numpy gradient() behaves differently between versi

Question

I noticed this when searching around numpy.gradient() usages. It seems that since numpy 1.13, the function treats it input argument differently than previous versions. Here is a simple example:

import numpy as np

x=np.linspace(0,10,100)
y=np.sin(x)

dx=np.gradient(x)

grad1=np.gradient(y,dx)
print grad1

grad2=np.gradient(y,x)
print grad2

I tested in numpy 1.13 and 1.11.3.

In 1.13, gradient(array1, array2) treats array2 as coordinates (x) for array1. But in 1.11.3, it treats array2 as the differentiation of coordinates (dx). So grad1 works in 1.11.3, and grad2 works for 1.13.

I think this is a rather dangeous trap, people may get wrong results if the same code is run using different versions of numpy. Or am I missing something obvious that clears this ambiguity?