Data alignment issue

The current if then else assume that the data aligned and have same length:

def _if_then_else(x1, x2, x3):
    """if x1 is nonzero (keep NaN), then x2, else x3"""
    return np.where(x1, x2, np.where(~np.isnan(x1), x3, np.nan))

However, with time series operator, the data usually don't have the same length