compute_gradient yields wrong results for multiple parameters

[Simon Funke]

The helper function compute_gradient returns in some cases None instead of the correct gradient.

This occurs for pScalarParameters which are part of the functional definition only (i.e. do not show up in the forward annotation) and if the gradient of 2 or more parameters is to be computed.

For example:
J = Functional((forward-x)*(forward-x)*dx) # x is a Constant that only shows up the the functional definition

m1 = [ScalarParameter("nu")] # nu is a Constant that show up in the forward model
m2 = [ScalarParameter("x")]

dJdm1 = compute_gradient(J, m1, forget=False) # => Constant(...)
dJdm2 = compute_gradient(J, m2, forget=False) # => Constant(...)
dJdm = compute_gradient(J, m1 + m2, forget=False) # => [Constant(...), None] (!!!!!)

A full code demonstration is attached.

[Simon Funke (simon-funke)]

This is now fixed in commit 784.