[
https://issues.apache.org/jira/browse/UIMA-2391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13262071#comment-13262071
]
Marshall Schor commented on UIMA-2391:
--------------------------------------
Some more discussion on if this is a bug, by analogy.
Adam said (above): An analogous case might be - what happens if I merge two
type systems that define the same type with the same named feature (call it
Foo.bar) but different range types. In type system 1, Foo.bar has range type A
and in type system 2 it has range type B. Would you want to say that in the
merged type system, Foo.bar could be A or B? Probably not."
If we modify this slightly: assume B is a subtype of A. Then B has all the
features A has, plus maybe more. A possible reasonable value for the merged
type system would be for Foo.bar to have range type B. Currently, UIMA Type
merging won't do this either.
The analogy discussed above is a case of a subtype "adding" features to an
existing type.
The "AllowedValues" attribute is unusual, in that, compared to it's supertype
(string), it "subtracts" possibilities. So, if you had some kind of merging,
you could make a case that the proper behavior for merging of these things is
to subtract the union of their negation - or in other words, if B has only
allowed value "a,b", but type A has "a,b,c", then the proper behavior is the
more restrictive one (a,b). This would satisfy the constraint that both users
of this would only see (a subset) of the allowed values either had declared.
> Uima type merging for string subtypes not working
> -------------------------------------------------
>
> Key: UIMA-2391
> URL: https://issues.apache.org/jira/browse/UIMA-2391
> Project: UIMA
> Issue Type: Bug
> Components: Core Java Framework
> Affects Versions: 2.3.1AS
> Environment: Linux on Power
> Reporter: Charles de Saint-Aignan
> Priority: Critical
> Attachments: UIMA-2391.patch
>
>
> The basic situation is that we are providing a UIMA-based core that other
> teams can extend to suit their needs. As such we are making use of UIMA type
> merging to allow them to add new features to existing types. This approach
> works fine since JCasGen merges the two definitions of the given type and
> produces a superset of the features. This is well documented here:
> http://uima.apache.org/d/uimaj-2.3.1/references.html#ugr.ref.jcas.merging_types.jcasgen_support
>
> However, in addition to this, we have the case where we have a string subtype
> with given allowedValues - lets say values a, b and c. The other team wants
> to extend this type and have additional allowedValues, say value d. Ideally,
> what I would like to do is the following (which follows the pattern used for
> adding features):
> Type Definition #1 (provided by core):
> <typeDescription>
> <name>com.ibm.Type</name>
> <description></description>
> <supertypeName>uima.cas.String</supertypeName>
> <allowedValues>
> <value>
> <string>a</string>
> <description></description>
> </value>
> <value>
> <string>b</string>
> <description></description>
> </value>
> <value>
> <string>c</string>
> <description></description>
> </value>
> </allowedValues>
> </typeDescription>
> Type Definition #2 (extension to core):
> <typeDescription>
> <name>com.ibm.Type</name>
> <description></description>
> <supertypeName>uima.cas.String</supertypeName>
> <allowedValues>
> <value>
> <string>d</string>
> <description></description>
> </value>
> </allowedValues>
> </typeDescription>
> In this case I wanted UIMA to recognize the two definitions at runtime and
> allow the superset of allowedValues. However, this does not do the trick -
> at runtime UIMA throws an exception saying that value d is not an allowed
> value for com.ibm.Type.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
