Theodore Ts'o wrote:
Olaf sent me two de.po files --- one in UTF8, and one in 8859-15. What are the tradeoffs between using one charset versus the other?
There are no tradeoffs; both encodings can encode the entire catalog just fine.
You always need the messages in your locale's encoding. So in de_DE.UTF-8, you need the messages in UTF-8; in de_DE.ISO-88599-15,
you need the messages in Latin-9.That makes it look like you need two catalogs, and indeed, you did, for older versions of gettext. Current gettext implementations transparently convert from the catalog encoding to the user encoding, so you need to ship only one catalog.
In theory, you should ship the one where the implicit conversion is less frequently needed, i.e. the encoding that the majority of users is likely to use. Unfortunately, this is difficult to determine and changes over time. Fortunately, it doesn't really matter.
Since, in the long run, some of us expect Linux to use UTF-8 uniformly for all languages, I'd recommend to ship the UTF-8 catalog. The "upstream" distribution may include both, for installations where gettext cannot perform automatic conversion (i.e. non-GNU or outdated gettext).
Regards, Martin