Suspicious performance degradation in 7.8

[edsko]

Consider

IDE backend tests
  GHC 7.4.2
    Compilation
      Reject a program requiring -XNamedFieldPuns, then set the option: OK (28.47s)
  GHC 7.8.3
    Compilation
      Reject a program requiring -XNamedFieldPuns, then set the option: OK (88.26s)

No idea why this is happening (or if it's even an issue with ide-backend or whether this is entirely a GHC problem). It seems an awful long time to compile this however. Running with --no-exe reduces the times dramatically and also brings them closer together, although still almost a factor 2 in the difference:

IDE backend tests
  GHC 7.4.2
    Compilation
      Reject a program requiring -XNamedFieldPuns, then set the option: OK (3.34s)
  GHC 7.8.3
    Compilation
      Reject a program requiring -XNamedFieldPuns, then set the option: OK (5.04s)

[edsko]

@Mikolaj Perhaps you could take a quick look at the buildExe part of this and see if this is something that we are causing or whether it's just ghc?

[edsko]

(In general it would be good to compare test times between 7.4 and 7.8; need latest version of tasty to get the these execution times.)