Re: Eclipse

[Tom Tromey]

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

Dalibor> Given that classes outside JAXP (beans, imageio) depend on
Dalibor> it, I'd be for making GNU JAXP (or another JAXP
Dalibor> implementation, if necessary) a hard requirement.

Me too, for the same reasons.  It is just like the PKI thing Casey
asked about last week.  Classpath's core has to be self-contained,
meaning all dependencies have to be integrated.

Dalibor> If someone really needs a JDK 1.0 subset of Classpath, they'll have to
Dalibor> put enough work into rewriting the core classes anyway that it would
Dalibor> warrant a fork, in my opinion. If someone just needs a simple way to
Dalibor> jar a subset of classes in the build, then that should be fairly
Dalibor> trivial to implement.

I think that things like J2ME and the various small profiles exist
because the reference implementation is not free software.  We can do
better than that.

For instance, someone could write a tool to cut down the class library
to custom-fit your application.  Or we could mark up the library
source with Classpath-specific 1.5 attributes that end up in the class
files, and that are used to remove parts of the classes later on.
(There are probably lots of cool things we could do with attributes,
something worth brainstorming about.)

I know this approach avoids some of the things needed for real
compatibility, for instance when semantics change over time.  But is
anyone really clamoring for JDK 1.0 compatibility?  My impression is
that when this has come up on this list it is just as a way to get a
release out that has 100% compatibility with *something* -- but for me
that is not an interesting end in and of itself.  In fact, I think a
release that is compatible with just JDK 1.0 is kind of laughable at
this date.

Tom