chore(Convolution2DPass): fix incorrect initial render in example

[Jo-Byr]

Add missing renderer.updateLightsGeometryToFollowCamera() to fix incorrect initial render and the first camera interaction changing the lighting of the scene

Context

In Convolution2DPass example, the ligting of the scene changes after the first camera interaction (rotation or zoom).

Results

The lighting is correct at the first render.

Changes

Added missing renderer.updateLightsGeometryToFollowCamera();

[finetjul]

LGTM

[PaulHax]

Drive-by observation: this looks like a general gap rather than an example issue. renderer.updateLightGeometry() exists but nothing calls it, so lights only get refreshed from the interaction path. Anything that moves the camera without a mouse ends up stale: view preset buttons, restored camera state, and a fair number of examples in here that call elevation/azimuth after setup.

For reference, VTK C++ calls it from vtkOpenGLRenderer::DeviceRender(), right next to UpdateLights():

this->UpdateCamera();
this->UpdateLightGeometry();
this->UpdateLights();
this->UpdateGeometry();

vtk.js has the equivalent OpenGL/Renderer buildPass calling only updateLights(), so the light geometry half never happens.

#2536 may be the same root cause. This one looks related too: https://discourse.vtk.org/t/issues-with-lighting-and-clipping-in-multi-renderer-vtk-js-setup/15495 where the workaround was setFocusedRenderer(), which fits since only the poked renderer gets its lights updated.

Might be fixable in one place, either there or by updating lights when the camera is modified. Happy to open an issue if that is useful.

[Jo-Byr]

The issue I have is that in C++, vtkOpenGLRenderer inherits from vtkRenderer, and as such has access to UpdateLightGeometry, but in vtk-js, vtkOpenGLRenderer extends vtkViewNode.

vtkOpenGLRenderer holds vtkRenderer instances indirectly through model.renderable.getRenderWindow().getRenderers().

Should we call updateLightGeometry on each of them ?

[finetjul]

vtkOpenGLRenderer holds vtkRenderer instances indirectly through model.renderable.getRenderWindow().getRenderers().

Isn't model.renderable what you are looking for (i.e. the vtkRenderer paired with vtkOpenGLRenderer)?

[Jo-Byr]

vtkOpenGLRenderer holds vtkRenderer instances indirectly through model.renderable.getRenderWindow().getRenderers().

Isn't model.renderable what you are looking for (i.e. the vtkRenderer paired with vtkOpenGLRenderer)?

Absolutely, I fixed it

[finetjul]

Is the light geometry updated sufficiently often ?
i.e. is buildPass() called for each new camera position ?

[Jo-Byr]

is buildPass() called for each new camera position ?

Yes, maybe even too much. It's called every few ms as long as the left click is held down (in Convolution2DPass and SimpleCone examples).

[finetjul]

is buildPass() called for each new camera position ?

Yes, maybe even too much. It's called every few ms as long as the left click is held down (in Convolution2DPass and SimpleCone examples).

Sounds good then.

[finetjul]

LGTM

[finetjul]

Is there an example to test if #2536 is fixed ?

[finetjul]

Apparently some test(s) are failing