Comments on Scala Blog: Manifests: Reified Types

your example is great, it helped more than anythin...

2011-09-19T03:29:53.671-07:00

your example is great, it helped more than anything elsebuy research papers

Richard: That's a very neat trick! I'm im...

2011-01-31T20:12:56.125-08:00

Richard:

That's a very neat trick! I'm impressed. The downsides are that you can't retrofit others' generic classes with type parameters, there's no way to guarantee such a class is instantiated in such a reified manner, and you can't propagate a requirement that a generic type be reified up the call stack.

To answer your question though: Scala can compile multiple specialized implementations of a class. Look at Scala's @specialized annotation.

--j

Actually, you can obtain compile time information ...

2011-01-31T18:37:22.930-08:00

Actually, you can obtain compile time information at runtime with Java, overcoming type erasure.

This is possible if you employ anonymous classes. It's a dirty trick, in fact, but it can be done. See my article about this:
http://www.jquantlib.org/index.php/Using_TypeTokens_to_retrieve_generic_parameters

So, once ...

1. it is possible to keep type information at runtime (as I explained in my article)

2. scalac is a compiler which can potentially "implement it right", creating all data structures needed at runtime when RTTI is needed (and eventually not needing at all any trick like I explained),

...my question is:

Why scalac is not able to produce multiple "incarnations" of a class, pretty much what C++ templates do?

Thanks

Richard Gomes
http://www.jquantlib.org/

Hi Jorge,

Yes, I agree that .NET got it rig...

2008-12-16T20:48:00.000-08:00

Hi Jorge,

Yes, I agree that .NET got it right in that respect.

I would also like to see a compiler plugin for that. In fact, I wanted to work on one, but time (or lack thereof) is the issue as usual.

Yes, I am quite interested to see how those efforts turn out.

By the way, two related things that are interesting are the -optimise flag for scalac and scalar replacement for HotSpot. I intend to do some tests with the former, and I've done some with the latter:

http://blog.juma.me.uk/2008/12/17/objects-with-no-allocation-overhead/

Hi Ismael,

Yup, there are lots of pitfalls ...

2008-12-03T07:05:00.000-08:00

Hi Ismael,

Yup, there are lots of pitfalls on the JVM. Unfortunately this is another area where C# "got it right".

As you explore in your blog post, this can be remedied with some compiler hacking. It'd be interesting to see an open source compiler plugin that produced specialized generic functions for use with primitive collections.

And thanks for the links! I wasn't aware of those efforts, and they certainly are interesting.

Hi Jorge,

Just be careful not to use anonym...

2008-12-03T02:42:00.000-08:00

Hi Jorge,

Just be careful not to use anonymous functions to work with the primitive collections or the results might not be what you expect.

In terms of generic specialisations, the following links are interesting:

https://lampsvn.epfl.ch/trac/scala/changeset/16665

https://lampsvn.epfl.ch/trac/scala/wiki/AnnotatedTypeParams

Ismael