Brian Dessent wrote:
Brooks Moses wrote:
In short, from what I could tell from a quick scan of that PR, the
problem is that you've got LD_LIBRARY_PATH set in such a way that it's
not including the GMP header files.
If you're using the standard Cygwin-package installation of GMP, I'd
guess this is because you haven't installed the "development" GMP
package that includes the header files; since Cygwin puts that into a
separate package, you need to install both. (MPFR is the same way.)
I don't think that applies in the case of Cygwin, since Windows does not
know anything about LD_LIBRARY_PATH. (Cygwin emulates it for dlopen()
but I don't think that's relevant here.)
I can say that I was able to successfully build and test the 4.1.2
20070129 (prerelease) version under Cygwin using the Cygwin-packaged
mpfr (2.2.1) and gmp (4.2.1) without any setting of LD_LIBRARY_PATH and
it passed all tests except for gfortran.dg/secnds.f which is a known
bug. So unless the released 4.1.2 tarball differs from the RC1 tarball
in this area, it's something else and not a gmp/mpfr/libgfortran
problem.
Well, that, and there's also the point you graciously didn't make, that
LD_LIBRARY_PATH has nothing to do with header files anyway. :)
I agree that LD_LIBRARY_PATH is a red herring, and I should have made it
clearer in the original post that I thought the real problem was a
missing package, not that it needed to be manually set. I'll still bet
that the problem is that the original poster hasn't installed all of the
relevant Cygwin GMP packages -- but part of that bet is that I vaguely
remember having similar problems myself with thinking I'd installed
everything and then finding I'd missed one of the packages.
- Brooks
