Transifex and Trac

Transifex is a modern, open-source localization platform. It’s a web system which automates the translation workflow for complex international projects.

Transifex is indeed a nice tool and can be seen as much more attractive for potential translation contributors than having to go through the somewhat complex procedure described in the TracL10N page, involving svn, Babel, make, etc.

However…

Things are not so simple, or rather, we should take an extra step to make it really as simple as it looks.

The problem is that we might have two sources of changes, and it will make no good to anyone if the two sources are not in sync:

• a Transifex translator might work on an out-of-date catalog, redoing the work the translator coordinator committed directly in the catalog file in the repository
  • worse, pulling updates from Transifex and blindly committing them could lead to regressions
• conversely, a translator coordinator could easily miss the changes done by another translator on Transifex
  • here as well there's a risk for him to overwrite changes made by Transifex contributors if he pushes changes carelessly

In the following, we'll try to present a workflow which avoids those pitfalls and enable changes to flow in both directions.

See also the related discussion in Trac-dev, ​Trac translations and Transifex.

Roles

Once you're familiar with Transifex (and they have some very good ​documentation on their site), you can apply for membership to one translation team for a language, or create a new team if your language of choice isn't yet represented there (be aware that undertaking a new translation for Trac represents a lot of work).

Translation team coordinator

It's much better if the translation team coordinator on Transifex is also the committer responsible for that language in the Trac project itself.

As a committer, he's ultimately in charge of the synchronization work, pulling changes made by members of the translation team on Transifex and pushing back changes made directly to the repository.

Here's an overview of the different teams and status.

Language name	Repository	Committer	Transifex	Coordinator
1) The languages for which we have a committer and a team coordinator
Chinese (China)	zh_CN​	zengjie	​zh_CN	gnozil
Chinese (Taiwan)	zh_TW​	zclin	​zh_TW	zclin
French	fr​	cboos	​fr	cboos
German	de​	fschwarz	​de	fschwarz
Dutch	nl​	lholst	​nl	trbs
Hebrew	he​	itamaro	​he	Yaron
Italian	it​	lgaifax	​it	seccanj
Japanese	ja​	jomae	​ja	jun66j5 (same ;-) )
Portuguese (Brazil)	pt_BR​	lwatter	​pt_BR	mfcn
Spanish	es​	raragon	​es	codeadict
Spanish (Argentina)	es_AR​	rmorales	​es_AR	cramm (same)
Russian	ru​	aprudnikov	​ru	bishop (same?)
Swedish	sv​	mrelbe	​sv	mrelbe
2) The languages for which we have a committer but no team coordinator
Armenian	hy​	narine_martirosyan	​hy
Catalan	ca​	jordi	​ca
English (United Kingdom)	en_GB​	mrelbe	​en_GB
Esperanto	eo​	wdiestel	​eo
Hungarian	hu​	nzoltan	​hu
Norwegian Bokmål	nb​	osimons	​nb
Slovenian	sl​	vezjakv	​sl
Turkish	tr​	asurion	​tr
3) The languages present in the repository and no committers but have a team coordinator
Czech	cs​		​cs	Tsbook
Estonian	et​		​et	lkraav
Finnish	fi​		​fi	mikkorantalainen
Greek	el​		​el	Evropi
Polish	pl​		​pl	wiget
Spanish (Mexico)	es_MX​		​es_MX	tzkmx
Ukrainian	uk​		​uk	sgalashyn
4) The languages not yet present in the repository but have content and team coordinator
Bosnian			​bs	kenan3008
Danish			​da	jenshyllegaard
5) The languages present in the repository, but no committers and no team coordinator
Galician	gl​		​gl
Korean	ko​		​ko
Persian	fa​		​fa
Portuguese	pt​		​pt
Vietnamese	vi​		​vi
Romanian	ro​		​ro
6) The languages for which there's no content, neither in the repository, nor in Transifex
Afrikaans			​af	pjdelport
Hindi			​hi	Sangeeta
Indonesian (Indonesia)			​id_ID	sumodirjo
Panjabi (Punjabi)			​pa	aalam
Turkish (Turkey)			​tr_TR	zeugma

• for the languages in #group1, the translation maintainers are primarily in charge for doing the synchronization
• for the languages in #group5, I'm currently going through a check before pushing back to Transifex
• for the languages in #group6, there's nothing to do

Translation team member

Updating translations on Transifex is only one part of the story, since you need to get your changes integrated into the source code. For that, you need to inform the translator coordinator for your language (see tickets for translation coordination). She will pull the changes from Transifex using the procedure described below, verify them, and commit them. If the translator coordinator doesn't react in due time or is not active anymore, a member of the Trac team will likely notice the activity on the ticket and he could do the same but without the language verification step, obviously.

