[Gdal-dev] On Feature.setGeometryDirectly SWIG bindings

[Tamas Szekeres]

2007/3/18, Ari Jolma <ari.jolma at tkk.fi>:
> Andrea Aime kirjoitti:
> >
> > One that has been tricky to identify is Feature.setGeometryDirectly.
> > This issue is there because every SWIG wrapper
> > does deallocate its C++ counterpart when garbage collected,
> > and thus the OGR Feature that got is geometry thru setGeometryDirectly
> > is left with a dangling reference.
>
> AFAI understand the SWIGTYPE *DISOWN that is used with that method
> should take care of the issue. The garbage collection system of the
> bound language should not deallocate the C++ object if it does not own it.
>
> On the other hand, one method which seems to be missing DISOWN, is the
> constructor of Feature. Since it assumes the ownership of the
> FeatureDefinition (which however is not usually owned by the bound
> language to begin with) it should be disowned.
>

The current implementation of OGRFeatureDefnShadow is confusing and I
doubt if it's working correctly as it stands. We can see some kind of
reference counting implemented inside OGRFeatureDefn, however the
destructor of OGRFeatureDefnShadow calls OGR_FD_Destroy regardless of
the actual reference count value. It would be better to call
OGR_FD_Release instead.

Apparently we should not mix the disown and refcount philosophy
because it may lead to various unpredictable problems in the future.

Best regards,

Tamas