This exists to help determine our best route to manage translations. Since our move to SVN soon will already require us to alter every translated file (we can no longer depend on CVS Revision numbers), we may as well reconsider our current practice and go from there.
Currently each translated file has a tag as such:
<!-- EN-Revision: 1.3 Maintainer: takagi Status: ready -->
That means this file is up-to-date and in sync with version 1.3 of the English file. The number 1.3 is the CVS $Revision$ of said English file.
We have tools to track when these our out of date, and this simplistic approach is how we manage translations today.
The move from CVS to SVN will change every revision number, so EN-Revision may be 456 instead of 1.3 now. And these numbers are not really sequential, so the next version of this EN file may be 987. This means all of our tools no longer work, and it's difficult to manage.
Several routes exist.
In the above, moving from comments to markup means implementing a revision attribute into the DocBook itself. So, something like this:
<!-- EN-Revision: 456 Maintainer: takagi Status: ready --> <reference xml:id="reserved.variables" xmlns="http://docbook.org/ns/docbook">
<reference xml:id="reserved.variables" xmlns="http://docbook.org/ns/docbook" revision="456">
Whereas totally redoing the system would essentially change everything, namely, every translated file and how we manage them.
Many systems today use a gettext approach, so in this case the English version of the manual would remain the same (DocBook XML) but the translations would be managed using po files.
Projects that use this gettext approach:
The SVN conversion will be completed sometime in the first quarter of 2009, let's say March 1. We should have our system essentially in place by that time. A sequential move is possible too (Start using SVN numbers, then later gettext) but that wouldn't be ideal. However, given the size of our manual, it may be required.
So far I've (philip) am having success converting translations to po files using po4a. After additional tweaking the plan is to setup pootle and see what happens. We may use transifex instead so we'll see but the key is getting the po files going. The main alternative to po4a is xml2po but so far I like po4a so don't see a need to investigate xml2po much further.
So, the current plan: