On Fedora's Localization platform

[Petr Viktorin]

On 07/18/2014 12:55 PM, Dimitris Glezos wrote:
> On Thu, Jul 17, 2014 at 3:43 PM, Petr Viktorin <pviktori at redhat.com> wrote:
>> Tools are *not* just tools. Relying on superior tools that are graciously
>> provided, but can be taken away at any time, can be quite dangerous.
>
> The analysis on the level of danger and cost/benefit is missing. This
> is Board's responsibility to define. The Board could demand an
> agreement with "we can use it for free and export all important data
> (+historical) for the next 10 years" -- and that could be good enough
> for the Board. There are other areas where as a project we're taking
> the risk, like with some of the hardware of our infrastructure which
> runs closed firmware we don't control.

Right, but the hardware is a commodity. It doesn't lock us in.
If Transifex was a commodity frontend for translations, I'd have no 
problem with it.

> I'd expect the L10n project to be primarily concerned about what
> matters most to it: making Fedora L10n a hugely successful project. We
> should be the ones willing take risks to achieve the core project's
> goals, push hard to have what we need, and the Board should be pushing
> back.. not the other way around. This, of course, requires strong
> leadership and governance.

I don't know what "We" is, bu I think it would be a sad world if only 
the Board pushed for Fedora's first foundation...

>> It has been mentioned on this list that sharing the translations between
>> multiple sites, so the "pragmatists" can get their features and the "purists"
>> can keep their freedom, is not feasible. This makes me sad. Why can't
>> Transifex or Zanata be just frontends for the data, sort of like Github is
>> for Git?
>
> Short answer? We tried it. This is how the first version of Transifex
> worked: it read git and committed back. I could write a huge list with
> the benefits. In short, users hated it. User experience sucked,
> translators were unhappy and developers simply did not translate their
> apps. In the end, this was important for 1% of the users. Put simply,
> what we originally designed is not what the users wanted. Plus, it did
> not achieve the core goals of the localization platform: to make
> localization as widely used as possible (= what future users want).

Yes, I agree the solution with direct access to Git sucked. The current 
Transifex is certainly better, especially for the average/casual user.

> GitHub's story is similar. GitHub is so much more than a git
> front-end. The true power of GH is the collaboration and
> communication. First it was ACLs and team management and then came
> pull requests, reviews/comments on code, integration with Issues and
> Wiki, integration with chat/issue tracking etc. This is too much and
> too complex metadata (+ functionality on top of them) to store in an
> external DB/repository.

Nevertheless, Github is the best option around if you *just* want a Git 
repo hosted, without any of the frills.
It works great for mirroring a project hosted elsewhere.
Also, it's easy to contribute to a Github-hosted project if you don't 
want to touch Github -- mail a patch, maintainer merges&pushes, done.
Transifex? IMO not so much; the the platform wants to be the canonical 
location of all the data. And when I merge a PO patch upstream, updating 
Tx is inconvenient (Will translations be lost if I just upload the PO? 
Is it possible to roll back if I screw up?).

> When Transifex started with VCS integration we had less than 500
> projects translated. Today there are 20.000. And the more important
> numbers are even better (number of completed languages per project,
> number of contributors, activity/month, proofread phrases Vs
> only-translated etc).

In short, Transifex grew. Congratulations!
Surely that doesn't that mean every individual design decision you made 
along the way was perfect.

> Successful L10n Community Managers aim high on
> these metrics and track them like crazy. Where the files are actually
> stored is gravy.

As Yuri Chornoivan says in a separate mail.... not necessarily true.
(He presented it as his personal opinion, but consider: for a project, 
Yuri is easily worth dozens of average translators. In FreeIPA, 
Ukrainian is always up 100% a few hours after I upload the POT. I don't 
know how the man does it, really.)

What the average future user wants is not what power users want.

-- 
Petr³