Marco Trudel writes:
> Andrew Haley wrote:
> > Stuart Ballard writes:
> > > Andrew Haley <aph <at> redhat.com> writes:
> > > > No, that's not what I mean by illegal. It's illegal in the sense that
> > > > the specification places requirements on the implementors of
> > > > subclasses, and that a subclass which does not meet these requirements
> > > > is not a well-defined Java program.
> > >
> > > public class Foo {
> > > private String x;
> > > public Foo(String x) {
> > > this.x = x;
> > > }
> > > public boolean equals(Object o) {
> > > if (!(o instanceof Foo)) return false;
> > > return x.equals(((Foo) o).x);
> > > }
> > > public int hashCode() {
> > > return x.hashCode();
> > > }
> > > }
> > >
> > > This class is clearly intended to meet the spec of equals() and
> > hashCode() and
> > > does so correctly in the vast majority of cases. But not quite: new
> > > Foo("").equals(new Foo(null)) is false, but new Foo(null).equals(new
> > Foo("")) is
> > > NPE.
> > >
> > > Is this not a well-defined Java class?
> >
> > AFAICS it isn't: the spec requires equals() to be symmetric for all
> > non-null x and y:
> >
> > "for any non-null reference values x and y, x.equals(y) should return
> > true if and only if y.equals(x) returns true."
> >
> > The result of any operation that involves this class is therefore not
> > well-defined, and whatever happens, it's not a bug.
>
> Allow me to split the discussion in two parts:
> 1. commit the patch
> 2. add a mauve testlet
>
> 1:
> Are you against or for committing this patch? Everyone so far agreed to
> commit it. I think the reason is clear: Keep classpath as usable as
> possible and don't handicap users. They might not even be responsible
> for the bug in the code they use.
Fine by me.
> 2:
> A mauve testlet is not really required. A comment would do the same
> thing as well (see new attached patch).
> Anyway, a mauve testlet would be more reliable. There's no need for a
> not legal or not well-defined program. See attached example.
Fine by me. It's clear that it's not a conformance test, but a
compatibility test.
Andrew.
