Re: More finalize woes

[Tom Tromey]

>>>>> "Dalibor" == Dalibor Topic <address@hidden> writes:

Dalibor> Your generator ends up with a different set of test cases for
Dalibor> Classpath and for JDK. This may not be what you want but it's
Dalibor> what you get when methods are left out ;)

This is an interesting example.  I'm not so sure that what you end up
with is incorrect in a significant way, though.  Or perhaps you'd
prefer to write your test generator to look at all the methods,
inherited or declared, in a given class.

Dalibor> beside, the 'least surprise' argument applies here as
Dalibor> well. If I extend Classpath's Deflater to implement my own
Dalibor> SuperfastNativeDeflater, then I'd like to rely on finalize()
Dalibor> calling end(), as the spec says.

In cases where there is a finalizer with a documented effect, I agree
we must implement it.  I'm just talking about trivial finalizers -- I
think we don't need those.

Tom