Saxon

I checked that there are tests in this area, e.g. open-040 et seq in the Saxon test contributions; so it's not something that's completely untested or unimplemented, although it's certainly not widely used.

I think that the problem is around UserComplexType line 1947 where it does

                    if (contentModel.isEmpty()) {
                        extendedParticle = null;
                        if (variety == ComplexVariety.ELEMENT_ONLY) {
                            variety = ComplexVariety.EMPTY;
                        }
                    }

to set the content model to empty when it finds that a complex type derived by extension and its base type both have empty content models. I suspect it shouldn't be doing this if there is an open content wildcard that applies to empty elements.

Indeed, this appears to fix the problem. Change if (contentModel.isEmpty()) in this code to

if (contentModel.isEmpty() && (openContentWildcard == null || !openContentAppliesToEmpty))

so that the content model remains ELEMENT_ONLY rather than EMPTY if there is an open content wildcard.