> I agree that's the wrong fix; I generally suppress the kernel
> definitions. For most of the things I've run into, the only things which
> are in the <linux/...> header files which are not provided by glibc are
> file-system-related things -- mount and linux-nfs in particular use these
> a bunch. For that, it's not an easy thing to figure out what the 'right'
> thing to do, since as you mention, some of the types may differ between
> the kernel and the libc.
Yes; I've heard about that before. The general feeling is that the least
evil approach is for programs like `mount' to contain their own private
header files with copies of the definitions they need. Since they're
Linux-specific anyway this ought not to be a big problem for portability.
The other main offenders are the network stuff, which hopefully we now
have covered by the files in /usr/include/net, and SCSI, which is still a
bit of a dubious area.
p.