+1 on both.

I think freeze is difficult to coordinate and a bit disruptive. and I think
it is relatively easy to add missing phrases quickly.
And yes. We should definitely have reminders before release - and having a
template to follow and [ANNOUNCE] is a good idea to trigger adding missing
phrases.

Also maybe one proposal: we seem to have **some** not huge, but quite
regular changes in patchlevel releases - and I don't think it's something
temporary, it will likely happen all the time in the future - maybe we
should have a light version of such reminder also before patchlevel
released - to add missing phrases in v3-X-test branch. Those changes are
not often cherry-pickable, but applying the few (usually) small phrases
should not be a big issue.

J.


J.


On Sat, Dec 6, 2025 at 5:48 PM Shahar Epstein <[email protected]> wrote:

> Hey everyone,
>
> While making updates to the translations I’m responsible for, I realized
> that although the translation freeze mechanism was very effective just
> before the initial release of the i18n feature (with all 18+ languages
> included), it does not seem necessary for every minor or major release and
> could become a potential bottleneck for both the release managers as well
> as contributors.
> In addition, with AI-based tooling, completing missing terms has become an
> easier task for translation owners to bring their locales above the
> required threshold in no-time - so for completing only dozens of terms for
> the most time, there's no good justification for the freeze to be applied.
>
> For that reason, I created PR #59136
> <https://github.com/apache/airflow/pull/59136/files>, which introduces the
> following changes in the policy:
> 1. Instead of requiring a freeze for *every *major or minor release, a
> freeze will be applied only when the median* coverage across all languages
> is below the 90% threshold, or when deemed necessary by the release manager
> (e.g., when a critical UI feature introduces many new terms). The idea is
> to use the freeze when *many *changes need to be applied across *many
> *translations
> (well above 100 terms), and not when specific translation(s) are simply
> unmaintained for too long.
> 2. A simple completeness check *should *be performed in every minor and
> major release, after which a thread should be posted on the dev list asking
> code owners to ensure completion (90%+) by the RC release (a mail template
> is included in the PR). Non-completed translations after the due date
> should be tracked for the subsequent release.
>
> * - median and not average to reduce sensitivity for outliers.
>
> I'll be happy to hear your opinions regarding it before making it to a lazy
> consensus in the upcoming days :)
>
>
> Shahar
>

Reply via email to