Loop unroll fixes

[Joe Buck]

Joseph Myers writes:
> A harder case might be where the bug was shown up by a small testcase from
> a proprietary testsuite (see e.g.  
> <URL: http://gcc.gnu.org/ml/gcc-patches/1998-11/msg00257.html >).  Do we
> have any legal guidance on what is and is not safe to do in such cases?

We got some legal advice from Eben Moglen (FSF counsel) on test suite
issues.  We could perhaps ask him for advice on how to legally create a
test case that we can distribute.  But the following procedure is
generally considered safe:

programmer A, with access to the test case, writes up a report on what
the test is doing and what the failure is.

(for the extremely paranoid: lawyer reads report, sprinkles holy water,
but probably unneeded in this case)

programmer B uses the report to write a test case that shows the same
failure.