Post by Stefan Monnier
But it ends up being more like =` than `eql` (i.e. it also
excludes -3.5e0 and -7/2).
Putting on my grandest language lawyer hat (the one with the tallest feathers sticking out of the crown) I'll point out this isn't _necessarily_ true.
The ANS description of FLOAT (see the dictionary entry in chapter 12) doesn't restrict the value of the float base _b_ (aka the radix), and in an implementation with radix a prime other than 2 there will be no float value that is exactly equal to -7/2. So the float approximation might be slightly greater or less than -7/2.
Of course, every known implementation uses IEEE in which the float radix is 2, so this problem is theoretical only.