The MSVC compiler emits a warning about precision loss when converting int64_t(0x8fffffffffffffffLL) to a double. We'd like our compiles to be warning-free. The correct solution here is probably to suppress the warning on Windows, since we're intentionally trying to lose precision for this test.