> Past experience shows that non-IO intensive benchmarks are useful.
> However, keep in mind that just having a base program in
> itself is not sufficient -- can you think of a useful
> tracematch to apply to it? Will it be expressible in other
> formalisms? Only if you have a very good trace property
> connected with it is it worthwhile trying to debug it...
Good point. So far I have only thought about the standard ones like the
hash set and iterator examples. Aprove makes intense use of hash sets
and indeed it's this tool where this example arised from, because that
bug was actually causing some trouble in Aprove in the last year. I
haven't thought about any other, more specialized properties yet - but I
will.
Eric
Received on Fri Jan 20 16:13:56 2006
This archive was generated by hypermail 2.1.8 : Sat Jan 21 2006 - 03:20:09 GMT