[hibernate-dev] Could we have a Hibernate 5.3 compatibility layer that includes the ORM 5.1 Hibernate Session class

[Sanne Grinovero]

On 10 February 2018 at 13:32, Scott Marlow <smarlow at redhat.com> wrote:
> Hi Sanne,
>
> Good question about what is public API exactly.  In WF modules, we have a
> way to mark the entire module as private.  That doesn't help here, and that
> would only be a "how" not answering "what extensions".
>
> Having said that, it would be interesting to know the "how" as well.
> Perhaps documentation could state what APIs in Hibernate 5.3 that could be
> expected to *not* be the same in 6.0+.  This wouldn't include new method
> signatures added but instead only existing method signatures removed.  As
> well as, existing methods that are going to be repurposed in some way.
>
> I am not sure if this is enough but we could explore this idea if it would
> give us enough flexibility to move from 5.3 to 6.0+, later.

That sounds like a workable idea. Users won't have to wait for 6, and
we can document our intentions.

We've previously used labels such as "experimental" on some APIs but
that hasn't been very effective.
To avoid misunderstandings, I'd prefer an explicit "opt in" for what
we consider stable API. So rather than listing what is *not*
supported, I propose we mark/annotate/document the methods we consider
stable. Would that work for WildFly's needs?

This way we'll likely be very conservative in 5.3 with what we
document to be "stable", providing essentially JPA and a few more
cherry picked essentials; we could then relax restrictions in smaller
steps while the new stuff matures.

I expect some cathegories of users will find our selection too
restrictive, but those more demanding are free to use ORM 5.1 for the
time being and let us know what they'd like to see supported in the
next round.
I also expect the majority of users to not care at all what we
document being stable or not, so everyone will have options ;)

Thanks,
Sanne


>
> Scott
>
> On Feb 8, 2018 5:09 PM, "Sanne Grinovero" <sanne at hibernate.org> wrote:
>
> Hi Scott,
>
> remember ORM 6.0 doesn't exist yet. Technically 5.3 doesn't exist yet
> either, but it's close.
>
> I'd start working to get 5.3 included so that we can address all
> integration issues, and additional unknowns caused by having multiple
> versions. If when we're all done fixing such problems 6.0 will be out,
> we'll re-evaluate the situation.
>
> In the meantime we'll check if 6.0 is going to be strickly compatible
> with 5.3: hard to answer on that in a definitive way when neither is
> ready. Unless Steve knows of some *public API* which definitely needs
> to break between 5.3 to 6.0? AFAIR last time we had such a
> conversation there was no clear agreement on what extensions should be
> considered public API *in the scope of WildFly's requirements*. Scott,
> do you have an update on that, or maybe refresh our memories?
>
> Thanks,
> Sanne
>
>
> On 8 February 2018 at 19:54, Scott Marlow <smarlow at redhat.com> wrote:
>>
>>
>> On 02/08/2018 01:50 PM, Steve Ebersole wrote:
>>>
>>> What is meant by "(JPA/native application) compatible with the Hibernate
>>> ORM versions that we include"?
>>
>>
>> It means that for a while, we can only upgrade WildFly to newer Hibernate
>> ORM versions that are compatible with one of the two Hibernate versions
>> that
>> we include with WildFly 12+.
>>
>>>
>>> Both 5.3 and 6.0 are JPA 2.2, but 5.1 is JPA 2.1.  Does that mean they
>>> violate that requirement?
>>
>>
>> I think the 5.3 versus 6.0 question, is more will applications written
>> against the (application level) Hibernate 5.3 ORM classes, work correctly
>> against 6.0.  For determining which classes exactly, are application level
>> classes, the Hibernate team should decide that.
>>
>> If the design for Hibernate ORM code changes, misses an application
>> compatibility issue that we miss, that cannot be avoided.  However, when
>> the
>> WildFly team looks at the latest/greatest version of Hibernate ORM to
>> include, we will face pressure against bringing in any newer ORM version
>> that requires application coding/configuration changes.
>>
>>
>>>
>>>
>>> On Thu, Feb 8, 2018 at 12:30 PM Scott Marlow <smarlow at redhat.com
>>> <mailto:smarlow at redhat.com>> wrote:
>>>
>>>
>>>
>>>     On 01/31/2018 10:49 AM, Steve Ebersole wrote:
>>>      > Not to mention, I'm really not even sure what this request
>>>     "means".  As
>>>      > we all understand 5.1 -> 5.2 unified
>>>     SessionFactory/EntityManagerFactory
>>>      > and Session/EntityManager, and that caused us to have to make
>>>     changes to
>>>      > certain method signatures - most notably `Session#getFlushMode`
>>>     was one
>>>      > of the problems.  Session defined that returning a FlushMode;
>>> however
>>>      > JPA also defined this same method, although poorly named IMO since
>>> it
>>>      > instead returns JPA's FlushModeType (so why the method is not
>>> called
>>>      > `#getFlushModeType` is beyond me.  Anyway the point is that there
>>>     is no
>>>      > way to rectify these - there is no way that we can define a
>>> contract
>>>      > that simultaneously conforms to both.
>>>      >
>>>      > As Sanne said, and as we all agreed during f2f, the best approach
>>>     is to
>>>      > have both versions available for use.
>>>
>>>     Which Hibernate ORM release would be best for the second version that
>>> we
>>>     include?  ORM 5.3 or 6.0?
>>>
>>>     Agreed that we will still include ORM 5.1, in WildFly.  For the
>>> second
>>>     ORM version that we include (whatever the version is), we have an
>>>     additional requirement now.  Future releases of WildFly need to be
>>>     (JPA/native application) compatible with the Hibernate ORM versions
>>> that
>>>     we include.
>>>
>>>     We will be releasing WildFly more frequently, and want users to be
>>> able
>>>     to able to keep up with our pace, as we release more often.
>>>
>>>     Scott
>>>
>>
>
>