Hi GHCers, I've noticed that the behaviour of Simon Marlow's buildbot has changed since ~Oct 5th when libffi was upgraded in the ghc-tarballs library. Prior to this date, the buildbot would either succeed, fail during cloning, or fail during testing. After this date, it fails a *lot* more. Some of these errors appear to be Alex related, but at least two classes of recurrent errors are related to libffi: 1. "cannot stat `libffi/build/.libs/libffi_convenience.a'" - see e.g. http://darcs.haskell.org/ghcBuilder/builders/simonmar-win32-head/411/8.html 2. "include/ffitarget.h:1:1: error: expected identifier or '(' before '!' token" - see e.g. http://darcs.haskell.org/ghcBuilder/builders/simonmar-win32-head/414/8.html
My Windows build is currently broken with error 2 and I have no idea how to resolve it. It looks like the in-tree GCC (from the ghc-tarballs) is being invoked on a file that transitively #includes libffi/build/i386-unknown-mingw32/include/ffitarget.h. However, this file is just a symlink to ./libffi/build/x86/include/ffitarget.h, and it looks like GCC is reading the *contents of the symlink* (which is some binary rubbish) rather than the *contents of the file it points to*. Has anyone else encountered this problem? If so, how have they fixed it? (This is all on Cygwin, by the way. I'm building with mingw/msys right now in another and it seems to be working.) Max _______________________________________________ Cvs-ghc mailing list [email protected] http://www.haskell.org/mailman/listinfo/cvs-ghc
