The function partially sorts the elements so that identical values are
adjacent. `NA`

values come last. This is guaranteed to be
stable, so ties are preserved, and if the data are already
grouped/sorted, the grouping is unchanged. This is useful for
aggregation and is particularly fast for character vectors.

Under the covers, the `"radix"`

method of `order`

is
used, and the same caveats apply, including restrictions on character
encodings and lack of support for long vectors (those with
\(2^{31}\) or more elements). Real-valued numbers are slightly
rounded to account for numerical imprecision.

Like `order`

, for a classed R object the grouping is based on
the result of `xtfrm`

.