Some days ago I announced having success in porting ghc-6.8.1 to FreeBSD-7.0-amd64. As the compiler seemed to be able to compile itself, I was confident to make the binary package available to the public. Now I have found out that the port is rooten in a very weird way.
From the files in testsuite (6.6.1) it has 4707 expected passes vs. 669 unexpected failures. Among the most unexpected of these is the failure of arith003 (conversions between Int and Integer near 64bit boundary). It seems there is still some 32/64-bit confusion left from the building process.
Still worse is, that any attempt to use the partially usable compiler as bootstrap tool fails for some reasons. On a freshly configure 6.8.1 tree (with ./configure --with-gmp-includes=/usr/local/include --with-gmp-libraries=/usr/local/lib ) the build process fails at libraries/base/Control/Concurrent.hs because gmp.h is not found by ghc-inplace I fixed this by copying /usr/local/include/gmp.h to ghc-6.8.1/includes. This is, of course, a risk of confusing the 32/64bit version of this library. Anyway, the resulting compiler in stage2/ghc-6.8.1 dumps core. Perhaps I can get better results, if I use the most conservative settings possible for stage1 build, but I have no idea what they are. The recommended porting process (via .hc files) fails, too. I am going to make a separate bug report about this. -- Dipl.-Math. Wilhelm Bernhard Kloke Institut fuer Arbeitsphysiologie an der Universitaet Dortmund Ardeystrasse 67, D-44139 Dortmund, Tel. 0231-1084-257 PGP: http://vestein.arb-phys.uni-dortmund.de/~wb/mypublic.key