Re: [osgi-dev] SCR: ServiceInjection into Fields that are Optional

Mon, 28 Sep 2020 08:41:41 -0700

I kind of like the idea of Optional<T> for field injection. As Neil says, it is short hand for cardinality=ReferenceCardinality.OPTIONAL.
 
Since R8 supports, Java 8 as the base language level, we can now add support for Optional as a field (and constructor) injection type. Bnd can infer the actual service type from the generic T.
 
So
 
@Reference
Optional<Foo> foo;
 
would be a static optional reference to a Foo service, The foo field would be injected with either an empty Optional if there is no target Foo service, or a non-empty Optional with the bound Foo service. Like all static references, the foo field would not be changed by SCR during the life cycle of the component instance. This can also be used in constructor injection with a Optional<Foo> constructor parameter.
 
@Reference
volatile Optional<Foo> foo;
 
would be a dynamic optional reference to a Foo service. The foo field would be injected with either an empty Optional if there is no target Foo service, or a non-empty Optional with the bound Foo service. Like all dynamic references, the value of the foo field can be changed by SCR at any time during the life cycle of the component instance. So when the component wishes to use the field, it should copy the field's value into a local variable before inspection and use.
 
Optional<Foo> localFoo = foo;
if (localFoo.isPresent()) { do something with localFoo; }
 
I do not think Promise is a proper type to use here. Promise is about a future result and is a one-time latch on resolving the value. At the time SCR will inject the field, the state of the target service must be known. For dynamic references, SCR must replace the field as the bound service changes. So there is no real utility in using Promise here as you still would have know way to know there was a new Promise object to which you might want to attach a callback.
 
I will open an issue to add Optional support for field and constructor injection to DS 1.5 for R8.
 
--

BJ Hargrave
Senior Technical Staff Member, IBM // office: +1 386 848 1781
OSGi Fellow and CTO of the OSGi Alliance // mobile: +1 386 848 3788
hargr...@us.ibm.com
 
 
----- Original message -----
From: Mark Hoffmann via osgi-dev <osgi-dev@mail.osgi.org>
Sent by: osgi-dev-boun...@mail.osgi.org
To: Neil Bartlett <njbartl...@gmail.com>
Cc: OSGi Developer Mail List <osgi-dev@mail.osgi.org>
Subject: [EXTERNAL] Re: [osgi-dev] SCR: ServiceInjection into Fields that are Optional<Service>
Date: Wed, Sep 23, 2020 10:14
 
Hi Neil,
 
exactly I thought of a fluent API for a ServiceTracker.
 
Regards,
 
Mark Hoffmann 
M.A. Dipl.-Betriebswirt (FH) 
CEO/CTO 
 
Phone: +49 3641 384 910 0 
 
 
Data In Motion Consulting GmbH 
Kahlaische Strasse 4 
07745 Jena 
Germany 
 
Geschäftsführer/CEO 
Mark Hoffmann 
Jürgen Albert 
 
Jena HRB 513025 
Steuernummer 162/107/05779 
USt-Id DE310002614
 
 
-------- Ursprüngliche Nachricht --------
Von: Neil Bartlett <njbartl...@gmail.com>
Datum: 23.09.20 16:01 (GMT+01:00)
An: Mark Hoffmann <m.hoffm...@data-in-motion.biz>
Cc: OSGi Developer Mail List <osgi-dev@mail.osgi.org>
Betreff: Re: [osgi-dev] SCR: ServiceInjection into Fields that are Optional<Service>
 
Ah okay thank you. So you're not talking about the existing Promise class, rather a class that would keep track of the changing state of the service.
 
As I think you said, that's basically a ServiceTracker. However the existing ServiceTracker class has a lot of complexity and still uses ancient APIs like Dictionary. So the requirement could be to define a new interface ("ServiceHandle"?) that has methods like:
 
    interface ServiceHandle<S> {
        void ifPresent(Consumer<? super S>); // from Optional
        Stream<S> stream(); // from Optional
        Promise<S> whenResolved();
        Promise<Void> whenUnresolved();
    }
 
SCR could inject an instance of this interface into a component (component authors would not be able to instantiate their own instance).
 
