We are looking at ways to reduce memory footprint of our iris code in https://github.com/metoppv/improver and are somewhat stuck on calculating weighted averages with iris.cube.Cube.collapsed. It looks like Iris insists on weights having the same shape as the collapsed cube even though the underlying numpy.ma.average in iris.analysis.MEAN can happily work with 1D array of weights. Why this restriction?