The mocked class here is org.apache.lucene.document.TextField. setStringValue is void.
My Specification looks like this...
given:
...
TextField textFieldMock = GroovyMock( TextField )
// textField is a field of the ConsoleHandler class, ch is a Spy of that class
ch.textField = textFieldMock
// same results with or without this line:
textFieldMock.setStringValue( _ ) >> null
// NB I explain about this line below:
textFieldMock.getClass() >> Object.class
the corresponding app code looks like this:
assert textField != null
singleLDoc.add(textField)
writerDocument.paragraphIterator.each{
println( "textField == null? ${ textField == null }" )
println( "textField ${ textField.getClass() }" )
textField.setStringValue( it.textContent ) // NB this is line 114
indexWriter.addDocument( singleLDoc )
The output from the printlns is
textField == null? false
textField class java.lang.Object
... which tends to prove that the mock is happening and getClass is being successfully replaced. If I get rid of the line textFieldMock.getClass() >> Object.class I get this output:
textField == null? false
textField null
In both cases the fail happens on the next line:
java.lang.NullPointerException
at org.apache.lucene.document.Field.setStringValue(Field.java:307)
at org.spockframework.mock.runtime.GroovyMockMetaClass.doInvokeMethod(GroovyMockMetaClass.java:86)
at org.spockframework.mock.runtime.GroovyMockMetaClass.invokeMethod(GroovyMockMetaClass.java:42)
at core.ConsoleHandler.parse_closure1(ConsoleHandler.groovy:114)
Line 114 is the setStringValue line. Field here is the (non-final) superclass of TextField.
To me it appears that something funny is happening: as though Spock were saying to itself: "ah, this class TextField is final, so I'll consult its parent class, and use the method setStringValue from there... and I find/decide that it is not a mock..."
Why is setStringValue not being mocked (or "substituted" or whatever the right term is for a method...)?
later
I went and took a look at Field.java in the package in question. The relevant lines are:
public void setStringValue(String value) {
if (!(fieldsData instanceof String)) {
throw new IllegalArgumentException("cannot change value type from " + fieldsData.getClass().getSimpleName() + " to String");
}
if (value == null) {
throw new IllegalArgumentException("value must not be null");
}
fieldsData = value;
}
... line 307 (implicated for the NPE) turns out to be the first throw new IllegalArgumentException... line. Quite odd. Suggesting that fieldsData is null (as you'd expect).
But why does Spock find itself processing this bit of code in class Field at all? Illogical: it's mocking, Jim, but not as we know it.
PS I later tried it with a (real) ConsoleHandler and got the same results. I have just noticed that when Spock output suggests you use a GroovyMock it says "If the code under test is written in Groovy, use a Groovy mock." This class isn't... but so far in my testing code I've used GroovyMock for several Java classes from Java packages, including others from Lucene... without this problem...
PPS workaround I got nowhere and in the end just created a wrapper class which encapsulates the offending final TextField (and will sprout whatever methods are needed...).
I have had experience of struggling with Lucene classes in the past: many of these turn out to be final or to have final methods. Before anyone makes the point that you don't need to test packages that can already be trusted (with which I agree!), you still need to test your own use of such classes as you develop your code.
