[Sugar-devel] Sugar module build is broken by translations

[Chris Leonard]

On Wed, Jun 13, 2012 at 11:41 AM, Simon Schampijer <simon at schampijer.de> wrote:
> On 06/13/2012 05:15 PM, Daniel Narvaez wrote:
>> the build on sugar git master has been broken for a few days for me on
>> Fedora 17.

. . .

>
> Thanks Daniel for the heads up. I have already asked Chris to take a look.
>
> Sadly, we break rather often lately after pushes from pootle. There is no
> verification done inside pootle itself :/


Ahem . . before we leap to conclusions about Pootle not doing it's job
correctly, let's take a closer look at what is going on with the PO
and POT files.

If you examine aym and id (which appear to be the source of the build issues).

http://translate.sugarlabs.org/projects/glucose/

You will notice that they have different word counts than most of the
other languages.

For example: en_GB (not touched since 0.96 release)

sugar-base.po
1 file 5/5 words (100%) translated [5/5 strings]

sugar-toolkit-gtk3.po
1 file 71/71 words (100%) translated [35/35 strings]

sugar-toolkit.po
1 file 73/73 words (100%) translated [37/37 strings]

sugar.po
1 file 1594/1594 words (100%) translated [389/389 strings]


en_US (freshly updated from templates)
These are the same stats as lang-aym and lang-id

Note change in word and string count in sugar-toolkit-gtk3.po and sugar.po

sugar-base.po
1 file 5/5 words (100%) translated [5/5 strings]

sugar-toolkit-gtk3.po
1 file 76/76 words (100%) translated [40/40 strings]

sugar-toolkit.po
1 file 73/73 words (100%) translated [37/37 strings]

sugar.po
1 file 1582/1582 words (100%) translated [387/387 strings]

The word and string count changes can happen by an upload with
overwrite (possible source of localizer error) or they can happen by
an "Update from Templates", which would trace back to a change made by
a developer to the code on Gitorious, cascading down into the POT
automagically (via the potupdater cron job) and then to the PO file
(after an "Update from Templates") and then back to Gitorious via
commit from Pootle (as appears to be the case here).

I would speculate that in this case, it is the fact that the POT has
not been frozen after the 0.96 release that may be the source of the
issues encountered.  I am not 100% confident in that speculation, but
I am working with Simon to try to determine how to get the builds back
on track.

I freely admit to perhaps being more aggressive than usual in trying
to make sure that localizers work is included in builds (the "Push
many PO files" commit, for example), and if I am the source of the
problems, I sincerely apologize for the inconvenience and promise to
do whatever is needed to make things right, in addition to being more
cautious in future.   However, at the moment, my intuition is causing
me to suspect the late (possibly post-release?) POT churn as the root
of this problem.  The solution for which would be to branch in
Gitorious (and Pootle) ASAP and to document our coordination between
release and L10n better for future use.

cjl
Sugar Labs Translation Team Coordinator