It's even a better idea to contact and discuss with the translator coordinator before updating the catalogs on Transifex, as some might not be up to date, and you will risk to duplicate work and make integration harder. Transifex provides all the facilities for smooth communications between team members. There's also the tickets, or plain e-mail on trac-dev@… or directly to the translators.

Synchronization with Transifex

As discussed above, this is normally the responsibility of the translation team coordinator who has the commit permissions on the catalog files in the Trac subversion repository.

Typically you'd do this for approving and committing changes if you're a translation coordinator, but even if you're not a committer you could do this if you want to prepare a patch for submitting those changes for review on a ticket.

Prerequisites

You first need to install the ​Transifex client.

Well, it's usually nothing more than:

easy_install transifex-client

(or one ​variation of the above).

If you're on Windows, you'll also need a ​hack…

Of course, you also need to have a checkout of the source, of the branch where you want to work. branches//0.12-stable​ will be maintained for the foreseeable future, so you may want to continue contributing there, but branches//1.0-stable​ is now the official stable release branch, so don't neglect it. trunk​ is where the new interesting stuff happen (TracDev/Proposals/ConfigEnumTranslation?).

Now let's go through a complete example. I made one update to the resource corresponding to the messages catalog on 0.12 for the ​French translation (more specifically, on that page: ​Transifex:resource/0_12-stable-messages-pot/l/fr/).

Checking the status

As I tried to emphasize in the previous sections, what is really important is to not lose translations, in one direction or the other. Well, at least on the Trac side we have version control, which doesn't seem to be the case on the Transifex side…

Back to the command line, for checking the status. First thing to do is to verify that everything is correctly configured:

cboos@linux:~/trac/0.12-stable$ tx status
No authentication data found.
trac -> 0_12-stable-messages-js-pot (1 of 2)
Translation Files:
 - en: trac/locale/messages-js.pot (source)
 - ca: trac/locale/ca/LC_MESSAGES/messages-js.po
 - de: trac/locale/de/LC_MESSAGES/messages-js.po
 - en_GB: trac/locale/en_GB/LC_MESSAGES/messages-js.po
 - en_US: trac/locale/en_US/LC_MESSAGES/messages-js.po
 - eo: trac/locale/eo/LC_MESSAGES/messages-js.po
...

If you get the No authentication data found. line, you first need to authenticate yourself:

cboos@linux:~/trac/0.12-stable$ mkdir tmp
cboos@linux:~/trac/0.12-stable$ cd tmp/
cboos@linux:~/trac/0.12-stable/tmp$ tx init
Creating .tx folder...
Transifex instance [https://www.transifex.com]: 
Creating skeleton...
Creating config file...
No entry found for host https://www.transifex.com. Creating...
Please enter your transifex username: cboos
Password: 
Updating /home/cboos/.transifexrc file...
Done.
cboos@linux:~/trac/0.12-stable/tmp$ cd ..
cboos@linux:~/trac/0.12-stable$ rm -fr tmp

Alternatively, you can directly edit your ~/.transifexrc file, and add:

[https://www.transifex.com]
hostname = https://www.transifex.com
password = <your actual password>
token = 
username = <your account name>

token can be left empty, but must be present.

Then try again:

cboos@linux:~/trac/0.12-stable$ tx status | grep fr
 - fr: trac/locale/fr/LC_MESSAGES/messages-js.po
 - fr: trac/locale/fr/LC_MESSAGES/messages.po

A very important preparation task that must be done before pulling is to make sure the .po file was updated. Even if you think it must be because there are no local changes, it's a good idea to make update-.. (fr in this example). Perhaps the Babel version you're using at the moment is different from the one which was used before the last version was committed, or a manual change was done at the last minute, etc.

Doing a catalog update ensures that we have a "normalized" content to compare with the version of the catalog pulled from Transifex, after the latter has been "normalized" as well by another Babel update, as the automatic formatting done by Transifex is completely different.

Make sure to "freeze" that updated Trac version, by making a commit in a local branch (if you use git or hg) or even a "maintenance" commit in Subversion.

Pulling the French catalogs from Transifex

Now, before pulling changes from Transifex, make sure you have no local modifications to the catalogs you're about to update. Otherwise, those local modifications will be lost (well, on Windows at least; seems that on Linux it manages to make better checks with the timestamps between the local file and the file on the server).

