I am added my remarks in the text.
Daniel W. Davis
[email protected]
Asger Blekinge-Rasmussen wrote:
Hi
I am not sure I understood your post entirely. I have inlined some
questions on the parts I find unclear.
On Mon, 2009-07-06 at 17:15 +0200, Daniel Davis wrote:
I am concerned that the Fedora persistence/service model and the
global information model are being conflated. A public API is needed
to facilitate persistence operations but it is easy in Fedora to
overlap the persistence/service model with the public information
model which is intended to suitable for use in graphs and in the Web
architecture. Extending Fedora with writable service methods may make
the separation more clear but for now it is easy to conflate the
global information model and the persistence/service model.
Indeed!
It is not helpful that there are two ways to access content in
Fedora, datastream disseminations and DO Service Methods (aka
Operations)(formerly known as disseminators). Unfortunately,
datastream disseminators are easy to understand and use while service
methods are much harder and most applications avoid them.
Indeed.
Fedora must provide a clear means to encapsulate the content or
Fedora will have problems evolving. Technically, datastream
disseminations don't break encapsulation but having them facilitates
thinking in terms of concrete datastreams instead of abstract stable
resources.
Indeed.
If you make statements about the datastream itself you may step over
the line since the internal representation will change given enough
time and may break the truthfulness of statements made in global
models if internal information is exposed.
This is the old conflict between interfaces and implementations.
Agreed but it is useful to have the discussion in terms of Fedora
because Fedora is not a programming language, or a relational database,
Web server, triple-store (...) though there are conceptual overlaps with
other aspects of the other computing paradigms, and now with semantic
technologies (and the Semantic Web) the lines are even more blurred. I
hope these discussions will help us think clearly regarding where Fedora
needs to go and I think your work represents a very important part.
Internal use of RDF to support the encapsulation is a wonderful
generic approach to support encapsulation. But statements in the
external (global) model must only be about facts which should be known
externally (globally) and for a number of reasons, particularly
preservation, should be relatively stable.
Indeed.
Externally, datastreams should likely not be objects and it would be
best if external models avoid statements about concrete datastreams
(and likely datastream disseminations) going forward. Statements
about the internal structure of the DO should not show up in external
models. External models should ignore the existence of datastreams
and assert statements about the DOs, services and methods---you can
get the same functionality without making statements about internal
persistence artifacts that must change over time if Fedora is to
evolve.
I agree in the "should" statements. External models "should" not depend
on the internal representation. I will say, however, that currently this
"should" is not possible. To much is directly linked to the internal
representation, that not depending on it is achievable.
I have a general problem with the Fedora API, which is perhaps
understandably, given that I am the guy with the EnhancedContentModels.
The API is object centric, not class centric.
You can get the list of methods/services on an object, but there is no
easy or even hard way to get the list of services implied by a content
model.
As such, you have individual objects, decorated with services, not
classes of objects.
So, the one method that I would really like to see would, on a content
model, give me the list of services that the subscribing data objects
would have.
Fedora needs to evolve. Let us (the community) debate extensions of
APIs or replacement APIs like MediaShelf's contribution of the REST
API. Fedora is moving to a more community-driven model managed with new
committers being added from outside the Duraspace/Fedora Commons
organization. Please note that I am slowly moving out of the Duraspace
organization so I think of myself as moving into the application
community. The Fedora Repo can have many APIs some which are improved
capabilities and some which are less capable than but support widely
used standards. For example, there are projects underway to provide
WebDAV and JCR interfaces.
I agree with the sense of your comments and would like to think about
how it could be implemented. As to the approach for your ideas, I think
keeping the number of operations in the repository APIs to a minimum is
an important design principle (the classic hourglass design). The
methods you suggest could be implemented consistently on all CModel
objects conceptually like disseminators. Currently the Fedora Repo has
a limited ability easily or optimally add the functionality associated
with those methods but it can be done. Improvements with the modularity
especially if Fedora can be moved to an OSGi platform will make adding
custom functionality easier. One big problem is that there is no
implemention on the new REST API for getDissemination which is needed
for an object-centric, service-oriented approach. If I had my way, we
would eventually deprecate the datastream disseminations and use the
service (dissemination) approach because it can provide a uniform
interface for all dissemination at the cost of a slightly longer URI/URL.
