Re: Persistency & Polymorphism

Thanks Axel,

this works just like you said it. Thinking about it now, it makes a lot of sense, too!

Bye, Christian

On Mon, 2007-02-12 at 16:19 +0100, Axel Naumann wrote:
> Hi Christian,
>
> i/o is done in a recursive way, where ROOT iterates through the base
> classes. Your base class AbsTrackRep will need a dictionary. TObject has
> one as well, even though almost nobody ever stores a pure TObject to
> file: it's needed for all the classes derived from TObject, too. And it
> does not matter that AbsTrackRep is abstract.
>
> So just add
> #pragma link C++ class AbsTrackRep+;
>
> Cheers, Axel.
>
> Christian Hoeppner wrote:
> > Hello,
> >
> > I have thew following problem I want to solve with ROOT:
> >
> > I want to make an instance of class called "Track" persistent to a ROOT
> > file.
> >
> > "Track" contains a collection of TrackReps (which are track
> > representations used with a Kalman filter). There is an abstract base
> > class "AbsTrackRep" (which inherits from TObject), and derived from that
> > are e.g. "HelixTrackRep" and "StraightLineTrackRep" (or in general
> > "ConcreteTrackRep").
> >
> > I figure that the right collection is a "TObjArray", since it contain a
> > collection of whatever, as long as everything inherits from TObject.
> > Please note, that the Array will never contain any instances of
> > "AbsTrackRep" but only different "ConcreteTrackReps".
> >
> > However, when trying to persist, I get an error message:
> >
> > No dictionary for AbsTrackRep
> >
> > This is completely true, because this abstract base class doesn't have
> > one, it can't even be instanciated. The "ConcreteTrackRep"s however have
> > dictionaries, and these are the ones which should be used.
> >
> >
> > Here is what I could track down what happens when the TObjArray Streamer
> > is called:
> >
> > TObjArray has a method Streamer which is used for writing to the IO. It
> > uses the << operator of TBuffer to do so. This operator uses the
> >
> > Int_t TBuffer::WriteObjectAny(const void *obj, const TClass *ptrClass)
> >
> > method. In this method something strange seems to happen:
> >
> > temp -= clActual->GetBaseClassOffset(ptrClass);
> > WriteObject(temp, clActual);
> >
> > This code somehow gets the adress of the base class and tries to call the
> > dictionary for the base class. And this is exactly where the "no dictionary"
> > error comes from. I tried this one the CINT command line after I loaded my
> > shared lib. The clActual points to a HelixTrackRep and temp to AbsTrackRep.
> >
> > Finally my question: Does anyone know how to work around this. E.g. is there
> > some pragma for the linkdef to tell root not to look for the base class
> > in the WriteObjectAny method?
> >
> >
> > Cheers, Christian
> >
> >
> >
Received on Mon Feb 12 2007 - 18:42:52 CET