Once you made sure you have no pending local modifications, you may now pull:

cboos@linux:~/trac/0.12-stable$ tx pull -l fr
Pulling new translations for resource trac.0_12-stable-messages-js-pot (source: trac/locale/messages-js.pot)
 -> fr: trac/locale/fr/LC_MESSAGES/messages-js.po
Pulling new translations for resource trac.0_12-stable-messages-pot (source: trac/locale/messages.pot)
 -> fr: trac/locale/fr/LC_MESSAGES/messages.po
Done.

If you see 'Skipping…' messages, the Transifex client thought it should better avoid updating your local copy. If you're sure you know better, you can pass the -f flag so that it updates it nevertheless.

Examining the changes

If at this point you want to check the difference … good luck! Transifex introduces a bunch of unrelated formatting changes in the translations, folding multiple lines into one, etc.

You can get rid of these spurious changes by doing an update with Babel, which can be seen as a kind of "normalization" step:

cboos@linux:~/trac/0.12-stable$ make update-fr
python setup.py update_catalog -l fr update_catalog_js -l fr
running update_catalog
updating catalog 'trac/locale/fr/LC_MESSAGES/messages.po' based on 'trac/locale/messages.pot'
running update_catalog_js
updating catalog 'trac/locale/fr/LC_MESSAGES/messages-js.po' based on 'trac/locale/messages-js.pot'

Now the changes are hopefully much more significant:

trac/locale/fr/LC_MESSAGES/messages.po

@@ -1 +1 @@
 # French translations for Trac.
-# Copyright (C) 2007-2008 Edgewall Software
+# Copyright (C) 2012 Edgewall Software
 # This file is distributed under the same license as the Trac project.
 #
+# Translators:
+# Christian Boos <cboos@edgewall.org>, 2008,2012.
+# Christian Boos <cboos@neuf.fr>, 2008.
 # Emmanuel Blot <emmanuel.blot@free.fr>, 2007.
-# Christian Boos <cboos@edgewall.org>, 2008.
 # Michel Briand <michelbriand@free.fr>, 2008.
-# Stéphane Raimbault <stephane.raimbault@gmail.com>, 2009
+# Stéphane Raimbault <stephane.raimbault@gmail.com>, 2009.
 msgid ""
 msgstr ""
-"Project-Id-Version: Trac 0.12\n"
-"Report-Msgid-Bugs-To: trac-dev@googlegroups.com\n"
+"Project-Id-Version:  Trac\n"
+"Report-Msgid-Bugs-To: http://trac.edgewall.org/\n"
 "POT-Creation-Date: 2012-01-29 14:04+0100\n"
-"PO-Revision-Date: 2011-01-25 20:28+0100\n"
+"PO-Revision-Date: 2012-10-27 13:13+0000\n"
 "Last-Translator: Christian Boos <cboos@edgewall.org>\n"
-"Language-Team: fr_FR <trac-dev@googlegroups.com>\n"
+"Language-Team: French <trac-dev@googlegroups.com>\n"
 "Plural-Forms: nplurals=2; plural=(n > 1)\n"
 "MIME-Version: 1.0\n"
 "Content-Type: text/plain; charset=utf-8\n"
@@ -2654 +2655 @@
 #: trac/ticket/default_workflow.py:251 trac/ticket/default_workflow.py:271
 #, python-format
 msgid "The owner will be changed from %(current_owner)s"
-msgstr "Le propriétaire changera de %(current_owner)s"
+msgstr "Le propriétaire ne sera plus %(current_owner)s"
 
 #: trac/ticket/default_workflow.py:259
 #, python-format

It's a good idea to get rid of the changes in the first hunk. With git, it's easy:

