[osg-users] Who is frustrated not to serialize their bufferobject configuration?

[Robert Osfield]

Hi Julian,

I don't have the time right now to dive in to a complex discussion.  A
few general feelings on the topic so you know where I stand:

First up sharing VAO's would be best done by shading a
VertexArrayBlock container class for per vertex arrays, that
osg::Geomertry objects would share.  The VertexArrayBlock would map to
VAO's.  At this point in time I don't think the performance benefit of
adding this extra level of complexity into the OSG is probably not
justified - in tests I did with your early VAO sharing code
performance was worse than the non shader VAO implementation that is
now in the core OSG.  You may have improved your VAO sharing
implementation to avoid the CPU overhead that was limiting
performance, but I wouldn't expect it'll make a big difference for
most OSG usage cases.

The BaseVertex extension to OpenGL is something that looks like a hack
for lazy OpenGL application developers that don't wish to update their
index and draw array offsets.  The osg::PrimitiveSet class has an
ofsetIndices(int) method makes it easy to readjust primitive sets when
vertex array are merged, so it's essentially the same feature, just
implemented on the OSG side and in a way that is compatible with all
OpenGL versions and compatible with scene graph code that traverses
the scene graph for things like intersection tests.  I am not aware of
any real benefit of base vertex for OSG users.

We are now not far from making the next stable OSG release, I still
have shader_pipeline functionality to complete for it, this is my
focus.  The shader_pipeline functionality will benefit the majority of
OSG users so in terms of bang for the buck this is where I need to
spend my time working on rather on niche areas of the OSG/OpenGL
usage.

Are there are small and uncontroversial changes to the OSG that will
help your usage case then please put them forward.

Robert.