dolecek@ics.muni.cz said:
>  Most likely, the filesystems will accept the desired codeset as mount
> option. 

Wouldn't it be possible to pass the filename information to userland
without loss and do the conversion there, controlled by the user's
locale settings? Not that I expect strangers to mount my windows
disks (if I only had any:-) anytime soon, but it would keep bloat out
of the kernel.

With UTF8, this could even work without changes to the system interface.
(Correct me if I'm wrong!)

> The engine tries hard to DTRT with Unicode characters which do not map
> into target codeset. In that case, the character's lower 8 bits are
> used to form a random 8bit character used as a replacement, 

There is already some kernel code which deals with Unicode characters:
wscons passes unicode to the display drivers. (At the time the terminal
emulation initializes - later display specific font indices are
transferred for performance reasons.) There is already the need to
deal with character replacements. For display purposes, it focuses on
similarity from the visual point of view of course, which might
differ from semantical ones.
In any case, it would look strange if we had two code subsystems for
more or less similar tasks. So if it shows that in-kernel filename
translation is really necessary, we should try to get this together.

best regards
Matthias