It's an interesting idea, but I believe it would be better to keep it as a separate requirement from the "support Optional-typed field injection" requirement.
 
Neil
   
 
On Wed, 23 Sep 2020 at 14:24, Mark Hoffmann <m.hoffm...@data-in-motion.biz> wrote:
Hi Neil,
 
thus is why I said something like a Promise. I think in my first post this morning I mentioned e.g. the missing unresolve or reresolve.
 
On the other side, if we had such an special Object, we could also bring the feature of callbacks back to field injection.
 
Just to get you right, I am not against the idea, I just wanted to get a step further !? 🙂
 
 
Mark Hoffmann 
M.A. Dipl.-Betriebswirt (FH) 
CEO/CTO 
 
Phone: +49 3641 384 910 0 
 
 
Data In Motion Consulting GmbH 
Kahlaische Strasse 4 
07745 Jena 
Germany 
 
Geschäftsführer/CEO 
Mark Hoffmann 
Jürgen Albert 
 
Jena HRB 513025 
Steuernummer 162/107/05779 
USt-Id DE310002614
 
 
-------- Ursprüngliche Nachricht --------
Von: Neil Bartlett <njbartl...@gmail.com>
Datum: 23.09.20 13:14 (GMT+01:00)
An: Mark Hoffmann <m.hoffm...@data-in-motion.biz>
Cc: OSGi Developer Mail List <osgi-dev@mail.osgi.org>
Betreff: Re: [osgi-dev] SCR: ServiceInjection into Fields that are Optional<Service>
 
Can you clarify how you would expect a Promise to work?
 
It sounds a bit like forcing dynamic behaviour into a static reference, i.e. you get a static field of type Promise which may later resolve to the bound service. But this disconnects the lifecycle of the component from the availability of the service. The problem is that the service can change state many times between available and unavailable, whereas a Promise only ever has a single state change (from unresolved to resolved). That is, a Promise cannot revert from resolved to unresolved.
 
It seems to me that we already have a mechanism for tracking the dynamic state of a service with callbacks, and that is method injection. The bind and unbind methods can be called as many times as necessary. I don't see how Promises make this any cleaner.
 
The idea of explicitly supporting Optional fields does not have any lifecycle implications for the component. We simply change the representation of "unbound" from null to Optional.empty() and eliminate a whole class of potential NPE-throwing code patterns.
 
Neil
 
On Wed, 23 Sep 2020 at 12:01, Mark Hoffmann <m.hoffm...@data-in-motion.biz> wrote:
Hi,
 
Neill I see the point. As already mentioned I would rather see something like Promise instead fd an Optional, where I could stay with one instance.
 
In addition to that I would get onResolve out-of-the-box.
 
Regards,
 
Mark Hoffmann 
M.A. Dipl.-Betriebswirt (FH) 
CEO/CTO 
 
Phone: +49 3641 384 910 0 
 
 
Data In Motion Consulting GmbH 
Kahlaische Strasse 4 
07745 Jena 
Germany 
 
Geschäftsführer/CEO 
Mark Hoffmann 
Jürgen Albert 
 
Jena HRB 513025 
Steuernummer 162/107/05779 
USt-Id DE310002614
 
 
-------- Ursprüngliche Nachricht --------
Von: Neil Bartlett <njbartl...@gmail.com>
Datum: 23.09.20 11:50 (GMT+01:00)
An: Mark Hoffmann <m.hoffm...@data-in-motion.biz>, OSGi Developer Mail List <osgi-dev@mail.osgi.org>
Betreff: Re: [osgi-dev] SCR: ServiceInjection into Fields that are Optional<Service>
 
I like Stefan's suggestion and I disagree with some specific points made by Mark... responses line below.
 
On Wed, 23 Sep 2020 at 05:35, Mark Hoffmann via osgi-dev <osgi-dev@mail.osgi.org> wrote:
Hi Stefan,
 
I believe Optionals are not optimal for that.
 
If a service is removed, you would need a new empty optional instance. Optionals doesn't support to hold a state.
 
