 At this time, these test files check the computations only, not the
printout.  If successful, testall.out should contain a lot of calls
to "aeq", each of which is immediately followed by "T", showing that
the test succeeded.
  You will also see a lot of warnings "X matrix deemed to be singular".
They are a result of how I am using coxph.detail, and can be ignored.

 One test of interest is found near to the end of the output from ftest2:

> c("Non-sparse log|H|"=det2a, "Sparse log|H|"= det2c)  #of interest
 Non-sparse log|H| Sparse log|H| 
          255.9002       255.903

This was a check to see how much inaccuracy was created by using a sparse
approximation to the Hessian.  I was very surprised at how little it is.

Note: a lot of the files end with rm(x, y, etc).  I got in the habit of
this due to latent bugs.  I would make some addition to the library, write
a new testfile, and get strange results.  Some temp variable within the
function would have been misspelled or not yet defined, and instead of getting
the "variable not found" error that would have helped, it instead picked up
one of my many temporary variables from earlier test runs.  So I keep the
total number of retained variable names in the directory minimal.
