On Fri, 23 Jan 2026 19:50:49 GMT, Joe Darcy <[email protected]> wrote:

> First cut at adding modeling interfaces for the algebraic structures of 
> rings, groups, and fields.

> Oh: I said all that because it came up _twice_; the other was:
> 
> > so at least ordered fields would be skipped.
> 
> But the real numbers are an ordered field and a type like `Double` (setting 
> aside the special values) IS modeling the real numbers. The fact that it's an 
> imperfect _representation_ of the type it models, that suffers from 
> saturation/overflow, imho does not change its nature. imho we should not 
> think of `Double` as having anything to do with algebraic "finite fields" 
> even though as a representation it is hopelessly finite.

To be pedantic, for IEEE 754 floating-point, the specific mathematical object 
being approximated is not the real numbers (a field with lots of fun bonus 
properties), but the _extended_ real numbers -- the reals augmented with 
positive and negative infinity values. Due to the infinities, the extended 
reals are none of a group, ring, or field, but they do have some useful 
algebraic properties.

-------------

PR Comment: https://git.openjdk.org/valhalla/pull/1957#issuecomment-3792791469

Reply via email to