On Thu, 21 May 2026 12:23:03 GMT, Fredrik Bredberg <[email protected]> wrote:
> I've had truly interesting discussions with @toxaart about how he > reverse-engineered this floating-point function from way back. With pen and > paper, he explained the logic in a way that took me from thinking, “this is > pure magic” to “okay, I think I understand”. > > Do I really have the skills to do a proper review of this PR? No, I don’t. > Few people do. But I deeply trust Anton’s mathematical abilities. I’m also > sure that @jddarcy has the expertise to review it more thoroughly than I can, > as well as having the right connections from way back to feel comfortable > with this patch. > > Understanding this source, and fixing a bug in it, is about as non-trivial as > it gets. Well done, Anton. Thanks @fbredber . I asked retired numerical analyst Doug Priest to look at an early version of Anton's fix. Doug's suggestions have been incorporated into the final patch. As a math library function, `pow` is "interesting" and known to be difficult to compute an accurate result. ------------- PR Comment: https://git.openjdk.org/jdk/pull/31210#issuecomment-4510183814
