Shouldn’t hashcodes be calculated using all fields...
# codereview
r
Shouldn’t hashcodes be calculated using all fields? By definition that makes hashcodes the indicator of similarity, right?