If mapping to the Web architecture, the DO-service-method-parameters
should preferably be a stable URI which also can be often be used in
the Fedora architecture to cause the streaming of a serialized
representation. It would be desirable that external URIs refer to
some essential characteristic of the DO and its content that will
always be true even if the concrete implementation of the object
changes.
Indeed.
Internal models should be free to make statements about the internal
structure of the Fedora object and, I think but am not sure, can use
statements which are derived from external views of objects since such
statements are supposed to be stable.
Indeed.
However, there needs to be a separation of concerns between the
external (global) and internal views. Care should be taken about the
visibility of internal statements. They cannot be held to be globally
true. Internal information can be presented to the external model
through the use of URIs established through the CMA using
DO-service-method-parameters exercising care that there is an
abstraction placed between the internal model and the external model.
This enables statements made to global URIs to be globally true.
I am not sure I understand what you are saying here.
I am suggesting that we enable accessing virtually all the information
about the Fedora DOs through services (disseminations) on objects
(including specializing CModel objects). Also, it is feasible to design
services which write to Fedora DOs or cause side-effects. There are a
few places where this is not an optimal approach but I think it should
be a primary way of interacting with the Fedora Repo.
There still needs to be methods that permit access to the internal
structures of the DO so that "privileged" applications can create the
concrete persisted internal content. These operations need to know
about datastreams.
And, now, for the first time, do I understand what caused the separation
between API-A and API-M.
I am not sure we know enough to create APIs which absolutely and
clearly separate the external and internal models. I think that we
need to keep the separation of the models clear in our minds and
exercise care when we use them and in the evolution of Fedora's
design.
I think it might be time to rethink the API and the evolution in Fedora
based on this separation. I am not aware if there is design guidelines
in regards to the separation.
I agree. There is material but it is scattered and finding the entry
point is hard. Now there is a good Wiki and tracker we can make this
information more accessible. It would be great because we need to
articulate the design patterns if we want consistent evolution of
Fedora. Fedora is very pattern driven which, as you know, is a major
theme in software development these days we can apply. I like to think
of you work as an important step toward a "Model-driven Content
Architecture".
In particular, I think we need to be careful when we extract this
information into a triple-store where it is easy to combine/conflate
the two models and inadvertently mix statements whose truthfulness is
long term with transient implementation-specific information.
So, we should hold on integrating RELS-INT?
I am just adding my comments to the debate and posing questions. The
Fedora developers have been very careful in introducing RDF because its
technology was immature, there was insufficient open source support
libraries, and performance was low. First and foremost Fedora needed to
be a trustworthy platform for the creation of repositories and a degree
of being cautious is important.
Things have changed regarding RDF. It is probably practical now to
describe the internal relationships/structure of a Fedora DO using RDF
both as a class (model) and for instances. Likely some of those
internal relationships need to reference external entities (what are the
rules).
A classic example is versioning which is currently done using
hierarchical XML. But what happens if we have a DO containing Computer
Aided Design information which takes five datastreams? After a future
format migration the CAD information is contained within one datastream
(keeping the old version around because that is a policy of your
institution). RDF can easily represent that but doing it in the current
versioning schema is difficult. Lets not even start on representing
email collections.
I will try to add more pointed comments about RELS-INT in a separate
note the review needs all of us and I was hoping to provide some
background information I felt relevent to the review. Ultimately the
committers will decide but I am glad that Steve stepped up to take this
on. RELS-EXT and RELS-INT has been a tough long term design debate
which I don't think is over yet.
By the way, RELS_INT could be renamed though likely RELS-EXT needs to
stay the same for a while to avoid causing breakage.
Regards
Daniel W. Davis
http://fedora-commons.org
[email protected]
(607) 255-6090 (Office)
Asger Blekinge-Rasmussen wrote:
Hi Frank
Thanks for the reply.
Yes, you definitely nailed down the the missing points.
RELS-INT and RELS-EXT are misnamed, for the very reason you wrote. No
contest there.
About the RELS-EXT relations to datastreams in the object, that was a
hack.
