On Monday, November 3, 2014 4:30:05 PM UTC+1, Francesco Mazzoli wrote:

Hi all,

Apologies if this is not the right space to ask this question.

We set up CI for Agda with Travis: https://travis-ci.org/agda/agda.

The problem is that we're having trouble with older versions of GHC. For example, with GHC 7.4.2, the build complains that ScopedTypeVariables is not enabled, even if it is. You can check the build failure here: https://travis-ci.org/agda/agda/jobs/39834738#L795, and the relevant file here: < https://github.com/agda/agda/blob/9a82afaf7657a97e85e7e31ea23f6386f99acd35/src/full/Agda/TypeChecking/Serialise.hs>.

As you can see from the pragmas ScopedTypeVariables is enabled, and in fact when building locally I have no problems.

We also got some spurious messages regarding syntax.

Has anybody encountered similar problems? I don't know if it's something depending on Travis or some configuration mistakes, but I'm a bit at loss on how to research this. Moving the OPTIONS_GHC pragma after the LANGUAGE pragmas seems to fix the issue but I'd like to know what's going on.

Are you sure that option order is accepted by earlier GHCs? You've checked locally with the same GHC version and build process, right? Last time I had problems (crazy transient issues, seemingly unrelated), Travis people asked me to open an issue at https://github.com/travis-ci/travis-ci/issues. Everything else failing, they might let you ssh to a machine to debug the issue. They seem to be very helpful and investigate quickly (even though I have next to 0 users). A few other things I've noticed while looking around: * For what it's worth, you're not really supposed to specify patch versions (7.4.2, as opposed to 7.4) of GHC in your Travis config. However, you can probably reuse ekmett's Travis config (not sure about licenses), since it seems more flexible and installs a different GHC, maybe with a better setup (Travis seems to be short on resident Haskell experts: https://github.com/travis-ci/travis-ci/issues/2690#issuecomment-52875095) https://github.com/ekmett/lens/blob/master/.travis.yml * I've encountered something "like this", for crazy values of "like" — as in, even stranger universe implosions on Travis for Haskell (though that looks some orthogonal setup problem): https://github.com/travis-ci/travis-ci/issues/2761 * That seems solved in later builds, but why the GHC version is changing midway here? I've found that linked in a comment from your Travis configuration. https://travis-ci.org/agda/agda/jobs/38699335#L402-L405