Re: running code vs. formal testing methods

[Graham Klyne <GK at dial.pipex.com>]

At 17:42 31/07/99 -0600, Marshall Rose wrote: 
>look at the number of standards-track rfcs that have  progressed past the
proposed level. i think you would be hard-pressed to find a  comperable
number of real-world protocols specified and validated using formal
methods.  history is pretty clear that the way to get success in the
protocol arena is to focus on well-written natural-language  specifications
iteratively developed with multiple interoperable  inplementations.  /mtr 


I think it's worth noting that there is SOME track record of formal
specification in RFCs -- the one that first comes to mind is ABNF.  On the
semantic front, RFC 821 uses state diagrams to describe key aspects of
processing semantics.

My point:  this isn't a black-and-white issue.  Some formal techniques have
a track record of demonstrated success (a kind of meta-running-code?).

I think the winning techniques tend to be those that address a well-focused
problem rather than attempt to solve all specification issues.  Maybe this
is the IETF culture at work:  progressing by incremental well-understood
steps rather than trying to digest the problem whole?

#g

------------
Graham Klyne
(GK at ACM.ORG)