On 07/18/2018 12:25 PM, Achim Gratz wrote:
> the established FP math says these
> two values should not compare equal anyway.

Sure, and = still does that. = is about numeric comparison; eql is about value 
comparison. Neither dominates the other: for example, they disagree about 0.0 vs 
-0.0 (= says they're equal, but eql says they're distinguishable values and so 
are not equal), and conversely they disagree about 0.0e+NaN versus 0.0e+NaN (= 
says they're not equal, whereas eql says they're indistinguishable values are so 
are equal).

> If you stray from that
> convention there should be a really good reason for that and it needs to
> be prominently documented.

There is good reason: eq, eql, and equal are about distinguishing objects (vital 
for hash tables, e.g.) whereas = is about numeric comparison. These are 
different things, and the numeric-comparison rule for NaNs is inconsistent with 
how hash tables should work. I gave a shot at documenting this by installing the 
attached patch.