FranÃois Revol wrote: > On 23/10/2012 12:47, Ingo Weinhold wrote: > > BTW, Haiku has support for the GNU xattr API (src/libs/gnu/xattr.cpp). > > That allows ported software that already supports xattr (e.g. rsync) > > to be built without change. The xattr support layer maps the BFS attribute > > and type according to the scheme "user.haiku.<attrname>#<type>" to xattr > > attribute names (<type> is either the four character variant, when all > characters > > are printable (isprint()) or an 8 char hex representation). > > An exception are BFS attributes with the type B_XATTR_TYPE; > > for those the name is identity mapped. In the other direction (xattr > -> BFS attribute) > > the scheme is simply reversed, i.e. the encoded (type, name) pair is > decoded and for > > genuine xattr attributes the name is identity mapped and they get the > type B_XATTR_TYPE. > > > > Ideally our ported FS modules that support xattrs (ext*, ReiserFS,...) > > would implement the same scheme. > > That would e.g. make it possible to rsync to a remote Linux, reboot > that machine to Haiku and read the > > typed attributes as they were originally. If tar supported xattr on > other systems and would use the same > > name/type encoding scheme all kinds of roundtrips were possible, which > would be very nice IMO. > > Except this scheme alone won't work when files are moved across other > non-linux filesystems, cf. > http://dcpapers.dublincore.org/index.php/pubs/article/view/3633 Yes, such a scheme would be nice. Alas it needs to be implemented on all platforms. If there only were a Haiku developer who cares enough to start with that in Haiku. :-P > Known attributes like the MIME type could be converted directly to the > corresponding linux one though, cf. > http://www.freedesktop.org/wiki/CommonExtendedAttributes > http://www.freedesktop.org/wiki/Specifications/shared-filemetadata-spec I'm a bit skeptical about that. Such a mapping would also have to happen in FS implementations which are supposed to store data without interpreting them semantically. I already dislike the extension based file type attribute faking in our FAT implementation. CU, Ingo