26 Mar
2018
26 Mar
'18
12:05 p.m.
Hi,
I am a bit skeptical about such a platform that:
* we'd need our users to register to (yet another account for the XWiki
contributor)
* we'd need to twist and turn so that it bends to our needs
* (and probably other things caused by the fact that I did not get to read
too much about it yet :) )
You mentioning PRs made me wonder if we couldn't simply ask contributors to
submit a standard PR for translation changes. That would take care of many
things and also give the author proper attributions.
The minimum requirements for this would be, IMO:
* the list of all translation files (i.e. links on GitHub; either manually
managed, automatically extracted or a mix) and
* the ability to search for a key or translation and find the relevant file
(most likely this would involve us indexing or caching the translation
files content somewhere on xwiki.org, kept up to date with webhooks? --
maybe on a very light translations dedicated wiki; a light l10n?).
* the ability to create a new language for a translation page, if it does
not already exist (to help the edge case when the language file is not
already there).
Something like this could even be done without an xwiki.org account, as
long as the user has a github account for editing (even with GitHub's web
edit UI) and submitting the PR.
Most importantly, IMO, the translations could become part of the
development process and not just a step at release time. A clear
consequence would be that snapshots would benefit from new and integrated
translations, added since the last release and we won't be seeing them just
after the release. AFAIU, weblate supports this as well, but by using its
own github account and submitting PRs in the name of the author (which
would not be the same effect as the actual author, but then again, only
actual GitHub users would care about this).
WDYT? Would it be too hard to use for non-techical users? Would we need
more that the described, in terms of UI (and would it justify the
effort/cost)?
Thanks,
Eduard
On Fri, Mar 23, 2018 at 11:56 AM, Vincent Massol <vincent(a)massol.net> wrote:
Hi devs,
We started discussing this yesterday on the xwiki chat.
Here’s the rationale for changing:
* l10n is slow and is preventing users to contribute translations (it’s
much harder than it should be)
* it’s costing us maintenance that we shouldn’t have to do (it’s not our
role to maintain a translation service, even though it was nice to eat our
own dog food initially). I see this very similar to when we switched from
our GWT-based WYSIWYG editor to CKEditor.
* It’ll allow us to benefit from new features/bug fixes the external
service develops
* Right now it’s taking an hour or more every time we release to integrate
the translation changes and this slows us down to increase our release
delivery speed
* We’re putting the onus on the RM only to validate that the proposed
translations are good. So only 1 pair of eyes.
* We have no time to fix any translation if they’re not correct. A system
where when someone proposes a new translation generates a PR that we apply
as they come in would be much better and solve both the review and lateness
issues.
Of course it’ll mean some plugins/customizations to develop in the service
we will use since it’s not going to support some custom formats we have
such as our XAR XML format. So we need to pick a tool that allows for this.
The proposal:
* Start investing in implementing XWiki translation using an external tool.
* Start by looking at weblate: https://weblate.org/en/ since
** it seems to offer lots of features we need (automatic PR - see
https://docs.weblate.org/en/latest/vcs.html#github, quality checks,
plugins). See https://weblate.org/en/features/
** it’s open source itself and in case the service goes down we can ask
XWiki SAS to host it for us (see https://github.com/WeblateOrg/weblate).
** We need to ask them if they can host us, see https://weblate.org/en/
hosting/ and the part about "Hosting for free software”.
** We could also ask XWiki SAS if they’d accept paying a “Basic” plan (200
euros/year which is affordable IMO vs hosting it ourselves)
** There’s a REST API so that if we want we can provide a view/status of
the translations from xwiki.org: https://docs.weblate.org/en/
latest/api.html . We could even imagine using this API to convert from a
format to our own format, if need be.
* Propose to have a developer work on this in an upcoming roadmap (to be
defined but as early as possible since l10n is not in good shape)
* Fix l10n as much as we can without spending too much time on it, until
we have the new translation service ready to be used
Things to look at:
* Ability to register custom formats or more generally how to handle our
custom format
* How do we handle deprecated translations keys
* How do we handle global rename/move of keys from one resource file to
another
* <add more here>
WDYT?
I’d like to have especially Thomas’s POV since he’s the one who spent the
most time on l10n and I’d like to make sure he’s ok with this.
Thanks
-Vincent