|
Hi Blaise,
Yes, I think that if a type were created through
JavaHelper.define() were given to XSDHelper.save(), then somehow the
annotation must be translated to a restriction.
Again, my motivation is mainly to make the metadata model
for types defined through java as consistent as possible with the metadata model
for types defined through XSD.
Fail fast exceptions are, in my opinion, a good thing,
especially for restictions like string length. But, like I said, whether
we do validation is a seperate issue than the question of whether the
restrictions can come from java or only from schema.
Have you thought about @SchemaInfo?
Ron
Hi Ron,
As of yet I'm not proposing anything, I'm just trying
to get a feel of why we're thinking of adding facets. Is your intent that
they be added solely as metadata? You mention that they are not
necessarily for schema generation, but would they be used
here?
-Blaise
Barack, Ron wrote:
7C3EF93EEBC6EB4A8B4470853DE86566707059@dewdfe18.wdf.sap.corp
type="cite">
Hi Blaise,
I see the facets mostly as metadata, for clients to use
or ignore. Not necessarily for schema generation, but also for
generating UIs and portlets. Originally, I had the goal to make the
types that could be generated from a java class just as rich as the types that
could be generated from XSD. In this proposal, I've move significantly
back from the goal, especially when it comes to registered dataTypes, but I
think, where possible, we should try to achieve a similar level of
expressiveness.
The question of if and when we do validatation is a bit
orthogonal, we would want exactly the same behavior for types defined through
JavaHelper.define as we have for XSDHelper.load. In our implementation
we do fail fast validation, but not of things like minOccurs (@Size(min=))
since they may be invalid only transiently. We also have some exception
handlers so that users may "catch" errors themselves. Before adding such
a construct to SDO, I guess we'd have to agree that fail fast should be the
default standard behavior. Are you proposing this? Again, since
the behavior would also apply to types defined by XSDHelper.define() I see
this as an orthogonal issue (to whether we have java annotations and what they
are).
I agree that we should learn from JAXB experience
and avoid on-demand validation.
I'm not sure I get the point of your last
paragraph... Are you implying that facets are too un-java-like to fit
comfortably in as Java annotations? Would you rather leave
them out of the proposal?
Ron
Hi
Ron,
What role do you see facets playing?
- Fail fast validation (i.e. set a string that is too long or add beyond
the max size of a list, then immediately throw an exception)?
- On demand validation (create a validator class that you can use to
confirm the object is valid)?
- More metadata for better schema generation?
In JAXB we found
more often than not the XML schemas users were using was more of a rough
guideline of how the XML should look as opposed to the exact
specifications. Where validation could occur in JAXB the spec group
added plenty of validation event handlers where users could programmatically
choose to ignore them. If we supported fail fast validation in
SDO, we would also need the mechanisms to allow users to override this
behaviour.
JAXB had support for on demand validation, but it was
deprecated in JAXB 2.0, ultimately all the participants agreed that nobody was
using this feature. In the almost 3 years since the release of JAXB 2, I
haven't had a customer ask for this functionality back.
In terms of
schema generation (from a JAXB perspective), Java users have different
expectations of XML schema than XML users do. JAXB Java users want an
XML schema they can use with Web Services, they rely on Java typing to
constrain there model. They choose int/long/float to constrain their
numeric types instead of relying on
maxInclusive/minInclusive.
-Blaise
Barack, Ron wrote:
7C3EF93EEBC6EB4A8B4470853DE86566706F9D@dewdfe18.wdf.sap.corp
type="cite">
Hi Blaise,
From Section
1.3
I agree, interfaces are the standard java
construct for static SDOs, and I think any clients that care about
portability have to use them only. The mapping from SDO
non-datatype Type to interface is established at least since SDO 2, and I
did not intentionally weaken this. In this paragraph, I wanted to say
that we are not standardizing that java interfaces/DynamicProxy always be
used, eg, clients should not rely on there being a client proxy
implemention. This allows implementation to chose other means of
implementing the interface (eg, a generated implementation class), and opens
the doors to other types of static SDOs, eg, POJOs or the generated class
itself.
Section 1.7.4
- @SdoProperty
My intention
was that @SdoProperty would only be necessary if the default mapping needed
to be modified (eg, to set the containment behavior). We want
unannotated interface to be usable, after all. On the other hand, I
think @SdoTransient would only make sense if we were dealing with an
annotated (POJO) class. On a Java interface (that is being implemented
using a DynamicProxy, say), it wouldn't make any sense. And since
interfaces are our only standard for static SDOs, having something like
@SdoTransient seems like something that should be left to be a vendor
extension. On the other hand, if we standardize bytecode weaving
as something that must be supported, then you are right, we need
@SdoTransient.
