On Wed, Aug 19, 2015 at 9:07 PM, Zibi Braniecki <
[email protected]> wrote:

> I got Settings to load properly, they're the opposite of FM radio. FM
> radio is localizing a single entity during launch. Settings are localizing
> above 100 before firstPaint.
>
> The magic? The memory overhead of l20n.js over l10n.js is *exactly* the
> same. ~150kb.
>

That's good to know.  And I feel like we have a pretty good idea of where
about half of that extra memory is used, right?  The module system?


> That's a really good news because it means that it doesn't scale with
> data, it shouldn't affect default vs. non-default and the whole
> optimization work should go into the init codework in l20n.js that I
> already profiled pretty well.
>
> I'm going to start doing that tomorrow. My plan is to take each major
> chunk of code that census shows as a major memory retenier (getMeta,
> plurals, pseudo, io.load) and try to optimize it and see the impact on
> raptor on FM and Settings.
> If I get a win on both, I submit a patch.
>

I'm looking forward to more data.  I hope lazy qps will save us some
memory.  Also, let's start talking about moving plurals into localization
files.  Would that even be feasible for CLDR rules?


>
> Also, performance wise Settings with l10n.js and l20n.js are exactly on
> par. That's another great news. I still think there's room to improve
> l20n.js, but it seems that the baseline is there.
>

Are on par for non-default languages as well?

-stas
_______________________________________________
tools-l10n mailing list
[email protected]
https://lists.mozilla.org/listinfo/tools-l10n

Reply via email to