LocalesThatDontSuck

Revision 8 as of 2005-11-02 22:07:32

Clear message

Summary

In order to serve our user communities better, we need to be more responsive and proactive with localisation updates.

Rationale

Getting locale changes into upstream glibc is unnecessarily hard. The glibc maintainer (correctly) requires proof of the correctness of a change before it's included, but the onus of demonstrating that change falls to the glibc package maintainers. The package maintainers are frequently not qualified to provide this proof, and cannot answer upstreams questions to the needed degree of satisfaction. The glibc upstream maintainer also has a well deserved reputation for being difficult to approach with changes.

In addition, updating locales requires a complete rebuild of the glibc package. This is a very time and resource intensive build just for the sake of arch-indep files.

Use cases

  • A user from an unsupported locale would like to submit his entry. We accept it via Rosetta, and it gets applied during the next lang-pack build.
  • New locales need to be integrated. The locales maintainer merges with lang-pack, and avoids having to force a rebuild of glibc.

Scope

  • Locales / Belocs-locales-data(universe) packaging
  • Langpacks

XXX:Does it affect all of every langpack? Surely not ;-)  -- smurf

Design

Currently locales come from glibc proper. Implementation will move locale data into lang-packs for easier maintenance, and to allow us to move to belocs-locales, which will provide more up-to-date data.

Implementation

First stage is to remove locales package from glibc build. The locales binaries still need to be provided by libc6, since they are used to generate the binary, arch-dependent locales. The locales-gen script from the locales package would need to be moved to libc6 aswell. The locales-gen script needs to be modified as noted below. libc6 would then need to conflict with locales.

Second, the belocs-locales data needs to be merged into language packs build. When a lang-pack is installed, it should place a file called /etc/locales/supported.d/<lang-name>. The postinst script would then call locales-gen with the lang-name. Locales-gen would parse this file, and (re)generate all the locales listed therein (which may be more than one, for e.g. english, there is en_US, en_CA, etc).

XXX: Would the existing locale selection dialog, with its large list of glibc locales, be kept-as-is / truncated-to-installed-locales / removed? -- smurf

Lang-packs would include all of the proper locales for that language (e.g. en_* for English).

Code

Requires somewhat extensive packaging changes for the lang-packs. The locales-gen script needs to be modified to understand /etc/locales/supported.d/<lang-name>.

XXX: Exactly what needs to happen?  -- smurf

Data preservation and migration

Belocs-locales seems to want to be a superset of glibc locales, but it appears to be missing several locales that are in glibc proper. If anything is missing from belocs, then the missing part needs to be investigated for inclusion as well. It may be that ones missing from belocs just don't work, or are no longer needed.

BoF agenda and discussion

Attendees generally agreed that having locales in glibc was sub-optimal, and that not being able to get locales merged upstream was just an added difficulty.

Consensus was that moving to lang-packs now, would also enable us move maint of the locale data to Rosetta at a later date (hopefully, dapper+1).

A new spec needs to be created for Rosetta, to plan for the move of locales data.

CategorySpec