Error message when accessing non-public type

[Gunnar Morling]

Yes, having a facility like that would help this case.

--Gunnar


2016-06-22 22:31 GMT+02:00 Alex Buckley <alex.buckley at oracle.com>:

> Hi Gunnar,
>
> Thanks. Yes, I suspected you were using ServiceLoader::load to get a dummy
> instance of each custom constraint validator class (in general, there might
> be multiple modules providing a validator for @Min-on-int), and then
> obtaining each dummy's Class object and doing your own instantiations of
> validator classes thereafter.
>
> It wouldn't help if ServiceLoader simply returned the Class objects for
> service implementations, because you still wouldn't be able to
> newInstance() them without the export. You would need ServiceLoader to take
> one of those Class objects back (call it cImpl), along with the Class
> object for a service interface (call it cIntf), and then instantiate cImpl
> on your behalf if a) cIntf.isAssignableFrom(cImpl) and b) <<some
> caller-sensitive conditions>>. No more export needed.
>
> Alex
>
>
> On 6/21/2016 11:55 PM, Gunnar Morling wrote:
>
>> Hi Alex,
>>
>> Good question, I should have mentioned some more details on this.
>>
>> Hibernate Validator is using the service loader here to discover
>> constraint validator *types*, but then it itself is going to create (and
>> configure) instances of these types based on the annotation properties
>> set for specific usages of a constraint. I.e. there will be one instance
>> of a validator per constraint usage (the instance obtained from the
>> service loader will be dropped once its type has been examined).
>>
>> E.g. two instances of the constraint validator type for @Min on int
>> would be created here:
>>
>>      class Foo {
>>          @Min(1)
>>          int bar;
>>
>>          @Min(2)
>>          int baz;
>>      }
>>
>> One might argue that this is a bit mis-using the service loader
>> mechanism and we should rather detect some sort of constraint validator
>> factory in charge of creating actual constraint validator instances (in
>> which case we could take the instance returned from the service loader
>> as is).
>>
>> But then it's working good enough in its current form and is simpler on
>> implementors that way (who just need to provide/expose constraint
>> validators and nothing more). But it explains why that export is needed
>> under Jigsaw.
>>
>> Alternatively, an API for obtaining just the types of services (not
>> instances) might help with this case.
>>
>> --Gunnar
>>
>>