A fedora object has some relation, fedora-view:disseminates I think, to
each datastream belonging to this object. Since this is the same
relation to every datastream, it is not possible to define a OWL
allValuesFrom restriction. In fact it is possible, but it has the effect
of demanding that all datastreams in the object is of the specified
class. Similarly, cardinality on that relation can only specify the
total number of datastreams.
I got around this by making my own relation (in RELS-EXT) to the
datastreams in the same object, but as you point out, these relations
could go to datastreams in another object.
Anyway, the introduction of RELS-INT does bring the current object
serialisation (foxml) into question. A datastream object conceptually
contains
A ID
Object properties (in RELS-INT)
Content (In the datastream proper)
Versioning (In the datastream proper)
Audit trail (in the AUDIT datastream in the Object)
So a datastream object are serialised into three datastreams, it self,
RELS-INT and AUDIT. And the fedora object then gets a relation to this
object.
To accomadate the new conceptual structure, it would probably be simpler
to serialise each datastream to it's own xml file, and make links from
the fedora object to each datastream it "contains".
The problem with this approach is that the traditional Fedora objects
will just become a collection of datastreams, and properties about this
collection, and not data in itself. This could easily be modelled with a
datastream object, and thus we have come full circle. Objects will in
effect reduced to having just one datastream.
This idea is starting to scare me somewhat....
Regards
On Mon, 2009-07-06 at 12:14 +0200, Schwichtenberg, Frank wrote:
Hi Asger,
I absolutely agree with you. That seems to be the logical enhancement of
Enhanced Content Models. :-)
I just wonder if the idea of RELS-EXT and RELS-INT holds. So, you are right
pointing out datastreams are entities (or objects), now. They have URIs and it
is possible to make statements in RELS-INT with datastreams of other Fedora
objects as object(-of-the-statement). So far your idea to enhance Enhanced
Content Models, which is great I think.
My criticism on "the idea of RELS-EXT and RELS-INT" would be: One can refer datastreams extern to
the Fedora object the RELS-INT belongs to. And it is possible to refer entire Fedora objects from RELS-INT.
Obviously it is possible to refer datastreams from RELS-EXT, also such of the Fedora object the RELS-EXT
belongs to. So "EXT" and "INT" seems to be out-dated. The difference between RELS-EXT and
RELS-INT has nothing to do with relations to external or internal entities. But with making statements about
the object (RELS-EXT) or about parts of the object (RELS-INT). So datastream URIs in statements (both in
RELS-EXT and RELS-INT) bring in possible complexity.
I don't want to say that's bad; just thoughts. Maybe that is something people
are waiting for. And the possibility to specify datastream cardinality (maybe
min and max) is great.
Maybe that just brings us back to the question, why not just allow datastreams
of RDF/XML content which are automatically get propagated to the resource index.
Regards, Frank
-----Ursprüngliche Nachricht-----
Von: Asger Blekinge-Rasmussen [mailto:[email protected]]
Gesendet: Freitag, 3. Juli 2009 20:39
An: [email protected]
Betreff: [Fedora-commons-developers] How RELS-INT breaks the Fedora
paradigm and opens the door for new and innovative solutions to old
problems
Hi
Steve Bayliss have just finished adding the RELS-INT datastream to
Fedora, as announced on this list. I have been in some discussion with
him, as also shown on this list. This discussion have granted me a
chance to fully understand the conceptual change that RELS-INT brings.
In the semantic web paradigm, everything with an URI is a thing, which
can have properties and so on. But in Fedora, so far only Objects could
have properties (relations)
This all changed with the introduction of RELS-INT. Steve Bayliss have
made a system for, in a fedora object, specifying object properties
with
a datastream id as subject. No more, no less.
So datastreams are now objects, so to speak. They have a URI, and they
can have properties themselves. Formerly, there was the Fedora Object,
which had datastreams (blobs of data) and properties. Now there is the
Datastream, which has ONE blob of data, and properties. Fedora objects
now has a list of Datastreams, and properties for the object itself.
So we have two levels of objects. This is the way the Fedora paradigm
is
broken.
