Hi Doug:

Larry is checking the LoD for general Java programs using Binoy's cd.
His question is about how to get at the interface of a Java class
for which only the .class file is given.
How does one do this kind of Java compile time reengineering?

Doug writes:
>Any chain longer than 1 is a violation, isn't it?

That is true according to the object form of the LoD. But Larry checks
the class form. But I think your suggestion is actually a very good practical
mixture of class and object form. We view any chain longer than 
1 as a violation. It is also much easier to check. Of course,
this.b().c() is length 1.

So Larry has then only to check calls like:
H  h = a.b();
This does not require an interface look up of b since we know the return-type
is H.

-- Karl





>From dougo@ccs.neu.edu Tue Dec  2 22:21:50 1997
>Cc: Karl Lieberherr <lieber@ccs.neu.edu>
>Subject: Re: Memory and java class interface questions
>
>Larry Dodds writes:
>
> > My other problem is dealing with calculating the violations of LoD. When a
> > statement like the following is executed:
> > 	H  h = a.b().c();
> > 
> > The problem is that I would normally look up the b() method and determine
> > the type of object that it returns so I could determine if the c() call is
> > a violation.  However, in Java, the b() could return a class that is not
> > in the Java system I am parsing or worse is not available to me in any
> > way.  For example what if b() returns a class that is in java.io or
> > java.util.  I this case, I have no access to the interface without parsing
> > the .class file.
>
>Demeter considers all external classes (i.e. those not in the class
>dictionary) to be terminal classes.  Thus you shouldn't need to know
>what methods are on them.
>
> > This problem obviously continues with longer chains. 
>
>Any chain longer than 1 is a violation, isn't it?
>
>--Doug
>