Re: ACTION-264: Discuss imports with Tim Redmond.

On Thu, Feb 12, 2009 at 11:44 AM, Peter F. Patel-Schneider
<pfps@research.bell-labs.com> wrote:
> From: Alan Ruttenberg <alanruttenberg@gmail.com>
> Subject: ACTION-264: Discuss imports with Tim Redmond.
> Date: Thu, 12 Feb 2009 08:04:25 -0800
>
>>
>> I had a talk with Tim on tuesday. He is concerned with the following
>> situation - a zip file of ontologies is sent, perhaps a development
>> version of a modular ontology. Someone wants Protege to open and edit
>> this ontology. He needs to examine the folder and figure out how to
>> resolve the various imports. For the purposes of this discussion we
>> can assume that all ontologies in the closure are in the folder.
>>
>> How can he do this? He points out that the language we use is
>>
>> If O contains an ontology IRI OI but no version IRI, then the ontology
>> document of O should be accessible from the IRI OI.
>>
>> He points out that this is different from saying
>>
>> ... should be accessed from ...
>>
>> or
>>
>> ... and is the one that would be accessed.
>
> How so?

I guess that if one only says where things ontologies might be, we
aren't saying actually what should be done.
But I am trying to best report my conversation with Tim, so he might
want to elaborate.

>
>> This could be fixed, in his opinion, by amending the description of
>> canonical parsing.
>>
>> Current:
>>
>> CP-2.1         Retrieve the ontology document DI from I as specified
>> in Section 3.2.
>>
>> Since 3.2 only specifies where one might retrieve the document from
>> (where it is accessible), tightening this to:
>>
>> CP-2.1 Retrieve the ontology document DI from from a location that I
>> 3.2 says it is accessible from.
>
> Huh?  How can I (an IRI) say where it is accessible from?

typo: CP-2.1 Retrieve the ontology document DI from from a location I that
3.2 says it is accessible from.

>> We also discussed that having a portable way of specifying the a
>> redirection mapping might better deal with this rather common case.
>
> Why would this be better than the current case?

Because along with the zip file such a specification could be
included, and that could precisely state which of the files were
ontology documents for each of the ontologies. Imagine, for instance,
a folder full of versions of the same ontology (he specifically cited
that this is the case for certain influential users of protege). Such
a specification could unequivocally state which of the files is to be
used.

>
>> -----
>>
>> Our discussion pointed to two more issues to note:
>>
>> 1) The current behavior of Protege 4 is that when loading an ontology
>> from a file system, it always looks in the same directory for
>> ontologies that are imported? Should this be the default behavior? I
>> would argue not. Tools may implement a redirection mechanism, and
>> protege 4 supplies one in the form of an ontology library mechanism
>> where a set of directories to search is specified. Therefore absent an
>> explicit mention of the "." directory in the ontology libraries the
>> ontology should be accessed from IRI specified in the imports or
>> versionuri statement.
>>
>> In any case it would be nice if our document could say enough that the
>> appropriate behavior could be determined.
>
> What appropriate behaviour?  Are you arguing for something like "don't
> put all your ontologies in one directory"?  What difference does this make?

Protege currently would look at the imports statements and the
ontology headers of all the files and then tries to figure out, which
documents should be imported. He felt that our specification didn't
help him enough in this task. The example cited below, which is
ambiguous, is one example of this problem.

>
>> 2) He notes a case in the aforementioned zip file use case that can
>> not be resolved at all:
>>
>> Two ontologies in the zip file ("headers" below)
>>
>> ontology foo versionuri bar
>> ontology foo versionuri bar1
>>
>> The document that is loaded has the following:
>> import foo
>>
>> There is no way to determine which of the two documents on disk is to
>> be preferred over the other.
>
> Why not?  There are lots of ways that one could make this determination,
> such as via file names, via IRI-to-file-name redirections, etc.

He wants the specification to help him figure out how to do this, and
laments that it doesn't.

>
>> I don't see any way of repairing this, however it does suggest that
>> enumerating a couple of examples might be a useful addition to our
>> documentation.
>
> What kind of examples?  There are currently three examples in the
> document, and one of these is even for off-line processing.  I do not
> think that it is the job of the WG to provide examples for cases that
> don't related to the Semantic Web.

I gave a use case above. I think this *is* related to the Semantic Web.

-Alan

>
> peter
>

Received on Thursday, 12 February 2009 21:34:45 UTC