Re: CONSENSUS TEST: Fragmentation handling

[Francis Dupont <Francis.Dupont@enst-bretagne.fr>]

 In your previous mail you wrote:

   At 11:25 PM +0200 4/6/04, Francis Dupont wrote:
   >  In your previous mail you wrote:
   >
   >    I proposed was that the stateful fragment checking be a SHOULD, with
   >    the explicit intent that an implementation may choose to not support
   >    #3 because of performance considerations.
   >
   >=> I can't see why a MAY is not enough.
   
   well, I wanted to have one MUST between #2 and #3. Originally #2 was 
   MUST and #3 was MAY.  but Tero argued against the MUST for #2, so I 
   changed #3 to a SHOULD and #2 to MAY, as a compromise. hope that's 
   clear :-)
   
=> now I understand. I am in favor of MAYs for #2 and #3. BTW the #2
makes the wrong assumption that one can always get the upper protocol
from an initial fragment (there are counter-examples in IPv6).

   the SHOULD does apply to all implementations, but a SHOULD means that 
   the implementation SHOULD offer this feature, unless there is a good 
   reason not to. I gave what I considered to be the good reason.
   
=> I (can) consider you gave the good reason to give only a MAY to #3.

Thanks

Francis.Dupont@enst-bretagne.fr