[bv-dev] Do we need ConstraintValidatorFactory#releaseInstance()?

[Gunnar Morling]

Hi all,

The latest RI version [1] also implements the spec changes regarding the
integration of BV and CDI.

During implementation Hardy, Emmanuel and I got into a discussion regarding
the lifecycle of CDI-managed objects [2] which I'd like to continue.

The question is whose responsibility is it to release involved objects such
as message interpolator, traversable resolver or constraint validators.
Generally we think objects should be released by that party/code that
created them.

This means:

1) objects created by the user and passed via bootstrap API: to be released
by user
2) objects created by BV implementation (e.g. internal caches): to be
released by BV implementation
3) objects passed by CDI integration layer (e.g. CDI-enabled message
interpolator etc. based on BootstrapConfiguration): to be released by
integration layer

I think 1) is not in our focus, let the user deal with it. For 2) we have
ValidatorFactory#close(). Interesting is 3). A CDI portable extension could
make sure that its created managed beans are properly released using CDI's
pre-destroy hook [3].

But that way a CDI-enabled ConstraintValidatorFactory could also easily
release all the constraint validators it created:

CdiConstraintValidatorFactory implements ConstraintValidatorFactory {

  public <T extends ConstraintValidator<?, ?>> T getInstance(Class<T> key) {
    //create validators. keep track of them.
  }

  @PreDestroy
  public cleanUpValidators() {
    //release created validators
  }
}

That is, we'd again adhere to the principle that whoever creates objects
has to release them. When following this approach, we actually wouldn't
need ConstraintValidatorFactory#releaseInstance() anymore. I think this
would be a good thing as it IMO simplifies the contract between the BV
provider and code integrating with it (such as a CDI portable extension)
and clarifies responsibilities.

In the spec we would describe that a CDI-enabled CVF has to release its
validators when it itself is going to be destructed and for
ValidatorFactory#close(), that a BV provider has to release its internal
state but no objects passed to it.

One thing not possible with that approach is to release CV instances before
the application is shut down. The spec currently says that
releaseInstance() is "typically" invoked when the ValidatorFactory is
closed, but AFAICS it leaves room for releasing constraint validators also
in between. OTOH I wouldn't know when/how a BV provider would decide to do
that.

Any thoughts?

--Gunnar

[1] http://in.relation.to/Bloggers/HibernateValidator500Alpha2And431Final
[2] https://hibernate.onjira.com/browse/BVAL-338
[3] http://docs.oracle.com/javaee/6/tutorial/doc/gmgkd.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/beanvalidation-dev/attachments/20121203/9cae70bf/attachment.html