I have made some progress analyzing the build failures with gfortran I see under 64-bit MacOS X.
The derivation whose build fails is gcc-apple-4.2.1.5666.3. I have the basic version (no Fortran etc.) installed from the nix_unstable as a binary, but when I request gfortran, it is built again with langF77 active. That seems quite reasonable. However, the build fails for a reason completely unrelated to Fortran: the patch file no-sys-dirs.patch, apparently made for gcc-4.1.0, refers to files that don't exist in gcc-4.2.1. Closer inspection shows more strangeness: the derivation downloads two source tar files, one for gcc and one for libstdcxx, but the latter contains another copy of the gcc sources and in fact seems to have the right structure for applying the patches. I tried to build (using nix-build) the very same gcc-4.2.1 that I installed as a binary package, i.e. gcc without Fortran. As expected, it fails. Which makes me wonder how the binary package I installed could ever have been compiled. One idea I had is that maybe the binaries were built from an earlier versions of the nixpkgs sources. If my understanding of Nix is right, this should not be possible because the cryptographic hashes should then be different. Just in case, I tried to go back to earlier revisions, before recent changes to gcc/4.2-apple64. Those builds all fail for other reasons, which I don't really want to explore. Is there any regular test of the build scripts in nixpkgs? The gcc for darwin looks like no one tried to build it for a very long time. Konrad. _______________________________________________ nix-dev mailing list [email protected] http://lists.science.uu.nl/mailman/listinfo/nix-dev
