couple questions on method handle combinators

Paul Sandoz paul.sandoz at oracle.com
Wed Jun 21 17:08:30 UTC 2017


> On 16 Jun 2017, at 09:02, Roland Westrelin <rwestrel at redhat.com> wrote:
> 
> 
>> Most likely, there's something missing for Q-typed shapes which disables
>> SELECT_ALTERNATIVE intrinsic.
> 
> Thanks Vladimir, Paul and Maurizio. As I understand, the code shape that
> I worry about can't be generated currently with GWT. I'll revisit that
> issue when GWT has improved.

Even the GWT test i wrote that now works with you patch?


> 
> Do we expect to see a lot of uses of the loop method handle combinator
> with mvt? I wonder if it's worth looking into loop combinators more?
> 

I expect loop combinators to be used more. This will require us to beef up the byte code generation to be on par with that for L shapes.

However, in general i anticipate a combination of method handles and explicit code generation may be required by a developer. I sense it’s gonna be hard contain the value type energy within a method handle combinator container field. Those exotic value type particles are unstable and wanna collapse into their lower energy state boxes. Some code generation rods will help stabilise things.

Paul.


More information about the valhalla-dev mailing list