555178 - [compare-locales] filter.py and error logic not good enough for lorentz, and per-l10n.ini filter.py

Status: RESOLVED FIXED

(Mozilla Localizations :: Infrastructure, defect)

Description

[Axel Hecht [:Pike]]

The current logic in compare-locales for errors is either "error" or "ignore", but we need a "report" state as well for Lorentz.

The strings should be exposed to localizers, but the strings with backup strings in the code should bust.

With l10n-merge on, "report" strings will be treated like "ignore", but they'll show up in the output in some way.

The other thing we need is to check for filter.py next to each l10n.ini, so that we can actually "report" or "ignore" toolkit strings independent of app. That's needed so that we don't have to hack around 1.9.2 strings in more-or-less mozilla-central focused repos like mobile-browser or comm-central.

Comment 1

[Axel Hecht [:Pike]]

Attachment: patch queue as of now

This patch queue is the easy part, currently three patches deep (plus the one that's actual beef and in the works).

- .hgignore patch, just to ignore .pyc and hg files
- add handling of filter.py functions that return "error", "report", and "ignore" instead of just bools (True is "report"). Make them overrule in that order.
- add picking up filter.py from all ConfigParsers, not just the top one.

The next patch is the one that actually handles "report" and "error" differently. That patch might actually change how I do notify quite a bit, I'm not really all that happy with how that works just now. Aka, I'm confused by the code myself. Changing that might enable better reports down the road, too, like better source information, in particular in the html output.

Comment 2

[Axel Hecht [:Pike]]

Attachment: updated queue

Comment 3

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

adding releng, so we can start to track this.

Comment 4

[Axel Hecht [:Pike]]

Attachment: update queue, ready for review

Another update, using execfile again instead of exec, so that debuggers can step through filter.py code with source information.

I used urlparse().path, which introduces a dependency on python 2.5, if we didn't have that yet. But that's fine according to coop, and for life, according to me.

Comment 5

[Chris Cooper [:coop] (he/him)]

Comment on attachment 437855 [details] [diff] [review]
update queue, ready for review

Patch looks good. I gave it a spin and it didn't break l10n repacks on my local setup either.

Axel: I think you can safely bump the RELEASE_AUTOMATION when you're ready to land too. Feel free to include the patch from bug 558057 under that umbrella too if you want.

Alternately, I can push these changes out for you tomorrow morning PDT during a releng pseudo-downtime.

Comment 6

[Axel Hecht [:Pike]]

Marking FIXED.

Comment 7

[Axel Hecht [:Pike]]

This apparently doesn't work on windows, see bug 558219 comment 7.

Comment 8

[Axel Hecht [:Pike]]

Thanks to bhearsum to help investigating, I missed out on a call to url2pathname, http://hg.mozilla.org/build/compare-locales/rev/331d5bca7838.

Another round of release test is under way, but resolving this as FIXED.

I moved the RELEASE_AUTOMATION tag as well. Yes, getting access to windows again is planned.