is@jocelyn.rhein.de said:
> I think you need two conversons:
> kernel: filesystem-charset to utf-8
> then
> userland: utf-8 to LC_CHARSET. 

That's what I meant. The filesystem-to-utf8 part
would be trivial however.
It is perhaps too optimistic to assume that each filesystem
charset can be expressed as utf8... I know too little about
the matter to tie myself down to utf8 here. The important part
imho is that no information is lost on the way from the filesystem
to userland.
It's generally dangerous to pass more than ascii through the
existing system call interfaces - an unsuspecting application
can be confused, or the terminal messed up. While we can
harden most applications, we can't change the vt100 specification...
It would need a new API.

Hmm - just got an idea: What about making the codeset to
use in kernel-userland communication a property of the process
instead of the mount, doing translation in-kernel as Jaromir
suggests? Default would be "ASCII only", and a process could
select itself what it wants, based on environment and/or options?
Syscall or sysctl...

Complex matter... definitely deserves more thoughts.

best regards
Matthias