On Mon, 27 Oct 2025 14:27:44 GMT, Raffaello Giulietti <[email protected]> wrote:
> OK, I think I got the logic of the tests and, as a consequence, the need to > have slightly different values than the ones published. So again, take my > comments with a grain of salt. Right -- as the test is currently written, it wants to only allow the two floating-point numbers bracketing the true answer. The convention it is using is the reference value should be the endpoint closer to zero rather than the endpoint closer to the true result. Following that convention required the adjustments you found. ------------- PR Comment: https://git.openjdk.org/jdk/pull/27939#issuecomment-3452127809
