I created https://issues.apache.org/jira/browse/UIMA-1874.
I'm adopting the approach of keeping the "requirement" that users call
super.initialize() if they use the standard base classes. Note that they don't
*have* to use these base classes...
-Marshall
On 9/3/2010 3:28 PM, Marshall Schor wrote:
> A recent "bug" in the Snowball annotator arose as follows:
>
> 1) It didn't implement the full annotator set of methods, but instead,
> depended
> on the implementation skeleton done by a "base" implementation class.
> 2) That base implementation class included an implementation of reconfigure()
> which is implemented as:
>
> public void reconfigure() throws ResourceConfigurationException,
> ResourceInitializationException {
> destroy();
> initialize(mContext);
> }
>
> where "mContext" is a field which might or might not be set (it isn't set by
> the
> framework).
>
> It is set by the base implementation class's implementation of the
> "initialize(uimaContext)" method. If the user has their own initialize
> method,
> and doesn't set this variable (which he would not normally be aware of)
> himself,
> he can set it by calling "super.initialize(uimaContext)".
>
> The bug that arose happened when the annotator writer didn't call
> "super.initialize", and then the application writer (different people,
> different
> roles) called "reconfigure", which the framework eventually passed down to the
> annotator.
>
> The user didn't implement this, but instead depended on the base
> implementation
> class. This referenced the null value for mContext, which caused the error.
>
> ========
>
> There are different kinds of things one could do here. One would be to add
> some
> more notes to the manual(s) (I didn't check, the notes might actually already
> be
> there, but we could in that case make them stand out more :-) ) saying that if
> you use the base implementation classes as superclasses of your annotator
> implementation, you *must* call super.initialize.
>
> Another would be to add some support in the getContext() default method which
> checks for null, and if found, issues a longer error message stating that the
> in
> user's implementations of initialize which extend xyz base class, the user
> *must* call super.initialize..., before calling getContext() or other
> framework
> methods that call getContext() (such as getEmptyCAS - used in CAS
> multipliers),
> or words to that effect.
>
> Another thing that could be done is to have the framework examine the
> annotator
> class before it calls its initialize method, to see if it is extending one of
> the base impl classes which has this field and makes use of it in the
> getContext() method, and if it finds that to be true, the framework could set
> this field itself. This would eliminate the possible error from happening,
> but
> maybe it's overkill for what is a simple thing for users to do.
>
> ========
>
> Are there other approaches to this, and which kind of fix do you think we
> should
> do for this?
>
> -Marshall
>
> P.S., the Snowball Annotator in our sandbox has been fixed
> (https://issues.apache.org/jira/browse/UIMA-1861). The other annotators in
> our
> sandbox and our examples haven't been "reviewed" as yet to see if they have
> this
> issue.
>
>
