Description of problem: Apparently somehow llvm-3.1 is missing the ghc ARM support that was merged into llvm-3.0. Presumably an upstream mistake. Anyway I think we need to patch llvm-3.1 for ARM if we want have functioning ghc in F18 ARM. Currently Haskell programs linked with llvm-3.1 segfault very early on. How reproducible: 100% Steps to Reproduce: 1. try to build any ghc-* package with f18 arm 2. minimal testcase: $ cat > hello.hs main = putStrLn "hello" $ ghc --make hello.hs $ ./hello Actual results: Segfault! Expected results: no segfaults Additional info: http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-June/051119.html
Just to confirm: reverting to llvm-3.0-13.fc18 fixes the ARM segfaults.
According to the mail the original llvm-3.0 patch should work also for 3.1: http://lists.cs.uiuc.edu/pipermail/llvmdev/attachments/20111014/844ce355/attachment.pl
Referenced patch didn't apply cleanly on ARM. There was another one attached to the original email post that was much closer. http://lists.cs.uiuc.edu/pipermail/llvmdev/attachments/20120624/891b3ca2/attachment-0001.pl It needed one section manually applied so the attched patch I think is good. The following is a scratch build if someone could test that is fixes the problem I'll push the change I have in my local git and build it properly. http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=959544
Created attachment 597745 [details] regenerated ocaml patch
This is the build http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=959567
Thanks Peter for working on this. I will try to test your build.
(In reply to comment #5) > This is the build > http://arm.koji.fedoraproject.org/koji/taskinfo?taskID=959567 I successfully tested it on armv5tel and it fixes the ghc segfaults so looks good to me (tested both hello.hs and building the hscolour package in mock). I can try to test armv7hl later using the koji static repo (the arm rawhide seems to have some perl breakage today.)