Re: n versioning is there any other better approach

[Peter Amey <peter.amey@praxis-cs.co.uk>]

Jeffrey R. Carter wrote:
> Larry Kilgallen wrote:
> 
>>
>> Do you recall whether there was any commonality in the versions, other
>> than the specification ?   Like language, background of implementers,
>> deadline, etc. ?
> 
> 
> Sorry, no, just that the versions were developed independently.
> 

Are you perhaps thinking of the Knight-Leveson experiment?  The paper is 
  on line somewhere.

ISTR that they got teams of students independently to implement the same 
specification.  The interesting result is that the programs all tended 
to fail the same, hard, test cases - e.g. area of a figure in the case 
where all the vertices were 0,0 or intersection of lines that are 
parallel.  The conclusion was that N version gives some protection from 
random errors but that real errors aren't very random!

Peter