[RFC/RFA] Fix varobj.c value comparison problems

[M.M. Kettenis]

Andrew Cagney <cagney@gnu.org> wrote:

> Mark, the thing that's missing is an automated testcase.  Having
> something change from -0.0 to +0.0 should do it.

Ah, but that is the exact the opposite of what triggered this fix:

   NaN != NaN

even though the bit pattern is the same, whereas

   -0.0 == +0.0

even though the bit pattern is different.  I caught the first one with
the existing testsuite although that's merely by accident.  Writing a
test that does this reliable is tricky though as NaNs are handled
differently by different targets.  Testing the second one should be possible.  I'll look into it.

Meanwhile, I've checked in the patch.