Re: [fuse-devel] fuse path encodings

[John H. <joh...@or...>]

On 03/12/09 22:58, Matt Turner wrote:
> I was wondering about dealing with non-ascii characters in paths. Simply
> put, in what encoding are paths passed to a libfuse client?
>
>    


Unless something has happened since I last looked (and I don't think it 
has) the kernel and fuse neither know nor care what the encoding used 
for a file name is.  With two exceptions.  The octets 0x2f (/) and 0x00 
are used to separate pathname components and terminate pathnames 
respectively.   Of course, you would have trouble creating a file called 
"." or ".." (because they already exist) and if you had a file system 
that didn't have those all kinds of things would break in interesting ways.

So, I can create a file called "\x88\x68", "\x68\x88", "\xe8\xa1\xa8" or 
"\x95\x5c" (representations of U+8868 in UCS-2LE, UTF-16, UTF-8 and 
SJIS) and how they look depends on the locale that the terminal is using 
to interpret character representations.  The last one is interesting 
because it ends with 0x5c (\) but the file system doesn't much care 
about that.

It is possible to create a file system that, for example, stores its 
pathnames as vectors of utf16 strings but you would have to make a 
decision about how to convert what the is passed into the kernel into 
UTF-16 -- that would have to be a mount option or something that you 
derive from somewhere (eg nl_langinfo(CODESET))  but it wouldn't stop 
someone creating a file called "\x95\x5c" because they happen to want to 
use SJIS but you were expecting UTF-8.

Does that make sense?

jch