Status of Debian Installer i18n framework improvements/changes

[Christian Perrier <bubulle@debian.org>]

When coming back from holidays, on Aug. 15th, I sent a long proposal
about improvements to the D-I i18n/l10n work methods.

This proposal could be resumed to two main points:

1) move towards the use of a single file for all "core" D-I packages
   translations

2) better categorize what we currently call "second stage" and "third
   stage"

Point 1) is aimed at offering a strong simplification to translators
work, especially those who are the only translator for their language.

Point 2) is aimed at having a better overview of the translation
status of the whole Debian installation process, from the user point
of view, and help translators prioritize their work

Two weeks after, it is time for me to make a progress status about
these projects.

1) Transition to a "master file"
--------------------------------

Technically speaking, the scripts/l10n/l10-sync script is the core
component of this process.

I have worked as deeply as possible for improving it in order to make
it robust to potential conflicts (avoidn commiting files with conflict
markers) while minimizing the number of commit it makes.

This has been nearly achieved and the generated noise is now quite
low, except in one case : when an upstream package has changed
templates, the script will make one commit per handled language. I'll
work on improving this, but at low priority.

The script now runs on my own system, every 3 hours, at 0:00, 3:00 and
so on (UTC).

The next goal is having it run on a Debian system, probably
gluck. This needs some investigation, though, as the script uses
po-debconf utilities, which are not available on Woody.

At the moment I write this, 22 languages are handled through this
"new" system. This means that 21 still use the old system. Most of
these are "one translator teams".

The current number of strings in the PO files is 1246 while the
combined size of all D-I packages is over 1400 strings, IIRC (the
difference are identical strings in different packages).

A few members of translation teams have expressed concerns about this
new scheme, including some in the French team. The main argument
against this scheme is more problems for dispatching the work and
update work. Especially when it comes at update management, I'm not
completely convinced that multiple files are easier to handle....and
I'm convinced they're more subject to inconsistencies.

Anyway, I respect the translation teams work methods and I'm currently
thinking about a system allowing both work methods.

Denis Barbier objected that the problem of PO files spread all around
the D-I repository is solved by the special repository set by Joey
Hess by using SVN hooks. This repository is IMHO really too slow for
being easy to use as a work basis.

Another argument I want to emphasize for the new method is the
documentation : currently, documenting the translation work is quite
complicated and needs explaining how to get the material for each
package and how the work should be prioritized. With the new method,
this is just "pick up template.pot, rename it to xx.po and commit back
the new file".

Anyway, the work will continue during the next weeks. I have asked to
remaining "single" translators about their agreement for switching
their language. I will soon talk more closer about this with the
translation teams (German, Dutch, Chinese, Portuguese (both), Italian,
Spanish...).

The D-I RC2 process and the associated string freeze will certainly
give us a better idea about the robustness of the whole system and how
it helps in translators work.

2) New way of counting the progress
-----------------------------------

I have proposed a new scheme, replacing the current "stages". The
rationale are the curernt inconsistencies in the way we have sorted
translations into either 2nd or 3rd stage. Another rationale is being
able to count the number of complete translations for a full Debian
base system (which is what the user really sees).

The proposal divided translations in "levels", with 4 levels:

1 : core D-I files
2 : all packages which prompt users during a *default* install at
    high priority for a Debian base system
3 : all packages which prompt users during other type of installs
    of a Debian base system
4 : all packages which *output* messages during any type of install
    for a Debian base system

No progress has been made yet on that subject. this needs some
redesign of the translation status pages and some rewrite of Dennis
Stampfer scripts which maintain this page. The new status page needs
to be operational while the old one is still used.

Dennis began to work on the design of this script and I hope that some
first versions will soon show up. Unfortunately, repetitive SVN/haydn
repository flaws didn't help in this work.

This will be the next priority during the next weeks.

3) Documentation
----------------

I have finalised the new i18n documentation version. Please review it
in installer/doc/i18n/i18n.xml. Built HTML and TXT files are available
in the same directory.

Please be aware that this file documents some currently *not* existing
methods, mostly the "levels" system.

--