"Provides" and "with" type relationships

[Alex Buckley]

There are also some improvements to make to the ServiceLoader API spec, 
and at that time we'll lock in alignment of API spec and JLS. 
(Consistent terminology; duplication of rules only where essential; 
appropriate "call outs" to module system concepts such as service binding).

Alex

On 3/17/2016 11:19 AM, Jonathan Gibbons wrote:
> Konstantin,
>
> The compiler checks that the service implementation implements the
> service type.
> I have confidence that when the final specification is published, there
> will be suitable assertions (either explicit or implicit) to back up
> that check.
>
> -- Jon
>
> On 03/17/2016 06:02 AM, Konstantin Barzilovich wrote:
>>
>> Thanks.
>> It is described in API spec in detail.
>> But from compiler point of view, it is allowed to use implementation
>> without inheritance.
>> It would be grate to add some of these statements to compiler spec, if
>> it is possible.
>>
>> Thanks,
>> Konstantin.
>>
>>> Yes. 'uses' and 'provides' are nothing more than static declarations
>>> that configure java.util.ServiceLoader, so all questions can be
>>> resolved by looking at the ServiceLoader spec:
>>> http://download.java.net/java/jigsaw/docs/api/java/util/ServiceLoader.html
>>>
>>>
>>> Alex
>>>
>>> On 3/16/2016 10:24 AM, Konstantin Barzilovich wrote:
>>>> Sorry, if this question was asked before.
>>>> Does service implementation need to inherit service interface?
>>>>
>>>> Thanks,
>>>> Konstantin.
>>>>
>>>>> // Ignore last mail (mail client did a surprising thing)
>>>>>
>>>>> A 'provides' clause specifies two things: a service interface and a
>>>>> service implementation. Using those terms helps to avoid confusion.
>>>>>
>>>>> A service interface does not have to be an interface; it can be an
>>>>> abstract class or even (not recommended) a concrete class.
>>>>>
>>>>> A service implementation must not be an interface, or an abstract
>>>>> class; it must be a concrete class.
>>>>>
>>>>> Therefore, it's legal (but not recommended) for a concrete class to be
>>>>> specified as both service interface and service implementation. It's
>>>>> illegal for an interface (or abstract class) to be specified as both
>>>>> service interface and service implementation. JCK will be writing
>>>>> tests for edge cases like this.
>>>>>
>>>>> Alex
>>>>>
>>>>> On 3/15/2016 12:39 PM, Paul Benedict wrote:
>>>>>> Thanks for your response Alex. If I am understanding you correctly,
>>>>>> "provides" is "not constrained to be an interface" because it can
>>>>>> be "a
>>>>>> single interface or abstract class". So shouldn't my concrete
>>>>>> class for
>>>>>> "provides" be rejected by the compiler? And is it okay that both
>>>>>> types
>>>>>> were identical?
>>>>>>
>>>>>> Cheers,
>>>>>> Paul
>>>>>>
>>>>>> On Tue, Mar 15, 2016 at 2:26 PM, Alex Buckley
>>>>>> <alex.buckley at oracle.com
>>>>>> <mailto:alex.buckley at oracle.com>> wrote:
>>>>>>
>>>>>>     The first operand to 'provides' (the "service interface") is not
>>>>>>     constrained to be an interface by "Modules in the Java
>>>>>> Language and
>>>>>>     JVM". This is because the spec of j.u.ServiceLoader ("a
>>>>>> service is
>>>>>>     represented by a single type, that is, a single interface or
>>>>>>     abstract class").
>>>>>>
>>>>>>     The second operand to 'provides' (the "service
>>>>>> implementation") is
>>>>>>     constrained not to be an interface or an abstract class by
>>>>>> "Modules
>>>>>>     in the Java Language and JVM". This is also because of the
>>>>>> spec of
>>>>>>     j.u.ServiceLoader ("provider classes must have a zero-argument
>>>>>>     constructor so that they can be instantiated during loading").
>>>>>>
>>>>>>     Bear in mind that the JCK team can easily set up abstract test
>>>>>> cases
>>>>>>     like this. What they can't do is check whether YOUR
>>>>>> application runs
>>>>>>     on JDK-9-with-Jigsaw, or whether arbitrary JARs on YOUR classpath
>>>>>>     work as automatic modules.
>>>>>>
>>>>>>     Alex
>>>>>>
>>>>>>
>>>>>>     On 3/15/2016 12:07 PM, Paul Benedict wrote:
>>>>>>
>>>>>>         module z {
>>>>>>               exports z;
>>>>>>               provides z.Main with z.Main;
>>>>>>         }
>>>>>>
>>>>>>         The SOTM says "Service-provider declarations can be further
>>>>>>         interpreted to
>>>>>>         ensure that providers (e.g., com.mysql.jdbc.Driver)
>>>>>> actually do
>>>>>>         implement
>>>>>>         their declared service interfaces" (section 4, para. 8).
>>>>>>
>>>>>>         I see javac checking that they are related types, but
>>>>>> javac is
>>>>>>         not checking
>>>>>>         that "provides" is an interface type. That is what I was
>>>>>>         expecting based on
>>>>>>         the reading material.
>>>>>>
>>>>>>         The other unexpected outcome was that provides/with allows
>>>>>> the
>>>>>>         identical
>>>>>>         type. I don't know if that's intended, but please advise.
>>>>>>
>>>>>>         PS: I did go through the open tickets this time (thanks Alan)
>>>>>>         and do not
>>>>>>         see any similar reports. If I missed it, I apologize; just
>>>>>>         trying not to
>>>>>>         waste your time by reporting a duplicate.
>>>>>>
>>>>>>         Cheers,
>>>>>>         Paul
>>>>>>
>>>>>>
>>>>
>>>>
>>
>>
>