(abs -0.0)

[moon-chilled]

It's not explicitly specified what the absolute value of -0.0 is, if -0.0 exists. There is an example:

(eql (abs -0.0) -0.0) => true

But examples are not normative, and this disagrees with standard practice (checked sbcl, ccl, abcl, ecl; clisp has no -0.0). Presumably, the intent of the standard was that -0.0 and 0.0 are both valid results for the absolute value. I suggest to say that (abs -0.0) must always be a positive zero, because:

• agrees with ieee 754
• agrees with standard practice
• agrees with float-sign
• under the limit interpretation of signed zeroes, it is the more useful result

It does break the invariant that (not (minusp x)) implies (eql x (abs x)). (Or, put another way, it eliminates (if (< x 0) (- x) x) as a potential implementation of abs for reals.) But signed zero already breaks the putative invariant that, if two numbers are = and they have the same type, then they are eql, so that does not seem like too big of a deal.