Plan for L10n for Fedora 11

[Asgeir Frimannsson]

----- "Dimitris Glezos" <dimitris at glezos.com> wrote:
> I'm very happy to announce that we've now landed support for
> translation
> statistics in Transifex.
> 
> Lately we've been working heavily in re-writing Transifex and getting
> v0.5 in
> the best shape possible for Fedora 11. Being 3 weeks ahead of the
> string
> freeze, we have plenty of time to test statistics on Transifex
> 0.5-devel.
> 
> For Fedora 11, my plan is to switch the old DL for Tx 0.5-devel for
> statistics, and continue to use the proven Tx 0.3 for submissions.
> I've
> requested [1] a publictest instance from the Infrastructure Group to
> install
> it and start putting data and testing it.
> 
>   [1]: https://fedorahosted.org/fedora-infrastructure/ticket/1191
> 
> To allow 2 weeks of testing, the instance should be ready by 23/2.
> When we
> see that everything works out as it should, we can discontinue the old
> DL.
> 
> Our goal with Transifex 0.5 is to include support for both submissions
> and
> statistics, and this way we can put aside our old version of Damned
> Lies which
> is presenting outdated translations. Since only the commits are
> missing from
> Tx 0.5-final, it'll be out in 3-4 weeks, and at that point we go on
> and test
> submissions too, while having Tx 0.3 as a backup solution.
> 
> Some of the advantages of this approach is that using the statistics
> from Tx
> 0.5-devel does not even require hook-up with FAS (only submissions
> require
> authentication), we have a smoother upgrade path for Django/v0.5 and
> we have a
> codebase we know inside-out to build/invest on.

First of all, setting up a test-instance of Transifex 0.5alpha is a good start, so a big +1 from me. You should probably think of a migration script for transferring the data as well. I would suggest [1] as a good starting point for this, as it is based on Django and gets you quickly up and running.

That said, I have a number of concerns with this approach. Just a few weeks back you noted that 0.5 wouldn't be ready in time for F11, and that a good approach would be to migrate to the new Django-based Damned Lies for F11. Hence, I have spent roughly a week worth of development on this now, and I am announcing a public test period for this by the end of this week. The changes we have made to the upstream Gnome version include Fedora-theming (similar to the existing instance), FAS authentication for translators in the cvsl10n group, proper VCS locking and improvements to the way vcs sandboxes are managed, and now the last bit involves support for publican statistics (for e.g. release notes and selinux-guide). 

However, I have no issue with throwing that work away if we can demonstrate that Tx 0.5 will serve the needs of the L10N project better in the given time-frame. My main concern is the translators and their work flow, and for F11 we owe translators a less crap system than what we've been using for the last two releases. My secondary concern is maintainability, and the core of my changes to DL has been ensuring that it doesn't fail as often, and providing better information when it fails.

Some of the benefits with the new Django-based DL:
- Translation teams can be better organized (Does Tx even have team pages?)
- Translators can indicate that they are working on a specific file
- Options of using a Translator->Reviewer->Committer work flow, I suspect linking up the actual commit functionality (migrating it from Tx 03 to DL) is less than a day's work.
- Django based DL has been in production upstream (Gnome L10N) for more than 3 months

So the gist of my message is that we should first look at the how the translation work flow changes by using these two tools. As I see it, we have four options:

1) Tx 0.5 for stats, Tx 0.5 for commits (not implemented yet)
2) Tx 0.5 for stats, Tx 0.3 for commits
3) DjL for stats, Tx 0.3 for commits
4) DjL for stants, DjL for commits (not implemented yet)

So far I'm a few days away from completing (3), and should also be able to get (4) done by early next week. If I can be convinced of a clear roadmap for (1) I'd be happy to help out with that, provided I can see the benefit to the translators with this approach. However, I am a bit afraid that option (2) might hurt Tx more than it gains in the long run. Why not focus on getting Tx 0.5 production ready and in shape to become a great translation platform, rather than simply rushing it in in time for F11? 

Maybe I'm missing something obvious, but *it is* great to finally see a test instance of Django-based Tx, so don't get me wrong, and I'm guessing a test instance of both systems will make it easier for us to find the best approach for F11 L10N. I know there have been plans of merging the Django-based Tx with DL as well, perhaps this might be a way of developing a migration-path in that direction as well, as I'm worried that the longer these applications stay separate with active development on both sides, the harder it's going to be to get e.g. Tx accepted for use in Gnome.

cheers,
asgeir

[1] https://fedorahosted.org/damned-lies-fedora/browser/stats/management/commands/migrate.py