You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@jackrabbit.apache.org by Jukka Zitting <ju...@gmail.com> on 2009/10/28 13:53:24 UTC

No repository map or set in o.a.j.core.RepositoryFactoryImpl

Hi,

As discussed in JCR-2360, I was looking at pushing the repository URI
support down to implementation-specific repository factories.

While doing so, I ran into the REPOSITORY_INSTANCES map and
ownRepositories sets in o.a.j.core.RepositoryFactoryImpl. Do we need
them, or would it be better to not keep track of the repository
instances across the factory instances.

The current behaviour is designed to make the following code pass:

    Repository a = JcrUtils.getRepository("file://path/to/repository");
    Repository b = JcrUtils.getRepository("file://path/to/repository");
    assertTrue(a == b);

Do we need this, or would it be better if we just instantiated a new
repository for each getRepository call and let the latest repository
access fail because the repository has already been started. I don't
think too many clients would be adversely affected by this change, as
such use of an embedded repository is typically only needed in cases
where you don't have multiple clients accessing the repository.

The benefit of not keeping the static REPOSITORY_INSTANCES map is that
there'd be no unexpected references to the repository instances and
thus they'd get normally garbage collected when no longer used. This
isn't too much of a problem now given the way TransientRepository
works, but it might prove troublesome later on.

Also, instead of the ownRepositories set used to track repository
instances created by a factory, we could just keep a single repository
reference in the factory instance. Any subsequent getRepository()
calls could simply throw an InvalidStateException, as the standard
access mechanisms documented for RepositoryFactory all call for a new
factory instance to be created for each getRepository() call.

These changes would simplify the RepositoryFactoryImpl implementation
and prevent potential memory issues in the future.

BR,

Jukka Zitting

Re: No repository map or set in o.a.j.core.RepositoryFactoryImpl

Posted by Thomas Müller <th...@day.com>.
Hi,

Remove REPOSITORY_INSTANCES: I think that's generally a good idea.

However it would be strange if a repository factory can only create
one repository. But instead of an ownRepositories we could add a 'back
reference' from the repository to the factory that created it.

Regards,
Thomas

On Wed, Oct 28, 2009 at 1:53 PM, Jukka Zitting <ju...@gmail.com> wrote:
> Hi,
>
> As discussed in JCR-2360, I was looking at pushing the repository URI
> support down to implementation-specific repository factories.
>
> While doing so, I ran into the REPOSITORY_INSTANCES map and
> ownRepositories sets in o.a.j.core.RepositoryFactoryImpl. Do we need
> them, or would it be better to not keep track of the repository
> instances across the factory instances.
>
> The current behaviour is designed to make the following code pass:
>
>    Repository a = JcrUtils.getRepository("file://path/to/repository");
>    Repository b = JcrUtils.getRepository("file://path/to/repository");
>    assertTrue(a == b);
>
> Do we need this, or would it be better if we just instantiated a new
> repository for each getRepository call and let the latest repository
> access fail because the repository has already been started. I don't
> think too many clients would be adversely affected by this change, as
> such use of an embedded repository is typically only needed in cases
> where you don't have multiple clients accessing the repository.
>
> The benefit of not keeping the static REPOSITORY_INSTANCES map is that
> there'd be no unexpected references to the repository instances and
> thus they'd get normally garbage collected when no longer used. This
> isn't too much of a problem now given the way TransientRepository
> works, but it might prove troublesome later on.
>
> Also, instead of the ownRepositories set used to track repository
> instances created by a factory, we could just keep a single repository
> reference in the factory instance. Any subsequent getRepository()
> calls could simply throw an InvalidStateException, as the standard
> access mechanisms documented for RepositoryFactory all call for a new
> factory instance to be created for each getRepository() call.
>
> These changes would simplify the RepositoryFactoryImpl implementation
> and prevent potential memory issues in the future.
>
> BR,
>
> Jukka Zitting
>