The state is in the component, not the Optional instance. References are static by default, so under Stefan's example the component would be destroyed if the field was bound to a service and that service instance became unregistered. The proposed change to the spec is that SCR in this case must reinitialize the field with Optional.empty() rather than with null. For a dynamic reference, the field would need to be volatile just as it is in R7. The only difference is that the field value would be replaced by Optional.empty() rather than null in the case where the reference is unbound.
 
There are two advantages that I can see. The first is brevity. When bnd sees a field of type Optional, it can infer a cardinality of 0..1, so we do not have to annotate with `cardinality=ReferenceCardinality.OPTIONAL`, which is more verbose. The second is a thread-safety issue for dynamic references. For example, suppose we have a dynamic optional reference in R7, i.e.:
 
    @Reference(cardinality=ReferenceCardinality.OPTIONAL, policy=ReferencePolicy.DYNAMIC)
    Foo foo;
 
When we want to use the field value we have to null check first, but the following code -- although very clear and intuitive -- is unsafe:
 
    if (foo != null) {
        foo.doSomething();
    }
 
It's unsafe because the value of the field can change between the null check and the invocation of the method. The following code patterns with a field of type Optional<Foo> are safe however:
 
    optFoo.ifPresent(Foo::doSomething)
 
or:
 
    Stream<String> barNames = optFoo.stream()
      .flatMap(foo -> foo.searchBars("*"))
      .map(Bar::toString);
 
They are safe because the value of the volatile field is only accessed once.
 
 
 
In addition to that you don't have callbacks where to get notified about adding, modifying or removing a service.
 
We don't have callbacks today with field injection in R7. If you want callbacks you need to use method injection.
 
 
 
I could imagine, to get an object similar to the promise injected, that supports the lifecycle callbacks, as well as resolving, unresolving or re-resolving as well as handling cardinality. 
 
A fluent API around the OSGi's ServiveTracker could be a solution. Injecting Optional for just a subset of the supported cases in DS would, from my perspective "pollute" the spec.
 
Maybe you can realize a custom solution using ServiceHooks?
 
I don't think that works because SCR and bnd would both have to explicitly understand and support fields of type Optional.
 
 
Regards,
 
Mark Hoffmann 
M.A. Dipl.-Betriebswirt (FH) 
CEO/CTO 
 
Phone: +49 3641 384 910 0 
 
 
Data In Motion Consulting GmbH 
Kahlaische Strasse 4 
07745 Jena 
Germany 
 
Geschäftsführer/CEO 
Mark Hoffmann 
Jürgen Albert 
 
Jena HRB 513025 
Steuernummer 162/107/05779 
USt-Id DE310002614
 
 
-------- Ursprüngliche Nachricht --------
Von: Stefan Bischof via osgi-dev <osgi-dev@mail.osgi.org>
Datum: 22.09.20 15:48 (GMT+01:00)
Betreff: [osgi-dev] SCR: ServiceInjection into Fields that are Optional<Service>
 

Hi,

I like it to use Optionals if it is possible that a field could be null.


In context of OSGi Services with SCR that means I have to handle it like this:

```

@Component
public class MyComponent
{

    Optional<Foo> oFoo = Optional.empty();

    @Reference(cardinality = ReferenceCardinality.OPTIONAL)
    void bindFoo(Foo foo)
    {
        oFoo = Optional.of(foo);
    }
}

```

What I really want to do is this:

```

@Component
public class MyComponent
{

    @Reference
    Optional<Foo> oFoo;

}

```


We have something like this in OSGi - CDI Integration Specification
https://youtu.be/7-UUJ4WkMsg?t=839

It would be nice to have this feature with the new version of the R8 DS Spec.

 

Regards

Stefan

_______________________________________________
OSGi Developer Mail List
osgi-dev@mail.osgi.org
https://mail.osgi.org/mailman/listinfo/osgi-dev
 

_______________________________________________
OSGi Developer Mail List
osgi-dev@mail.osgi.org
https://mail.osgi.org/mailman/listinfo/osgi-dev

Reply via email to