Jim Meyering <address@hidden> writes:
Mart, let me know if you can assign copyright to the FSF,
and I'll send you the paperwork.
Paul?
Thanks, Mart. Nobody has come up with a better solution, so let's go
with Mart's, with one proviso: it should be documented a bit more
conservatively, in that the documentation shouldn't promise any
particular behavior for unnormalized numbers. How about putting
something like this in the documentation, instead of the last
paragraph of what was proposed in
<http://article.gmane.org/gmane.comp.gnu.coreutils.bugs/6793>?
The sort order is defined only for input numbers that are normalized
for either powers-of-1000 or powers-of-1024 notation, such as those
output by other @acronym{GNU} utilities; it is not defined for
unnormalized numbers. For example, @command{sort} considers
@samp{1023M} to be less than @samp{1.0G}; this is correct for
powers-of-1024 notation since 1 gibibyte is 1024 mebibytes, and it is
also "correct" for powers-of-1000 notation as @samp{1023M} is an
unnormalized number in that notation.