Big deal? Yes. Because if the datastreams can have relations, they can
have the hasModel/rdf:type relation. So, suddently we have a framework
for talking about the classes of datastreams. Now, like the content
models, there is the possibility to specify restrictions and demands on
the datastream, both it's relations and it's content.
Some might remember the old problem with the DS-COMPOSITE-MODEL
datastream. There is no way to specify datastreams that might be there,
only datastreams that have to be there, and there is no way to specify
cardinality for datastreams.
With the use of RELS-INT and enhanced content models, we can now
specify
something close to a solution to this problem.
Enhanced Content Models give the ability to define an ontology for
subscribing objects. This could include relations from the object to
the
objects datastreams. On such relations, Enhanced COntent Models give
the
ability to make cardinality demands, and specify the class/content
model
of range.
So, in the RELS-EXT for an object you could make this blob
<rdf:Description rdf:about="info:fedora/demo:object1">
<fedora-system:hasModel rdf:resource="info:fedora/demo:cm1"/>
<demo:hasDCdatastream rdf:resource="info:fedora/demo:object1/DC1"/>
<demo:hasDCdatastream rdf:resource="info:fedora/demo:object1/DC2"/>
<demo:hasDCdatastream rdf:resource="info:fedora/demo:object1/DC3"/>
</rdf:Description>
Then in the ontology we would specify something like
<owl:Class rdf:about="info:fedora/doms:ContentModel_DOMS">
<rdfs:subClassOf>
<owl:Restriction>
<owl:onProperty rdf:resource="#hasDCdatastream"/>
<owl:minCardinality
rdf:datatype="http://www.w3.org/2001/XMLSchema#integer";>3</owl:minCardi
nality>
</owl:Restriction>
</rdfs:subClassOf>
<rdfs:subClassOf>
<owl:Restriction>
<owl:onProperty rdf:resource="#hasDCdatastream"/>
<owl:allValuesFrom
rdf:resource="info:fedora/demo:DCdatastreamcontentModel"/>
</owl:Restriction>
</rdfs:subClassOf>
</owl:Class>
This basically says that demo:object1 must have at least three
hasDCdatastream relations to things of the type
demo:DCdatastreamcontentModel
This in the RELS-INT in demo:object1
<rdf:Description rdf:about="info:fedora/demo:object1/DC1">
<fedora-system:hasModel
rdf:resource="info:fedora/demo:DCdatastramcontentModel"/>
</rdf:Description>
<rdf:Description rdf:about="info:fedora/demo:object1/DC2">
<fedora-system:hasModel
rdf:resource="info:fedora/demo:DCdatastramcontentModel"/>
</rdf:Description>
<rdf:Description rdf:about="info:fedora/demo:object1/DC3">
<fedora-system:hasModel
rdf:resource="info:fedora/demo:DCdatastramcontentModel"/>
</rdf:Description>
And voila, you have specified that objects of demo:cm1 must have at
least three datastreams, which all have a specific content model.
I have not fully thought everything above through, but I hope you get
the gist of it. I would like to hear other peoples thoughts on this.
Think of this as a preliminary on how RELS-INT can be used in enhanced
content models
Regards
Asger
Enhanced content models to be found on ecm.sourceforge.net
-----------------------------------------------------------------------
-------
_______________________________________________
Fedora-commons-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/fedora-commons-developers
-------------------------------------------------------
Fachinformationszentrum Karlsruhe, Gesellschaft für wissenschaftlich-technische Information mbH.
Sitz der Gesellschaft: Eggenstein-Leopoldshafen, Amtsgericht Mannheim HRB 101892.
Geschäftsführerin: Sabine Brünger-Weilandt.
Vorsitzender des Aufsichtsrats: MinR Hermann Riehl.
------------------------------------------------------------------------------
_______________________________________________
Fedora-commons-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/fedora-commons-developers
------------------------------------------------------------------------------
Enter the BlackBerry Developer Challenge
This is your chance to win up to $100,000 in prizes! For a limited time,
vendors submitting new applications to BlackBerry App World(TM) will have
the opportunity to enter the BlackBerry Developer Challenge. See full prize
details at: http://p.sf.net/sfu/Challenge
_______________________________________________
Fedora-commons-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/fedora-commons-developers
