I changed my version of contrib/cube to use double precision for most
things (the gist cost functions) in the package. This provides more
accuracy, but the exponents may still be limited or the gist cost
functions may supply bogus results.
I changed the output string allocation to use the dimension to figure out
how much space to request. I also use '%.16g' instead of '%g' to print
out elements.
I added four functions to make the individual elements and dimension accessible
(for making views in a different coordinate system) and a function that
returns whether or not a box is a point.
I haven't updated the regression test to test the new functions or account
for some output differences that come from increasing the precision in the
output format.
This is good enough for what I want. If these changes are something that
might get added to contrib/cube I can fix up the regression test and
make the changes available. If this isn't something of enough general
interest to worry about, then I won't.