RFC: New approach to locales - or - we already use localepurge anyway

[Daniel Ruoso <daniel@ruoso.com>]

Hi,

I was thinking about the size of the localized messages inside the
debian packages and about how this can affect the bandwidth load not
only on the mirrors side, but also on the client side. I had recently
installed sid on a 486 notebook that doesn't have CD or ethernet, so I
had to download all the system through the serial port (which is a
16550A, so, max 38.4kbps)...

well, personal experiences apart, here's my idea, it's quite undefined
yet, and I would like to listen to your comments...

Today, the debian packages includes the translations to all languages
inside it. All the translations are downloaded, but only one (in 99% of
the machines) is used. There is a well-known tool called localepurge
that every sane person uses to clean the hard disk.

So I started to think, why download several megabytes and delete after
that?

So, here is my draft proposal:
 * Remove localized messages from deb files, creating separated files
for each language.
 * These files will not be regular debian packages, but special packages
that only contains the locale files... let me say... ldeb packages...
 * The format for the file names would be
   packagename_1.0-3_pt-br.ldeb or
   packagename_1.0-3_pt.ldeb

I know 3 change points:
 1) dpkg-buildpackage would have to split the files inside the package
to create the special packages.
 2) apt would have to download the ldeb files for the locales the user
choosed.
 3) dpkg would have to know about this type of packages and stablish a
database of which locale files are installed for each package in a way
that these files could be deleted if the user de-select a locale or
required if the user select a new locale.

Transition:
I don't have a clue about how to migrate the archives or how to avoid
transition problems for users upgrading from current system to the new.

Important Point:
This change would not generate new entries in Packages file, maybe one
entry in control to say which locales are available, or a file listing
the ldeb files (a ls-lR file with md5sums), or even a file per language
listing the available ldeb files, or even none of the above, leaving to
apt to try to download and fail if it doesn't exists.

Possible features:
I don't know how yet, but this could lead us to separate the translation
upload from the package upload. So the translators could work without
pestering the developers (not that I dislike translators, but I know
that some people have problems with them)

Well,

It's a quite immature idea yet, but I think it could be interesting...

What do you think?

daniel