cboos@linux:~/trac/0.12-stable> git checkout -p
diff --git a/trac/locale/fr/LC_MESSAGES/messages.po b/trac/locale/fr/LC_MESSAGES/messages.po
index 00f3c3c..a33bf50 100644
--- a/trac/locale/fr/LC_MESSAGES/messages.po
+++ b/trac/locale/fr/LC_MESSAGES/messages.po
@@ -1,19 +1,21 @@
 # French translations for Trac.
-# Copyright (C) 2007-2008 Edgewall Software
+# Copyright (C) 2012 Edgewall Software
 # This file is distributed under the same license as the Trac project.
 #
+# Translators:
+# Christian Boos <cboos@edgewall.org>, 2008,2012.
+# Christian Boos <cboos@neuf.fr>, 2008.
 # Emmanuel Blot <emmanuel.blot@free.fr>, 2007.
-# Christian Boos <cboos@edgewall.org>, 2008.
 # Michel Briand <michelbriand@free.fr>, 2008.
-# Stéphane Raimbault <stephane.raimbault@gmail.com>, 2009
+# Stéphane Raimbault <stephane.raimbault@gmail.com>, 2009.
 msgid ""
 msgstr ""
-"Project-Id-Version: Trac 0.12\n"
-"Report-Msgid-Bugs-To: trac-dev@googlegroups.com\n"
+"Project-Id-Version:  Trac\n"
+"Report-Msgid-Bugs-To: http://trac.edgewall.org/\n"
 "POT-Creation-Date: 2012-01-29 14:04+0100\n"
-"PO-Revision-Date: 2011-01-25 20:28+0100\n"
+"PO-Revision-Date: 2012-10-27 13:13+0000\n"
 "Last-Translator: Christian Boos <cboos@edgewall.org>\n"
-"Language-Team: fr_FR <trac-dev@googlegroups.com>\n"
+"Language-Team: French <trac-dev@googlegroups.com>\n"
 "Plural-Forms: nplurals=2; plural=(n > 1)\n"
 "MIME-Version: 1.0\n"
 "Content-Type: text/plain; charset=utf-8\n"
Discard this hunk from worktree [y,n,q,a,d,/,j,J,g,s,e,?]? y
@@ -2654,7 +2655,7 @@ msgstr "à %(owner)s"
 #: trac/ticket/default_workflow.py:251 trac/ticket/default_workflow.py:271
 #, python-format
 msgid "The owner will be changed from %(current_owner)s"
-msgstr "Le propriétaire changera de %(current_owner)s"
+msgstr "Le propriétaire ne sera plus %(current_owner)s"
 
 #: trac/ticket/default_workflow.py:259
 #, python-format
Discard this hunk from worktree [y,n,q,a,d,/,K,j,J,g,e,?]? q

Note that you can also do make diff for a more up-to-the-point output (make diff vc=git if you're working in a git checkout).

Do something with those changes…

… do whatever you wish, for example commit the changes to the source with an appropriate log message, but, please!, before that:

• remember to check your catalog with make check (here, make check-fr)
• remember to check your catalog by compiling it (here, make compile-fr)

Once you're done, make sure you push back the changes to Transifex. See the other example below.

Pushing local modifications of the French catalogs to Transifex

After committing changes to the message catalog, it's a good idea to push those changes to Transifex so that other contributors can see these changes.

Even if you just pulled changes from Transifex (coming from the other example above), you most certainly did local adaptations like the removal of the spurious metadata changes, or small fixes, then you should also push those changes back to Transifex the same way, as Transifex won't pick up the changes made to the translated catalogs from the repository (it will do that for the catalog templates though, so the latter are always up-to-date).

But if you didn't do a pull just before, doing a push might overwrite changes contributed to Transifex, so you should be careful. The safest thing to do is to simply do a pull first, normalize the changes by running make update-.., compare, and only then, if there are no changes but spurious ones in the metadata in the comments in the header, revert those spurious changes and do the push of your committed state. If the changes were significant, you're simply back to the #pull-example situation described before.

cboos@linux:~/trac/0.12-stable$ tx push -t -l fr
Pushing translations for resource trac.0_12-stable-messages-js-pot:
Pushing 'fr' translations (file: trac/locale/fr/LC_MESSAGES/messages-js.po)
Pushing translations for resource trac.0_12-stable-messages-pot:
Pushing 'fr' translations (file: trac/locale/fr/LC_MESSAGES/messages.po)
Done.

For some languages, push can fail like this:

$ tx push -t -l gl
Pushing translations for resource trac.0_12-stable-messages-js-pot:
Warning: No mapping found for language code 'gl'.
KeyError: 'gl'

This is because there's no message-js.po catalog for the 'gl' translation, only a message.po one.

The workaround is to push explicitly this one, by using the -r option:

$ tx push -t -r trac.0_12-stable-messages-pot -l gl
Pushing translations for resource trac.0_12-stable-messages-pot:
Pushing 'gl' translations (file: trac\locale\gl\LC_MESSAGES\messages.po)
Done.

Notes about the Transifex "Releases"

Transifex is not exactly clear if a release should be a line of development (branch) or a given released version. It supports both and let you choose to organize you the way wish.

I opted to make the "releases" match our maintenance branches, and not individual releases (well, we still have a ​few of these, as apparently you can't delete a release).

The downside is that string freeze periods and release dates will need to be adjusted constantly on the same release. The alternative, creating a new "release" for each actual version we release bears as much overhead (more even, are there's no copy operation) and doesn't bring any benefit besides keeping a record of the versions we made, record that we obviously maintain here anyway in the roadmap.

OTOH, the big benefit is that all the links from t.e.o to Tx won't need to be updated after each minor release. The only updates needed will be at the occasion of introducing a new major release (i.e. for 1.2).