Re: setsockopt() compat issue

["Greg A. Woods; Planix, Inc." <woods%planix.ca@localhost>]

On 19-Oct-08, at 11:53 PM, Brian Ginsbach wrote:

> On Sat, Oct 18, 2008 at 05:53:22PM -0400, Greg A. Woods; Planix,  
> Inc. wrote:
> > A better argument as to why NetBSD uses I32LP64 on those  
> > architectures
> > where ILP64 might make more sense would be to say that it was  
> > simply easier
> > to follow the herd than to try to ride against them all no matter  
> > how bogus
> > were the arguments used by the herd to choose their direction.
>
> You don't really want to see a BSD networking stack on a ILP64
> system.  It isn't pretty.  Cray's UNICOS and UNICOS/mk used a BSD
> networking stack and both were ILP64.  The code is fully of fun
> stuff to make up for no native I32...


Which is one of the elephants in the room -- and a great testament to  
how un-portable the BSD networking code really is, or rather how  
massively abundant the assumptions are that it makes about the  
implementation language and the underlying target hardware architecture.

C programmers really are, in general, horrible at writing decent  
performing, easy to understand, and use, low-level, portable, data  
marshalling code.

--
                                        Greg A. Woods; Planix, Inc.
                                        <woods%planix.ca@localhost>

Attachment: PGP.sig
Description: This is a digitally signed message part