Feedback on LazyConstants (formerly StableValues)

[Chen Liang]

Hi David, before we move on, I need to ask more about your use cases.


  1.
How do you ensure that a singleton is sharable, that all constructor arguments passed by all callers result in an equivalent instance?
  2.
Why would you say using the original StableValue is "more cluttered" than if (instance == null) initialize; return instance?

I don't think we can have a meaningful discussion if we can't know how you are sourcing the arguments to ensure the instances are equivalent.

Regards
________________________________
From: amber-dev <amber-dev-retn at openjdk.org> on behalf of david Grajales <david.1993grajales at gmail.com>
Sent: Wednesday, September 24, 2025 9:20 AM
To: Maurizio Cimadamore <maurizio.cimadamore at oracle.com>
Cc: Per-Ake Minborg <per-ake.minborg at oracle.com>; amber-dev <amber-dev at openjdk.org>
Subject: Re: Feedback on LazyConstants (formerly StableValues)

Hi Maurizio and Minborg. Thank you so much for the response. I will focus on explaining the part with the parameters for building objects. This scenario is not as unique at it may seems, I used a singleton because I thought it would be the easiest example but it's true it may not be the most representative. I will hold my thoughts until I test the new API. but I still would like to propose something. The supplier based factory requires to capture a reference to an external variable in case we need to set some parameters to create or compute the lazy constant, which can't hurt performance a little if used often.

var conf = "conf";
var foo2 = LazyCosntant.get(() -> {
    if (conf.isBlank()) {
         // do some validations
    }
    return new Foo(conf);
});

I think it would be worth considering adding a function based factory that accepts an object T (a configuration class to pass the parameters), and a function that accepts the parameter and returns the lazy computed value.

private class Bar{
   public final String confParam1;
   public final int confParam2;
   public Bar(String param1, int param2){
      confParam1 = param1; confParam2 = param2;
   }
}


var bar = new Bar("conf1", 42);
var foo2 = LazyValue.get(bar, s -> {
        if(/*Set some predicate for validation*/)  {
         // do something
    }
    return new Foo(s);
});

I think it's cleaner and safer (also a little more performant since the parameter can be inlined and not captured as an external element, and since deferred initialization is pretty much about squeezing performance it may be worth considering this). besides it may internally check for T not null.

Thank you so much and best regards.



El mié, 24 sept 2025 a la(s) 7:02 a.m., Maurizio Cimadamore (maurizio.cimadamore at oracle.com<mailto:maurizio.cimadamore at oracle.com>) escribió:

On 24/09/2025 11:38, Per-Ake Minborg wrote:
> I think the other examples you show (albeit I didn't fully get how
> they were supposed to work) would have issues regardless of whether
> there were language or library support for lazy computation

I'd like to amplify this point a little.

Your example shows the use of a singleton -- an object that is
constructed once, then stashed in a static final field.

However, the construction of the singleton (getInstance method) depends
on a parameter.

This means that, effectively, getInstance will capture whatever
parameter value was passed the first time it was constructed.

Now, there might be use cases for this, but such a use case would also
not be supported if using Kotlin's lazy, Scala's lazy val, or our
LazyConstant API (all of them are equivalent from an expressiveness
point of view).

So, your claim that

> this may indicate that a keyword or annotation-based solution could be
> a better fit.
Feels a bit off -- either the example you provided is not what you
really had in mind, or, when you say _keyword_ you mean something other
than a lazy-like keyword (but as Per explained, while there are some
more obscure keywords in other languages that might provide more
flexibility, the semantics associated with such keywords feels a bit
ad-hoc, and surely not something we would like to permanently bolt onto
the language).

Cheers
Maurizio


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/amber-dev/attachments/20250924/7df019e5/attachment.htm>