[Tmrg] TCP evaluation suite

[fred at cisco.com (Fred Baker)]

On Jun 30, 2008, at 1:26 PM, Sally Floyd wrote:

> I haven't read this version yet, but I don't think it needs SHOULD,  
> MAY, etc.  Those are usually used only for protocols.  (For  
> Informational RFCs that were also targeted as Best Current Practice,  
> and became Best Current Practice RFCs, you could look at RFC 5033,  
> or RFC 2914.)

actually, they are intended for requirements documents. The first RFC  
where one could construe "SHOULD" being use that way is RFC 827, in  
which Eric indicates that in a certain circumstance a "gateway" SHOULD  
do something in particular. But it's not the word that is capitalized  
per se, it's two sentences. The first document in which it is defined  
as we use it now is RFC 1122/1123 and later 1812, and the question  
before the house (per section 1.3.2 of RFC 1122) is implementation  
compliance - an implementation can be said to be "conditionally  
compliant" if it implements all the MUSTs, and "fully compliant" if it  
also implements the SHOULDs. To be honest, I think most documents that  
use RFC 2119 language mis-use it.

I am amused by RFC 2119's "guidance":

    Imperatives of the type defined in this memo must be used with care
    and sparingly.  In particular, they MUST only be used where it is
    actually required for interoperation or to limit behavior which has
    potential for causing harm (e.g., limiting retransmisssions)  For
    example, they must not be used to try to impose a particular method
    on implementors where the method is not required for
    interoperability.

I wish he had said

    Imperatives of the type defined in this memo MUST be used with care
    and sparingly....

:-)