Section 1.7.6 -
@SdoFacets
Using JSR-303
is a great idea, and I will try to update the proposal accordingly. I
cases where the JSR does not provide an annotation that maps 100%
to an XSD restriction, how do you see proceeding? Should we
define our own annotatins, eg, commonj.sdo.validation.Pattern,
commonj.sdo.validation.MinExclusive? What about annotations by the JST
that do not match XSD restrictions, like @NotNull.
Really, the
fundamental problem here is that we don't have any standard SDO model
for restrictions/facets. I our implementation we have a bunch of
open content properties used for this purpose. Do you agree we
should standardize this?
Best
Regards,
Ron
Hi Ron,
Below
are some initial comments, more to follow.
From Section
1.3
"The implementation of the
returned DataObject MAY be a dynamic proxy, the result of byte code
enhancement or weaving of a POJO JavaBean or a class that was generated
during the application’s build process." Thinking
ahead to when the HelperContext creation issue is solved, and users can
freely switch between SDO implementations won't the only portable approaches
be to use only interfaces and dynamic proxy objects, or produce impl classes
using only byte codes? I'm not saying this is bad, only that this
appears to be what needs to happen.
Section 1.7.3 -
@SchemaInfo
Still working on comments for this one.
Section
1.7.4 - @SdoProperty
Does an accessor need an @SdoProperty annotation
to be included as a property? If not can we provide an annotation like
@SdoTransient to prevent some accessors from being converted to
properties?
Section 1.7.6 - @SdoFacets
Do we need to define
our own annotations or can we leverage JSR-303 Bean
Validation?
-Blaise
Barack, Ron wrote:
7C3EF93EEBC6EB4A8B4470853DE865666C77F9@dewdfe18.wdf.sap.corp
type="cite">Hi Frank,
Thanks for bringing this up, since I somehow left it up of my "main points" presentation.
There may need to be wording changes in the proposal, but my intention was:
SDO URI -> Java Package: Use JAX-B name-mangling.
Java Package -> SDO URI: uri = package.getName().
The disadvantage of the first rule is that when generating POJOs using JAXB and static SDOs from the same XSD, the default behavior will be to overwrite the classes. On the other hand, it save us from having to come up with a new algorithm.
I looked through the JAXB spec and I didn't find any behavior at all for describing the other direction. I don't like the idea of using the "" namespace. An important use case for us is going from unannotated classes to a reasonable SDO model. Using the "" package would mean that the class names have to be unique (over all packages), which I think is an unreasonable restriction.
The problem with the rule in my proposal is that the URI is not a legal URI in XML. This didn't really bother me, since it's still a legal SDO URI. The WebBeans approach sounds like a reasonable improvement on the proposal I sent out.
Ron
-----Ursprüngliche Nachricht-----
Von: Frank Budinsky [mailto:frankb@ca.ibm.com]
Gesendet: Dienstag, 10. Februar 2009 21:02
An: sdo@lists.oasis-open.org
Betreff: Re: [sdo] Proposal for Static SDOs
Hi Ron,
Regarding the package name to/from URI mapping, your document says they're
the same value, but that won't work. For example: "http://a/b/c" is not a
valid package name. I thought we were thinking about using the JAX-B
mapping rule for URI -> package name. In the other direction (package name
-> URI) JAX-B just uses "" for the URI by default (if there is no
annotation). Is that what we want to do as well? I heard that the WebBeans
spec is proposing a Java to XML namespace mapping of the following form: "
urn:java:<package qualified Java Class>". Seems kind of strange, but maybe
it's a good idea to make SDO3 and WebBeans behave in the same way.
Frank
"Barack, Ron" <ron.barack@sap.com>
02/09/2009 06:52 AM
To
<sdo@lists.oasis-open.org>
cc
Subject
[sdo] Proposal for Static SDOs
<<Proposal for Static SDOs.doc>> <<Static SDO.ppt>>
Hi Everyone,
Here is the proposed wording for Chapter 3 of the Java Spec, and a PPT
describing what I think are the highlights and main discussion points of
the proposal. Comments of course welcome. Hopefully, everyone will have
time to review this before next weeks virtual F2F.
Ron [attachment "Proposal for Static SDOs.doc" deleted by Frank
Budinsky/Toronto/IBM] [attachment "Static SDO.ppt" deleted by Frank
Budinsky/Toronto/IBM]
---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail. Follow this link to all your TCs in OASIS at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail. Follow this link to all your TCs in OASIS at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail. Follow this link to all your TCs in OASIS at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
|