Ok, then I don't see anything special that JCasCollectionReader_ImplBase does over CollectionReader_ImplBase other than calling cas.getJCas() before calling getNext().
I don't know the exact peculiarities involving the initialization of the JCas. Maybe it needs to be done at a certain point in time, e.g. before creating any annotations, such that calling it anywhere else than at the beginning or before getNext() doesn't work - which would b) and c). It may also explain a) if some seemingly redundant calls to getJCas() in uimaFIT 2.0.0 have been removed for 2.1.0. It's been a while, I don't remember. As I said, I've seen these problems largely with binary serialization, probably because I use JCasFactory() everywhere anyway. Cheers, -- Richard On 30.07.2015, at 15:09, [email protected] wrote: > Hi Richard! > > No, I don't use initialize() without args directly. I use > initialize(UimaContext context) and call super.initialize(context). > > Best, > Armin > > -----Ursprüngliche Nachricht----- > Von: Richard Eckart de Castilho [mailto:[email protected]] > Gesendet: Donnerstag, 30. Juli 2015 14:46 > An: [email protected] > Betreff: Re: DKPro NamedEntity ClassCastException > > On 30.07.2015, at 13:23, [email protected] wrote: > >> Hello Richard and Peter! >> >> The problem is solved but not really understood. >> >> a) Downgrading form uimaFIT 2.1.0 to uimaFIT 2.0.0 helped. So, it must be >> connected to 2.1.0 in some way. >> >> b) Adding an annotator with cas.getJCas() or jcas.getCas().getJCas() in >> process() to the begin of the pipeline does nothing. >> >> c) Adding cas.getJCas() or jcas.getCas().getJCas() to the collection reader >> does nothing. >> >> d) Changing the collection reader from CasCollectionReader_ImplBase to >> JCasCollectionReader_ImplBase solved it. I don't use JCas often. I prefer >> the more flexible CAS. > > Uhm - JCasCollectionReader_ImplBase does directly inherit from the UIMA > CollectionReader_ImplBase - nothing special. > > Did you possibly override the initialize() method (the one with no args) > without calling super? > ... although don't wouldn't see why that would make a difference either. > > Cheers, > > -- Richard
