog
updating catalog 'trac/locale\\fr\\LC_MESSAGES\\messages.po' based on 'trac/locale/messages.pot'
python setup.py update_catalog -l de
running update_catalog
updating catalog 'trac/locale\\de\\LC_MESSAGES\\messages.po' based on 'trac/locale/messages.pot'
checking catalogs for fr de...
check-fr: OK
check-de: OK
All catalogs checked are OK
python setup.py compile_catalog -l fr
running compile_catalog
compiling catalog 'trac/locale\\fr\\LC_MESSAGES\\messages.po' to 'trac/locale\\fr\\LC_MESSAGES\\messages.mo'
python setup.py compile_catalog -l de
running compile_catalog
compiling catalog 'trac/locale\\de\\LC_MESSAGES\\messages.po' to 'trac/locale\\de\\LC_MESSAGES\\messages.mo'
translation statistics for fr de...
stats-fr: 1096 translated messages.
stats-de: 932 translated messages, 164 untranslated messages.

Contributing

The various catalog files (messages.po) can be found in the repository under source:trunk/trac/locale. The messages.pot is the template file from which the various locale-specific files get created.

Adding translations for a new language

If you need to create a catalog for a new locale (e.g. et), do:

$ ./setup.py init_catalog -l et

Use ISO:639-1 language codes, only use national variant when there's a real need.

We also added support for translating messages in Javascript code. Those messages are collected in a separate catalog, so you need a separate command:

$ ./setup.py init_catalog_js -l et

Alternatively, you can do make init-<locale>, this will create the missing catalog(s) if needed.

Then proceed through the steps listed above starting at step 2.

How far would you like to go?

The amount of translatable text could be seen as overwhelming… You don't have to translate everything though in order to provide a good l10n experience to your users.

I think we can distinguish 3 levels:

  1. user-oriented translations: those are all contained in the messages.po and messages-js.po catalogs; if you reach 100% there you're already a hero (Hello Itamar! ;-) )
  2. administrator-oriented translations: since 1.0 we also introduced the possibility to translate the documentation information for the numerous TracIni settings, in the tracini.po catalog; for now those are primarily targeted at Trac administrators, so translating those could really be seen as an "bonus" level
  3. TracGuideTranslations — various people have attempted to translate the TracGuide, but this would really be a huge effort; maybe in the future we could isolate a smaller subset of user oriented pages of the TracGuide
  4. Translating the API documentation (and why not also the source code?). Hey, just kidding ;-)

Statistics

Those statistics reflect the coverage status of the .po files (catalogs) compared to the .pot files (catalog templates). The template catalogs themselves might be out-of-date compared to the source itself, though we regularly perform an automatic extraction.

messages.pot messages-js.pot tracini.pot
0.12-stable http://www.transifex.net/projects/p/trac/resource/0_12-stable-messages-pot/chart/image_png http://www.transifex.net/projects/p/trac/resource/0_12-stable-messages-js-pot/chart/image_png
1.0-stable http://www.transifex.net/projects/p/trac/resource/1_0-stable-messages-pot/chart/image_png http://www.transifex.net/projects/p/trac/resource/1_0-stable-messages-js-pot/chart/image_png http://www.transifex.net/projects/p/trac/resource/1_0-stable-tracini-pot/chart/image_png
trunk http://www.transifex.net/projects/p/trac/resource/trunk-messages-pot/chart/image_png http://www.transifex.net/projects/p/trac/resource/trunk-messages-js-pot/chart/image_png http://www.transifex.net/projects/p/trac/resource/trunk-tracini-pot/chart/image_png

Charts kindly provided by https://ds0k0en9abmn1.cloudfront.net/static/charts/images/tx-logo-micro.646b0065fce6.png, based on the content of the catalogs uploaded there, which should correspond to the content of the catalogs in the repository.

Note that if the statistics for a language suddenly drop to 0%, this is indicative of the presence of fatal errors (as reported by msgfmt --check) in the corresponding messages.po file.

If you want to generate those statistics by yourself, do make summary. Be aware that those statistics can only be accurate if the catalogs are actually up to date with respect to the catalog template, so you should do a make update before. As this is quite expensive, you may want to do only make update summary locale="..." for the locales you're interested in.

Editing Guidelines

There are tons of good guidelines on the web about how to write good translations. See for example: PO_Odyssey and step-by-step, and other resources from http://l10n.kde.org. (The references also applies to translators using Windows, though the set of translation editors seems to be narrower. However, see Poedit for a multiplatform dedicated PO editor.)

