Float (original) (raw)

scala.math.Ordering.Float

Orderings for Floats.

The default extends Ordering.Float.TotalOrdering.

Ordering.Float.TotalOrdering uses the java.lang.Float.compare semantics for all operations. Scala also provides the Ordering.Float.IeeeOrdering semantics. Which uses the IEEE 754 semantics for float ordering.

Historically: IeeeOrdering was used in Scala from 2.10.x through 2.12.x. This changed in 2.13.0 to TotalOrdering.

Prior to Scala 2.10.0, the Ordering instance used semantics consistent with java.lang.Float.compare.

Scala 2.10.0 changed the implementation of lt, equiv, min, etc., to be IEEE 754 compliant, while keeping the compare method NOT compliant, creating an internally inconsistent instance. IEEE 754 specifies that 0.0F == -0.0F. In addition, it requires all comparisons with Float.NaN return false thus 0.0F < Float.NaN, 0.0F > Float.NaN, and Float.NaN == Float.NaN all yield false, analogous None in flatMap.

List(0.0F, 1.0F, 0.0F / 0.0F, -1.0F / 0.0F).sorted      // List(-Infinity, 0.0, 1.0, NaN)
List(0.0F, 1.0F, 0.0F / 0.0F, -1.0F / 0.0F).min         // -Infinity
implicitly[Ordering[Float]].lt(0.0F, 0.0F / 0.0F)       // true
{
  import Ordering.Float.IeeeOrdering
  List(0.0F, 1.0F, 0.0F / 0.0F, -1.0F / 0.0F).sorted    // List(-Infinity, 0.0, 1.0, NaN)
  List(0.0F, 1.0F, 0.0F / 0.0F, -1.0F / 0.0F).min       // NaN
  implicitly[Ordering[Float]].lt(0.0F, 0.0F / 0.0F)     // false
}

Attributes

Source

Ordering.scala

Graph

Supertypes

Self type

Members list

An ordering for Floats which is consistent with IEEE specifications whenever possible.

lt, lteq, equiv, gteq and gt are consistent with primitive comparison operations for Floats, and return false when called with NaN.
min and max are consistent with math.min and math.max, and return NaN when called with NaN as either argument.
compare behaves the same as java.lang.Float.compare.

Because the behavior of Floats specified by IEEE is not consistent with a total ordering when dealing with NaN, there are two orderings defined for Float: TotalOrdering, which is consistent with a total ordering, and IeeeOrdering, which is consistent as much as possible with IEEE spec and floating point operations defined in scala.math.

This ordering may be preferable for numeric contexts.