The following computation on wolframalpha gives the evaluation at the end. If you click 'approximation', it gives a value of ~0. If you ask for more digits, it tells you it's actually -2*10^6.
Ironically, I found this as a result of a "Wolfram alpha can't" question.
I edited the title. General rule: Don't call something a bug unless and until it has been verified as such.