The comparison is using a two pass algorithm: in the first pass the properties of the warning are compared. If one property differs (e.g., line number) then this warning is moved to a second pass. In the second pass, the actual code is compared. This is currently done by checking the context above and below the warning (5 lines). Did the content (or whitespace) in the surrounding lines change?