Re: Secure Shell: Milestone Update.

[Bill Sommerfeld <sommerfeld%sun.com@localhost>]

On Thu, 2005-03-17 at 23:28, Sam Hartman wrote:
> >>>>> "Bill" == Bill Sommerfeld <sommerfeld%sun.com@localhost> writes:
>  You want to avoid having too
> many features; treating each MUST/SHOULD/mAY as a feature is clearly
> too much effort and in many cases results in untestable features.
> However if your feature list is too course you risk appeal or a
> discuss vote at the IESG.

though clearly major features which are known to have caused interop
difficulty in the past (rekey comes to mind) should be on the list as
distinct test points.

> Avoid getting yourself into a situation where you need to test a MUST
> NOT.  For example, it would be difficult to verify conformance if you
> listed "MUST NOT crash on badly formed input," as a feature.  In
> general, if testing something to go to draft requires you to write
> code to generate invalid input or to create a can't happen situation
> then you are doing something wrong and probably doing too much work.

That sounds like very good advice.  Such tests are often valuable but
are probably better left as regression/stress/abuse tests. 

							- Bill