Special values handling is inconsistent in DoubleDoubleSummation

TEST(Summation, ConvertSpecialToDecimal) { constexpr double infinity = std::numeric_limits< double >::infinity(); DoubleDoubleSummation sum; sum.addDouble(infinity); ASSERT_EQUALS(infinity, sum.getDouble()); // ok ASSERT_TRUE(sum.getDecimal().isInfinite()); // <-- fail here, it's nan ASSERT_FALSE(sum.getDecimal().isNaN()); // <!-- fail here, it is actually nan sum.addDouble( 1 ); // now once we add something to sum, everything is ok ASSERT_EQUALS(infinity, sum.getDouble()); // ok, still infinity ASSERT_TRUE(sum.getDecimal().isInfinite()); // ok, now it's infinity ASSERT_FALSE(sum.getDecimal().isNaN()); // ok, not nan }