Philippe Houdoin wrote: > Le 31 oct. 2012 11:10, "Ingo Weinhold" <ingo_weinhold@xxxxxx> a Ãcrit : > > > > Philippe Houdoin wrote: > > > Networking file systems are obvious clients for such move but > > > ont necessarily the only ones. > > > > Could you elaborate on that part? I don't see why the userland MIME > sniffing shouldn't work network FSs. > > Indeed. > The question then is why NTFS needs to do such mapping by itself, in fact ? Unlike our FAT implementation (and probably others) it doesn't do any kind of mapping/sniffing. It just supports a virtual (in-memory) "BEOS:MIME" attribute that can written and read (it's initially not set), so that the userland sniffing works. That is the sniffing part always works, but setting the "BEOS:MIME" attribute to the resulting type fails, if there is no attribute support. IMO, that's the most sensible solution. Hence Axel suggested to generalize it via a kernel module or an FS layer. CU, Ingo