Keeping `new Integer` source compatibility with unnamed factory method

[Mateusz Romanowski]

Hi Remi,
Thanks for your reply.

I completely forgot about code which needs the two references for new
Integer(2) to be different... no silver bullets here..

Unless... could specialization help -- int being a primitive species of
Integer while having another reference species available for _mature_ code?
I dimly remember both Fred and John, I think, making such remarks...

Regards,
Mateusz


On Wed, Nov 4, 2020 at 10:16 PM Remi Forax <forax at univ-mlv.fr> wrote:

> Hi Mateusz,
>
> ----- Mail original -----
> > De: "Mateusz Romanowski" <romanowski.mateusz at gmail.com>
> > À: "valhalla-dev" <valhalla-dev at openjdk.java.net>
> > Envoyé: Mercredi 4 Novembre 2020 21:07:35
> > Objet: Keeping `new Integer` source compatibility with unnamed factory
> method
>
> > Hi All,
> > regarding the summary of the last EG meeting (2020-11-04) [1], I can see
> a
> > future me being frustrated with having to keep some dusty `javac` only
> > because of some _mature_ code generation tool that "optimizes" by using
> > `Integer::new`.
>
> You can pro-actively already create a pull request to ask this mature tool
> to use Integer::valueOf instead of Integer::new.
> But you're right that this is a real issue that will hinder the adoption
> of Valhalla.
>
> >
> > Is the "something special" to keep `Integer::new` source compatibility
> > adding an `Integer::new` (and similarly `Object::new`) unnamed factory
> > method?
> >
> > Sorry if I have missed any material discussions related to such an idea.
>
> In the Valhalla world, Integer is not a concrete class anymore but a
> subtype of int, exactly a subtype of int sees as a primitive object), that
> why you can not instantiate it directly anymore.
>
> So currently, we are weighting our option more than we are making decision
> on which backward compatible strategy we want to use, source backward
> compatible vs binary backward compatible.
>
> So we may still allow to use new Integer()/Integer::new in the source
> code, to allow any tools that generates Java code to still work but it
> means that the compiler will have to rewrite it to be Integer.valueOf,
> which can be very surprising,
> by example,
>   new Integer(2) == new Integer(2)
> will now return true.
>
> And in that case, we have several ways to do it. We can do that in the
> library, creating an unnamed factory method that calls Integer.valueOf() as
> you are suggesting or we can do it directly in the compiler by rewriting
> all occurrences of new Integer().
>
> >
> > Cheers,
> > Mateusz Romanowski
>
> regards,
> Rémi
>
> >
> > [1]
> >
> https://mail.openjdk.java.net/pipermail/valhalla-spec-experts/2020-November/001446.html
> > [2]
> >
> http://cr.openjdk.java.net/~dlsmith/special-methods/special-methods-20200210/specs/factory-methods-jvms.html#jvms-2.9.4
>