From: Brian Candler <B.Candler@...>
Date: 2003-07-01T22:48:21+09:00
Subject: Re: Message catalogs (I18N) overnight hack...

On Tue, Jul 01, 2003 at 10:25:51PM +0900, Hal E. Fulton wrote:
> But there are several little issues like that, that I'm not addressing yet.
> Someone mentioned the decimal point issue. I hate to think about that.

And then you get into date formats (US middle-endian), three-letter
abbreviations for month names, ... ugh.

I'm sure these problems must have been gone through before. I noted a while
ago under FreeBSD that gmake had some strange dependencies:

# pkg_add gmake-3.79.1_1.tgz
pkg_add: could not find package libiconv-1.7_5 !
pkg_add: could not find package expat-1.95.2 !
pkg_add: could not find package gettext-0.11.1_3 !

It seemed strange to me that a 'make' utility would have a dependency on an
XML parser. It turns out that gettext is the GNU way to deal with this:

[man gettext]

DESCRIPTION
       The  gettext program translates a natural language message
       into the user's language, by looking up the translation in
       a message catalog.

I imagine that the source format for these message catalogues is XML, and
hence the requirement on an XML parser (although IMO gettext should be split
into two: a client side which just reads the message catalogues, which
appear to be in a binary format, and a -devel package which includes the
XML-to-message-catalogue tools. But I digress).

I note there's a ruby-gettext library already. A quick browse and it seems
to require indexing by message-ID rather than the original "untranslated"
text.

That approach makes logical sense to me - decouple *all* language text from
the source, rather than have language A in the source and languages A,B,C,D
in the translation database. (Otherwise, whenever you change a message in
the source you'd have to update the corresponding language A entry - a
violation of the DRY principle)

Cheers,

Brian.