I have written about the incubator before, it's a tool that should help localizations newly in CVS to work towards release. It's based on buildbot, and I whacked it pretty hard recently.
The result can be seen on the l10n server again. What's new?
- Incubator is now L10n Trunk, I'm running my tools on all localizations working on the trunk now. That's both localizations moving towards fx3 and new ones.
- It's using the python implementation of CompareLocales. It does check more, and it's output could potentially suck less. Red again means items missing, orange is obsolete items, green is good.
- I'm exposing stats on localizations. You'll find a 90% translated in the output box of the comparison step. More below.
- There's one column for trunk, and one for incubators. You can tell which build is yours by looking at the description of the comparison step.
- Technically, it doesn't sink the l10n server anymore on incubator check-ins.
- Only incubator builds do langpacks right now, and only if the comparison ain't red. There's a download link labeled [D] on the upload step that takes you there.
So what's the stats? First of all, I know that "translated" isn't the right term, but it's as wrong as anything else, so wth. What I'm doing is, I'm counting all entries in DTDs and properties files which key does not match [kK]ey. Then I'm comparing those values to the en-US values and then do a 100*changed/(changed + unchanged). There are still a bunch of values that don't need to be changed to be localized, so I'm not worried that we're not at 100% anywhere. I do think that the differences between the high 70s and the 90 are real, though. I'm confident that removing the keys made those stats comparable mostly, as it did fix Japanese.
Why the redo? As indicated above, the first implementation didn't scale, so I needed a bit more buildbot-ism. I didn't feel good about adding new languages to the incubator just for load reasons. That's solved. The amount of languages leading to numbers of columns turned out to be bad, too.
What the redo? I'll make follow-up post on that, this one is long enough.