In short, a translation unit is typically something like:

#: trac/ticket/templates/milestone_delete.html:41
#: trac/ticket/templates/milestone_view.html:105
msgid "Delete milestone"
msgstr "Supprimer le jalon"
  • the "#:" lines are comments indicating the locations of the message in the source.
  • msgid "..." is the message itself, this should never be changed
  • msgstr is your translation; you should keep there the same structure, as in the msgid. In particular the presence of "\n" at the beginning or the end of the msgid should be mirrored in msgstr
  • right after an automatic catalog update you'll see lots of:
    • # fuzzy comments: this identifies translations that are only approximated (the msgid has changed)
    • msgstr "" translations: those are newly added translations

There are a few more complex situations.

Here's a message containing parameters:

#: trac/ticket/templates/milestone_view.html:26
#, python-format
msgid ""
"Completed %(duration)s ago\n"
"              (%(date)s)"
msgstr ""
"Atteint il y a %(duration)s\n"
"                (%(date)s)"

Note that you need to keep the same set of parameters in msgid and msgstr, both for the name and the keyword (i.e. don't replace a %(id)s with a %(id)d), although you can reorder them as needed.

Another example, here's how a structured message looks like:

#: trac/ticket/templates/milestone_delete.html:45
msgid ""
"[1:Note:] See\n"
"      [2:TracRoadmap] for help on using\n"
"      the roadmap."
msgstr ""
"[1:Note :] consultez [2:TracRoadmap] pour obtenir de l'aide sur "
"l'utilisation de la feuille de route."

The [1: ...] groups are used when the message contains markup (this is Genshi specific). It is important to keep the same semantic structure. As with message parameters, the groups can be reordered but you should have the same groups in msgid and msgstr (see for example r9553). The various check_catalog_* commands (make check) now look for such errors.

Of course, you can have messages that combine markup groups and parameters names:

#: trac/ticket/templates/milestone_view.html:32
#, python-format
msgid ""
"[1:%(duration)s late]\n"
"              (%(date)s)"
msgstr ""
"[1:%(duration)s en retard]\n"
"                (%(date)s)"

Finally, there's the translation of sentences which have a singular and a plural form:

#: trac/ticket/templates/query.html:29
#: trac/ticket/templates/query_results.html:30
#: trac/ticket/templates/report_view.html:94
#, python-format
msgid "%(num)s match"
msgid_plural "%(num)s matches"
msgstr[0] "%(num)s résultat"
msgstr[1] "%(num)s résultats"
  • when such a message is newly added, you'll have
    #, fuzzy, python-format as the first comment. Be careful to remove only the fuzzy, part, but not the python-format keyword.
  • some languages (zh, ko, ja) don't have plural forms (plural == 1), so you only have to put a msgstr[0] line.
  • some other languages have more than 2 plural forms, so you need as many msgstr[] as needed.

Terms (Definitions)

Consistent and careful translation of terms like timeline, ticket, report is very important. These terms are used everywhere and must be easy to remember and comfortable to use.

The way to make up good translations of important terms is to discuss them before using everywhere. The easiest way to accomplish it is to set up wiki pages for different languages.

Look also at the various term definition pages: Am, Cy, Cz, De, El, Es, Fa, Fi, Fr, Ge, Gl, He, Hu, It, Ja, Ko, Lv, Nl, Pl, Pt, PtBr, Ru, Sv, Tr, ZhCn, ZhTw.

Translation Coordination

For the various languages there are already various tickets logged in which the work on them is tracked:

#1513
Multi-language support in the wiki
#11114
Translation of Trac to Arabic/عربى [ar_EG]
#7186
Translation of Trac to Argentinian Spanish/español de Argentina [es_AR]
#8991
Translation of Trac to Armenian [hy_AM]
#7353
Translation of Trac to Catalan/Català [ca]
#5472
Translation of Trac to Chinese/中文 [zh_CN]
#5592
Translation of Trac to Czech/Česky [cs_CZ]
#5473
Translation of Trac to Dutch/Nederlands [nl_NL]
#7417
Translation of Trac to English [en_GB]
#7578
Translation of Trac to Estonian/eesti [et_EE]
#6382
Translation of Trac to Farsi/پارسی [fa_IR]
#7178
Translation of Trac to Finnish/suomi [fi_FI]
#5474
Translation of Trac to French/Français [fr_FR]
#7918
Translation of Trac to Galician/Galego [gl_ES]
#5475
Translation of Trac to German/Deutsch [de_DE]
#5481
Translation of Trac to Greek/Ελληνικά [el_GR]
#9040
Translation of Trac to Hebrew [he_HE]
#5486
Translation of Trac to Hungarian/Magyar [hu_HU]
#5489
Translation of Trac to Italian/Italiano [it_IT]
#5471
Translation of Trac to Japanese/日本語 [ja_JP]
#5477
Translation of Trac to Korean/한국어 [ko_KR]
#7177
Translation of Trac to Norwegian/bokmål [nb_NO]
#5480
Translation of Trac to Polish/Polski [pl_PL]
#5476
Translation of Trac to Portuguese/Português [pt_BR]
#5470
Translation of Trac to Portuguese/Português [pt_PT]
#8008
Translation of Trac to Romanian/Română [ro_RO]
#5478
Translation of Trac to Russian/Русский [ru_RU]
#7487
Translation of Trac to Slovene/Slovensko [sl_SI]
#5469
Translation of Trac to Spanish/Español [es_ES]
#5487
Translation of Trac to Swedish/Svenska [sv]
#5482
Translation of Trac to Traditional Chinese/正體中文 [zh_TW]
#5488
Translation of Trac to Turkish/Türkçe [tr_TR]
#6629
Translation of Trac to Vietnamese/Tiếng Việt [vi_VN]
#9259
Translations to Esperanto [eo]

