On Wednesday, September 3, 2003, at 09:30  AM, Jonathan Stone wrote:

> Not taking sides one way or the other, but: surely someone brought up
> the point that userland trampolines doesn't work[*] for binary 
> emulations?
> (consider supporting the "old way", for old nebbsd binaries, as an 
> emul...)

Sure.  The kernel still continues to provide the trampoline for 
binaries which do not provide their own.  I was quite careful to 
maintain compatibility.

I guess the point is that there are already "different paths" (which 
really means "multiple small functions that convert the internal 
representation to the external representation"), and I just don't see 
the issue with continuing to provide the traditional sigcontext "path" 
for binaries that do not request siginfo.

         -- Jason R. Thorpe <thorpej@wasabisystems.com>