Skip to main content
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [qvto-dev] model extents for clone/deepClone 
Hi Ed, Christopher,

If we agreed with the following statements:
a) An extent specification makes sense only when objects are created.
b) A created object will always belong to a model extent.
c) Objects model extent can't change residence via model extent specification. They need to manually be assigned to the proper container (via containment references assignment or a library operation) 

I could recollect a joint Issue resolution towards this direction.
If you see reason about why any of those statements should not be true, please give me some lights with examples. 

More details in the TL;DR version

Regards,
Adolfo.
----------- TL;DR---------------
Yes, it looks like Ed missed one of the emails explaining why the issue 13103 resolution is not satisfying, IMO. Please have a look to it (title: Issue 13103: element creation was: [qvto-dev] QVT 1.2 RTF Ballot 3 preview 2). It also includes an revised resolution for the issue.
No, changing residence of an object which already has got a residence via this modelExtent specification is not good idea. IMO, it's hazards-full language decision. Actually, I'm inclined to always make an object have residence regardless when/how they are created. MappingParameter (section 8.2.1.16) goes in this direction: if the residence can't be inferred when it's not specified, an ambiguity error needs to be thrown. Actually the specification could be improved so this extent inference is statically computed (the extent is mandatory in the AS) rather than dynamically computed at runtime.
IMO, things get simple and coherent when we see the model extent specification whenever a object is CREATED/INSTANTIATED. In QVTo I conceive 4 ways to create objects: 

a) ObjectExp when the referredObject is null
b) In the implicit instantiation section of MappingOperations, when out mapping parameters are null after the init section is executed. 
c) Using InstantiationExp -> using the new keyword.
d) Using clone/deepClone library operations
QVTo specification says that you can explicitly especify the @modelExtent for a), b), c). Nothing is said about d). In the three cases are always related to "creating" objects, not when updating them. Updating objects could take place in in a) and b) and attaching/reattching to a different model extent should NOT take place.
I understand Ed's point since when you clone objects you will *usually* attach them to another object via a compose association (aka containment reference). However, Christopher case is valid and I don't think that QVTo provides means to do it: There is not to a Model::addElement(Element ) operation. Actually if it where pointless, we should also consider why is useful for InstantiationExps
Another point is that what if a Model::addElement(Element) should have been more convenient rather than the @modelExtent specification. The @modelExtent looks like a more handful mean to do the same. We don't neither want to break the specification to get rid of the @modelExtent specification in favor of a Model::addElement(Element) library operation.
So if it were my duty, with respect to the clone/deepClone the specification would change in the following way: - In Element::clone/deepClone library operation, clarify that the created element belong to the same model extent of object which is created from. 
- Add an Element::clone/deepClone(Model extent) operation.
This changes leads to the ideally situation in which in a m2m transformation every object will always belong to a model extent. 


On 05/02/2014 09:58, Christopher Gerking wrote:

Hi

Ok. This gives rise to residence changes, which is basically ok for me.

But then we need to consistently revise also 8.2.1.16, which mentions
“the potential created element” and therefore gives the impression that
the extent plays a role only on creation. The 8.2.1.16 revision proposed
by Adolfo would be invalid then, because it makes the very same
assumption that residence affects only the creation.

What about the extent inference rules? Should they apply only to mapping
parameters? Not in case of object expressions or clone/deepclone calls?
Should they apply at all?

Regards

Christopher

In 8.2.1.16,

In associations subsection:

Replace:

"The extent of the mapping parameter. If not provided, the extent is
inferred by inspecting the model types of the transformation. See the
inference rules below. Should be explicitly provided when there is an
ambiguity on the extent to own the potential created element
corresponding to this parameter."

*Von:*qvto-dev-bounces@xxxxxxxxxxx [mailto:qvto-dev-bounces@xxxxxxxxxxx]
*Im Auftrag von *Ed Willink
*Gesendet:* Mittwoch, 5. Februar 2014 10:17
*An:* QVTOML developer mailing list
*Betreff:* Re: [qvto-dev] model extents for clone/deepClone

HI

No. Read the proposed revised text for Issue 13103.

