P0863R1.pdf

The C++ committee reviewed the Consistent Comparison proposal and found an issue with the partialorder comparison algorithm, which checks a == b and a < b but not b < a, and always returns a value other than partialorder::unordered when the inputs are unordered. The current design assumes operator< provides at least a weak order, but this has unfortunate consequences. The desirable properties of the fallbacks include providing genuine strong, weak, or partial orders when the underlying type's operators implement the corresponding ordering. However, the current wording supports all properties except for property 3, which concerns partial orders. Modifying partialorder to test b < a and return partialordering::unordered if both return false would fix the issue but come at a cost, as properties 8 and 9 would no longer hold. The author argues that properties 7, 8, and 9 have no value and that the modification should be made to ensure correct results for partial orders. Additionally, the cost of allowing strongorder and weakorder on partially ordered types is considered, and it is suggested that SFINAE tests on these expressions cannot provide useful information about the comparison semantics of the type.