Today's HEAD produced 49 unexpected failures, most of which are caused by trivia. - tcrun006 and tcrun029 use datatype contexts which were removed from HEAD (7 ways each), can be fixed by adding a language pragma (cf. also http://hackage.haskell.org/trac/ghc/ticket/5229) - ghcpk01.stdout hasn't yet been updated to include the new trusted field - ffi005 fails to compile (7 ways) due to an ambiguous occurrence of unsafePerformIO (Foreign and System.IO.Unsafe) - cg005 and T4059 have unexpected stderr due to a warning about unsafePerformIO going to be removed from Foreign - T4437 fails to compile due to a type change: T4437.hs:9:39: Couldn't match expected type `(String, CmdLineParser.FlagSafety, ExtensionFlag, DynFlags.TurnOnFlag -> DynFlags.DynP ())' with actual type `(t0, t1, t2)' In the pattern: (ext, _, _) In a stmt of a list comprehension: (ext, _, _) <- xFlags In the expression: [ext | (ext, _, _) <- xFlags] Those are easily fixed by the attached patches (but some may need a more principled fix). Further, 17 unexpected failures are due to hpc output being formatted differently from the expectation (5x hpc_fork, 6x hpc001, 6x tough). Two unexpected failures (T4809, cgrun068) are due to mtl not being built the dyn way. The remaining 5 are - T4801 (allocating too little because there are fewer Generic instances for tuples) - dph-diophantine-opt (3 ways) - dph-words-opt On my 32-bit box, also T3294 unexpectedly fails since about a week ago due to too little allocation (a bit below 690M, minimum allowed 800M). If that's reproducible and not known to be temporary, the bounds should be adjusted. Cheers, Daniel