     Regards

         Ed

On 05/02/2014 09:02, Christopher Gerking wrote:

    As Adolfo pointed out, the extent of an ObjectExp applies only if
    the object is actually instantiated (not just updated, like in the
    cloning case).

    Would you agree?3.

    *Von:*qvto-dev-bounces@xxxxxxxxxxx
    <mailto:qvto-dev-bounces@xxxxxxxxxxx>[mailto:qvto-dev-bounces@xxxxxxxxxxx]
    *Im Auftrag von *Ed Willink
    *Gesendet:* Mittwoch, 5. Februar 2014 09:47
    *An:* QVTOML developer mailing list
    *Betreff:* Re: [qvto-dev] model extents for clone/deepClone

    Hi

    If the clone is to be root, then it probably occurs in main() which
    seems to have its own way of managing extents.

    Otherwise an ObjectExp may assign the clone to its extent.

         Regards

             Ed

    On 05/02/2014 08:40, Christopher Gerking wrote:

        Hi again

        What if the clone is meant to be the root of an extent? Then it
        does not participate in any containment.

        Therefore, I think we do need to attach a clone to an extent.
        Again, I find the current Eclipse QVTo solution very practical,
        which reuses the inference rules for mappings that do not
        specify an explicit extent.

        Regards

        Christopher

        *Von:*qvto-dev-bounces@xxxxxxxxxxx
        <mailto:qvto-dev-bounces@xxxxxxxxxxx>[mailto:qvto-dev-bounces@xxxxxxxxxxx]
        *Im Auftrag von *Ed Willink
        *Gesendet:* Dienstag, 4. Februar 2014 18:54
        *An:* QVTOML developer mailing list
        *Betreff:* Re: [qvto-dev] model extents for clone/deepClone

        Hi

        I see no point in a cloned object automatically belonging to any
        extent. The clone should become part of a Resource as soon as
        ity participates in a containment relationship.

        The clone can be explicitly extented using an ObjectExp.

        If the user neglects to provide an extent tough it's lost. If
        QVTo is kind, it could issue a warning, and possibly even put it
        in yet another Resource, or the trace model.

             Regards

                 Ed

        On 04/02/2014 17:12, Adolfo Sánchez-Barbudo Herrera wrote:

            Hi Christopher,

            To be honest I´m not up to date with your/EclipseQVTo
            clone/deepClone discussions, but the specification should
            clearly state what happens when cloning objects. I only see
            a couple of alternatives:

            1) They don´t belong to any modelExtent.

            2) They belong to the same modelExtent of the cloned object.

            and optionally:

            3) They can explicitly belong to a model extent, for
            instance with some library operations:

                  a) Element::clone/deepClone(Model extent) : Element

                  b) Model::clone/deepCloneElement(Element element): Element

            Cheers,

            Adolfo.






            _______________________________________________

            qvto-dev mailing list

            qvto-dev@xxxxxxxxxxx  <mailto:qvto-dev@xxxxxxxxxxx>

            https://dev.eclipse.org/mailman/listinfo/qvto-dev






            No virus found in this message.
            Checked by AVG - www.avg.com <http://www.avg.com>
            Version: 2014.0.4259 / Virus Database: 3684/7058 - Release
            Date: 02/03/14





        _______________________________________________

        qvto-dev mailing list

        qvto-dev@xxxxxxxxxxx  <mailto:qvto-dev@xxxxxxxxxxx>

        https://dev.eclipse.org/mailman/listinfo/qvto-dev





        No virus found in this message.
        Checked by AVG - www.avg.com <http://www.avg.com>
        Version: 2014.0.4259 / Virus Database: 3684/7062 - Release Date:
        02/04/14




    _______________________________________________

    qvto-dev mailing list

    qvto-dev@xxxxxxxxxxx  <mailto:qvto-dev@xxxxxxxxxxx>

    https://dev.eclipse.org/mailman/listinfo/qvto-dev




    No virus found in this message.
    Checked by AVG - www.avg.com <http://www.avg.com>
    Version: 2014.0.4259 / Virus Database: 3684/7062 - Release Date:
    02/04/14



_______________________________________________
qvto-dev mailing list
qvto-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/qvto-dev

Back to the top