Despite not updating any extent: Programmable filter "ERROR: [...] The update extent specified in the information for output port [...] on algorithm [...] is [...], which is outside the whole extent [...]."

I wrote a programmable filter which gives me the following error that I would like to get rid off:

ERROR: In /builds/gitlab-kitware-sciviz-ci/build/superbuild/paraview/src/VTK/Common/ExecutionModel/vtkStreamingDemandDrivenPipeline.cxx, line 878
vtkPVPostFilterExecutive (0x20bdd610): The update extent specified in the information for output port 0 on algorithm vtkPVPostFilter(0x1ff2ba90) is 0 800 0 84 0 256, which is outside the whole extent 0 800 0 88 0 0.

The confusing part is that inputs[0] has extent 0 800 0 84 0 256, and so does output which inherits the grid from inputs[0], does it not?

Below is a not so MWE. Can anyone help me out with the missing bit needed to get rid of this error?
I came across this clue, talking about using some UpdateExtent-function but I can’t quite figure out how to do it. Plus, in my case I am not even modifying any grid.

The filter gets two inputs with different rectilinear grids (different node counts).
Data of the second input (data_av below) are interpolated onto the grid of the first input and then substracted from those data.

So the data that I finally assign to the output (ufluc below) has the dimension of the output. Yet, Paraview complains about some “update extent”.
The produced data looks absolutely fine, too.

from paraview.vtk.numpy_interface import dataset_adapter as dsa
from paraview.vtk.numpy_interface import algorithms as algs
import numpy as np
from scipy.interpolate import RegularGridInterpolator

data = inputs[0]  # extent 0 800 0 84 0 256
data_av = inputs[1]  # extent 0 800 0 88 0 0

# Get extents
extents_av = data_av.GetExtent()
nx_ny_nz_av = tuple(extents_av[2*i+1]-extents_av[2*i]+1 for i in range(3))
x_av = np.array(data_av.GetXCoordinates())
y_av = np.array(data_av.GetYCoordinates())
z_av = np.array(data_av.GetZCoordinates())
xyz_av = [x_av, y_av, z_av]

avdims = np.where(np.array(nx_ny_nz_av) > 1)[0]
xyz_av_red = [x for i,x in enumerate([x_av, y_av, z_av]) if i in avdims]
xyz_av_red_g = np.meshgrid(*xyz_av_red, indexing='ij')
nx_ny_nz_av_red = [n for i,n in enumerate(nx_ny_nz_av) if i in avdims]

for ind, (block_in, block_out) in enumerate(zip(data, output)):

    extents = block_in.GetExtent()
    nx_ny_nz = tuple(extents[2*i+1]-extents[2*i]+1 for i in range(3))
    x = np.array(block_in.GetXCoordinates())
    y = np.array(block_in.GetYCoordinates())
    z = np.array(block_in.GetZCoordinates())
    xyz_red = [x for i,x in enumerate([x,y,z]) if i in avdims]
    xyz_red_g = np.meshgrid(*xyz_red, indexing='ij')
    nx_ny_nz_red = [n for i,n in enumerate(nx_ny_nz) if i in avdims]
    nx_ny_nz_red_ext = [n if i in avdims else 1 for i,n in enumerate(nx_ny_nz)]
    # Normalize existing fields
    for qname in block_in.PointData.keys():

        qname_av = qname + "_av"
        qname_fluc = qname + "fluc"

        if qname_av in data_av.PointData.keys():
            u = np.reshape(block_in.PointData[qname], nx_ny_nz, order='F')
            u_av_red = np.reshape(data_av.PointData[qname_av], nx_ny_nz_av, order='F').reshape(nx_ny_nz_av_red)
            u_av_interpolator = RegularGridInterpolator(xyz_av_red, u_av_red)
            u_av_interpolated = u_av_interpolator(tuple(xyz_red_g)).reshape(nx_ny_nz_red_ext)
            # taking advantage of numpy broadcasting rules, here
            ufluc = u - u_av_interpolated

            block_out.PointData.append(np.reshape(ufluc, (-1), order='F'), qname_fluc)

The origin of this bug is probably this one: vtkPVPostFilterExecutive (0x1e3ecaa0): The update extent specified in the information for output port 0 on algorithm vtkPVPostFilter (0x1f196a90)

Sorry for not having seen this earlier. I responded to the other post and the bug report. Here is the response:

As cryptic as this is, this is not a bug. By default, VTK filters that take multiple data objects in an input request the same extent, that of the first input’s WHOLE_EXTENT, from all input connections. Maybe not the best choice, but it is by design. To fix the issue in the script, you need to add the following to the RequestUpdateExtent script of your programmable filter:

SDDP = vtk.vtkStreamingDemandDrivenPipeline
inInfo = self.GetInputInformation(0,1)
extent = inInfo.Get(SDDP.WHOLE_EXTENT())
inInfo.Set(SDDP.UPDATE_EXTENT(), extent, 6)