[plug] RE: Linux backward compatability baggage

[Mike Holland]

On Sat, 7 Aug 1999, David Campbell wrote:

> Providing a clear distinction between kernel and program is kept 

This gets me thinking ... (danger!)

Why define a fixed sytem call interface at all? Isnt that just creating
baggage that all future kernels will need to support?
   Instead, all the "system calls" could become library functions.
That libray would include the current system calls. 
Applications would use that library to be independent of kernel version.
  Then the kernel interface would be free to change, possibly shrinking
and moving code to user-mode, where beneficial. Old system calls could be
mapped to
new ones in the library. Each new kernel would have its own user-mode
system library to map to system calls.
  This would add the overhead of a function-call to every existing system
call, but that would be insignificant compared to the context-switch.

  OK, I'm sure there are lots of very good reasons for not doing this, and
applications can already use libc. But why should applications be allowed
at all to make direct system calls?
  I admit I've never written a line of kernel code in my life (or do
MS-DOS device drivers count?), so dont be too hard on me.


Mike Holland <mike at golden.wattle.id.au>            Perth, Australia.
                          --==--
	Many are cold, but few are frozen.