Re: gnome

[Jimen Ching <jching flex com>]

On Tue, 8 Dec 1998, Havoc Pennington wrote:
> You are not considering all the design goals. One of them is to work on
> all modern Unix variants. We can't change the kernel on all Unix variants. 
> Therefore, any functionality we want to be an essential part of Gnome must
> be implemented outside the kernel. Period. 

I'm getting the feeling that all of the people who responded to my mail
either didn't read what I wrote or didn't understand.  In either case, it
is very hard to continue the discussion.

Just a short comment on the above qoute though; not all modern Unix
variants support networking and physical disks (no hard disks).

My conclusion still holds; it is a mistake to take a kernel level API and
place it in a library.  This applies only to monolithic kernels.  Kernels
like the HURD and Plan9 are perfect for this.  But those kernel support
the old method as well.  So in my professional opinion, it is wiser to
stick with the old method for now, until the new kernel philosophies are
better understood.  Putting a kernel API in a library is just bad design,
when applied to monolithic kernel.

Also, where are all the fears coming from?  It won't work on this system,
it won't work on that system.  Before MC was ported to FreeBSD, it didn't
work on that system either.  Everyone is worried about all of the systems
out there that doesn't support this kernel feature.  Why not worry about
getting a good design and a functional environment before we worry about
these systems?  Btw, the *BSD's does not have a problem with putting the
vfs in the kernel.  So stop worrying.  People are porting new file systems
into the OSS kernels all of the time.  I don't see people complaining that
it will bloat the OS or it will make the system unstable or it will make
maintaining the source harder or it will do this nasty thing.  I didn't
see this type of whining when the mtools were moved into the kernel.

--jc
--
Jimen Ching (WH6BRR)      jching@flex.com     wh6brr@uhm.ampr.org