Random numbers

[Simon B Jones]

Recently Jan Krynicky wrote:
> 
> IMHO, it'd be nice to have it again. Random numbers seem to me a 
> suitable place for droping ref.transparency. If I work with random 
> numbers I don't suppose to get always the same result.
> 
> Jenda
> 

But...
... (perhaps) the Clean compiler does common subexpression elimination (or
other similar transformations/optimizations) - or may do at some time in 
the future. Dropping the ref transp requirement may lead
to erroneous object code now, or a sudden change in program behaviour if
such transformation/optimization facilities are introduced in the future.
Ref transp guarantees that such changes do not alter program behaviour - in
software engineering considerations, I would say that that is a property we
must not sacrifice!

Regards  Simon

--
Dr Simon B Jones              
Dept of Computing Science
             and Mathematics    | Email: sbj@cs.stir.ac.uk
University of Stirling          | URL: http://www.cs.stir.ac.uk/~sbj
Stirling                        | Tel: (UK) 01786 467434 (Int) +44 1786 467434
SCOTLAND  FK9 4LA               | Fax:       .... 464551        ....... 464551