This is a way by which translators for the same language can coordinate their work.

It is also important to subscribe to the Trac-dev MailingList, in order to be notified of the release schedule, of the "string freeze" periods shortly preceding a release. During a string freeze we don't add new messages or modify existing ones, so this is the ideal period for finalizing a translation.

For Committers

If you have commit access to the translation catalogs in the SubversionRepository, then you should be careful to commit only error-less updates. Translations can be fuzzy or missing, that's not a problem, but the errors spotted by make check compile locale=<yourlocale> should really be fixed before committing. Please also take care of fixing any [n:...] nestings errors reported by make check, as those mistakes are quite deadly (#9171).

The commit message should have the following format:

<branch-name>/l10n/<yourlocale>: <some message> (<xy>%)

<additional text if needed>

<branch-name> can be omitted for trunk.

The percentage being the ratio of:

# translated messages from make stats-<yourlocale>
# untranslated messages from make stats-pot

Note that you can now easily get a template for the above by issuing the following command:
make summary-yourlocale

When you're adding a new catalog (make init-...), be sure to set "native" line-endings before committing (svn pset svn:eol-style native trac/locale/.../LC_MESSAGES/....po).

Now that all the maintenance branches have l10n support, special care should be taken when merging l10n changes from one branch to the next.

One problem to be expected with the merge is a high rate of conflicts due to line number differences. Even if the source text (the msgid) are often unchanged, it's very likely that their position in the file will be different.

The trunk/contrib/l10n_revert_lineno_conflicts.py can be used to remove those spurious conflicts.

Example workflow:

1.0-stable$ make update
1.0-stable$ svn merge ^/branches/0.12-stable --accept postpone
1.0-stable$ $ svn st -q | sed -e 's,\\,/,g' | grep ^C | cut -c2- > conflicted.files
1.0-stable$ python contrib/l10n_revert_lineno_conflicts.py `cat conflicted.files`
reverted 6 ignorable changes in trac/locale/fa/LC_MESSAGES/messages.po
reverted 1 ignorable changes in trac/locale/nb/LC_MESSAGES/messages.po
reverted 147 ignorable changes in trac/locale/nl/LC_MESSAGES/messages.po
no ignorable changes in trac/locale/nl/LC_MESSAGES/messages-js.po
[...]

At this point, one can inspect the remaining meaningful conflicts and resolved them manually.

Then:

1.0-stable$ make update

to remove the spurious line number difference, review and commit.

For Developers

Here's a condensed list of a few things useful to know about i18n support when coding:

title attributes
those attributes are collected automatically (i.e. no need for explicit ${_('...')} expressions), but only when the message doesn't contain an expression. In the latter case, they need an explicit ${_('... %(param)s ...', param=...)} wrapping
typographical translation
some languages, like French, have special typographical rules (non breaking space before ":" or ";"), so don't consider that those signs can be left outside of translation scope when part of the text (e.g. use _(";") when ";" is used as a separator, _("%(field)s:") instead of '%(field)s:')
messages consolidation
when adding new messages, always look if there's not already a very close message, and then reuse it (e.g. we now have 7 occurrences of one single "Repository '%(repo)s' not found" message, previously we had 7 messages, each slightly different from the other!)
non-translatable messages
in Genshi templates, you can use xml:lang="en" to mark the text in the element as not to be extracted (r10265)

Tips for translating plugins can be found in CookBook/PluginL10N.

Open Issues

#1513
Multi-language support in the wiki
#4636
Component field sort order doesn't use locale
#5441
Timeline shows incomplete information about status changes for customized workflow
#9472
Workflow actions not translated
#10027
TicketQuery groups without translation
#10339
Created, Modified and Ticket columns are not translated in SQL reports
#11536
Wrong label on wiki diff view between initial version and version 1


See also:

Last modified 18 months ago Last modified on Feb 2, 2013 6:05:22 PM

Attachments (8)

Download all attachments as: .zip