Rosetta/bzr

[Danilo Šegan]

Hi Milo,

Today at 14:52, Milo Casagrande wrote:

> Yes, but when working offline with a po, you mark fuzzy an entry you are
> not sure or that you want to check later (at least the Italian GNOME
> team, and I, do that).

That's what we did in Serbian team when we started as well, and only
got a mess out of it.  It simply doesn't work, unless you use msgmerge
without fuzzy matches (and you don't if you use PO files updated by
upstream/imported into Rosetta).

> "--use-fuzzy" is useful for testing new translation against the UI and
> check if it fits in well or, at least for Italian, if the translation is
> the right translations: we have different style for different things in
> the UI that could have the same base string, like "Select file" from a
> menu and "Select file" for the tooltip (that's a stupid one... but there
> are many). Working offline, if you have the src, is easy, you can go and
> see it directly; with Rosetta a test is needed.

I agree that there's a shortcoming in both GNU gettext PO format, and
by extension, in Rosetta—originally modelled on gettext support.

Still, using fuzzy marking for this is wrong, and won't work well.  If
it works well for you, great, but it's merely by accident (well, the
"accident" being that we probably don't import much half-translated
Italian translations, because of upstream teams being good at keeping
translations at 100%), and it's going to be broken for most other
languages.

>> Of course, there's a discrepancy between "fuzzy" and "needs reivew"
>> concepts in Rosetta as well (we definitely want them separated, and
>> that's another thing we're working on).
>
> So when I mark a string as "need review" and download the po file,
> doesn't it show up like "fuzzy"?

It does, and what the above paragraph was meant to say is: this is a
problem in Rosetta terminology.  "Needs review" is what translators
usually want, and "fuzzy" is a string which needs review because it
was machine-translated based on original string similarity match.

It is clear that a string needing review because of human translating
it and not being sure about it is very much different from machine
translating it (which *must* go through review).  It's unfortunate
that the same mechanism in Rosetta and PO files is used for both of
these.

And I hope we will fix this problem as well when we implement
  https://features.launchpad.net/products/rosetta/+spec/rosetta-fuzzy-merge
(the spec at least acknowledges the problem)

Cheers,
Danilo