Peter von der Ahé, Sun lead for javac
Veloso
der Ahé, (http://java.sun.com/developer/Meet-Eng/vonderahe/) Sun's
tech lead for javac, Sun's Java compiler, who
claims that if developers know how to use the Compiler API, it will
provide a lot of indirect benefits. But are the benefits sufficient to
repay the time and effort required? Does anyone have experience with
this?
-- He argues, in regard to generics: "Rather than trying to provide
reified types for all instances, why not simply accept that some
instances do not have reified type information, because they are
compiled with a 1.4.2 compiler, or for other reasons? This turns
reification into a best-effort problem: if you use a raw type, then no
type information is reified, but if you avoid raw types, you can take
advantage of the additional reified type information."
Does turning reification into a "best effort problem" make sense to
anyone?
-- His proposal for an alternative to type inference for local
variables involves having the compiler provide the static factories.
-- He discusses the Kitchen Sink Language, (https://ksl.dev.java.net/)
which is a place you can modify javac and play around like the
ultimate geek. Has anyone been engaged in any experimentation there?
He's obviously a smart, creative guy; I lack the competence to guage
the validity of his ideas though, so I thought I'd throw it up in the
air and see what lands...
Tom Hawtin
> Java.sun.com, Sun's main Java site, has an interview with Peter von
> der Ahé, (http://java.sun.com/developer/Meet-Eng/vonderahe/) Sun's
> tech lead for javac, Sun's Java compiler, who
> claims that if developers know how to use the Compiler API, it will
> provide a lot of indirect benefits. But are the benefits sufficient to
> repay the time and effort required? Does anyone have experience with
> this?
I think he's claiming benefits for users of certain pieces of software
that will be able now have a standard way of compiling Java source code,
that also has the advantage that it doesn't have to start up a new JVM.
> -- He argues, in regard to generics: "Rather than trying to provide
> reified types for all instances, why not simply accept that some
> instances do not have reified type information, because they are
> compiled with a 1.4.2 compiler, or for other reasons? This turns
> reification into a best-effort problem: if you use a raw type, then no
> type information is reified, but if you avoid raw types, you can take
> advantage of the additional reified type information."
>
> Does turning reification into a "best effort problem" make sense to
> anyone?
It doesn't introduce any extra incompatibilities. You can use
reification information if you want to. Seems just about perfect to me.
You can't use reification for code compiled with pre-JDK7 javac or
without generics, but then there aren't many schemes where you could.
> -- His proposal for an alternative to type inference for local
> variables involves having the compiler provide the static factories.
I don't see why type inference can't be done for constructed types (have
to tread quite carefully with terminology here - constructors have
exactly the same type inference as methods, but generic constructors
(whether to construct generic types or not) are not well known).
> -- He discusses the Kitchen Sink Language, (https://ksl.dev.java.net/)
> which is a place you can modify javac and play around like the
> ultimate geek. Has anyone been engaged in any experimentation there?
Nope. And I wouldn't really want to under GPL. Nothing stopped anyone
playing about with javac beforehand (like Neal Gafter has done).
Tom Hawtin