On 27.02.13 09:30, Henri Sivonen wrote:
On Fri, Feb 22, 2013 at 8:03 PM, Axel Hecht <l...@mozilla.com> wrote:
On 22.02.13 18:41, Henri Sivonen wrote:
On Feb 22, 2013 5:30 PM, "Axel Hecht" <l...@mozilla.com> wrote:
There's just no other way than post-mortem work. That's one of the
reasons why we're not taking arbitrary changesets to ship to any audience
beyond aurora and nightly, for beta and release, we got to have technical
checks in place.
Where should I file bugs to add checks to this set of checks?
Not sure which checks you're talking about, so I can't really tell what you
want.
I meant checks like flagging attempts to go to beta with either of the
following:
* Detector pref not being blank except for a specific white list of
particular values for the ru, uk, ja, ja-JP-Mac and zh-TW locales.
* Fallback charset set to UTF-8 in any locale that doesn't already
have it set to UTF-8.
I'm doing a source-based review, which at least catches regressions to
those settings.
And I think we're doing charset detector settings wrong. Let me see if I
get right what we're doing:
- most content should be labeled for charset
- if not, let's see if we can guess the encoding
-- if we assume the language of the content, we can guess better
-- many languages really only have one option
-- ru, uk, ja, zh-TW do have options, so we use a charset detector
Now, I don't think it's right to use the UI language to guess content
language. We have a list of user-preferred languages (with good defaults
based on UI language). We should go through that list, and pick charsets
to try for unlabeled content from there.
That's rather orthogonal to what you're currently trying to do, but it's
also indicating to me that we should remove all of those settings from
intl.properties, and just leave accept-lang, and deduce the rest.
You also mentioned in the bug that you didn't get the OK to use
telemetry to gather further data. I think if we just collect the data
about the charset optimization and how good it's doing, we should be OK.
I.e., at the point where the locale doesn't matter, but just cp-1252
etc, the entropy goes up a good deal. In particular for small locales.
I'd argue that this might even make sense to be part of health report.
Axel
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
