[PRL] Precision of Reachability analysis
Johan
johan at ccs.neu.edu
Tue Jan 13 19:11:10 EST 2004
William D Clinger wrote:
> As proof that I don't understand your question, I will comment on this:
>
>
>>Because of dynamic dispatch and interfaces, I'd suspect that the
>>analysis will quickly become very coarsegrained. In other words, I would
>>expect that typical java programs have control-flow graphs with a high
>>degree of both interconnection and thus strong-connection.
>
>
> A strongly connected control flow graph implies there are absolutely
> no leaf methods. Simple static counts typically show that 30 to 70%
> of the methods are leaf methods. So I doubt your inference.
>
> Will
Not all of it will be strongly connected; that would have a lorenz curve
flat lorenz curve. The question is how quickly the connection count
rises.
More information about the PRL
mailing list