Regenerate .travis.yml

[phadej]

Using haskell-CI/haskell-ci#106

[phadej]

I'll do the buildinfo trick @hvr mentioned (See unix package), and incorporate #245 here.

[kazu-yamamoto]

I don't understand why Travis CI is still failing...

[phadej]

the doctest suite doesn' work with new-build, things are in different places

[eborden]

There were some questions the last time this PR came around. Is new-build now the official and documented way to run cabal or is it still experimental?

Also my concerns about utilizing cabal-install-head still remain.

#254

[kazu-yamamoto]

@eborden The new features of cabal are used on Travis CI only. We can use any build tools locally as we like. So, I think we don't have to worry too much.

[kazu-yamamoto]

@phadej Please incorporate #245 and comment out the doctest.

[kazu-yamamoto]

If people agree, I will merge #245 to record @hvr's credit.
@phadej Then, you should rebase this PR onto master.

[phadej]

@kazu-yamamoto #245 or #254 ? :)

[phadej]

@kazu-yamamoto but please merge @hvr's #254, I haven't noticed he already did buildinfo.in trick! That's great. (We discussed that approach, and @hvr didn't even remember he already made a patch!)

[kazu-yamamoto]

Ah. I meant #254!

[kazu-yamamoto]

@eborden Do you agree with merging #254 (and #259 on the top of #254)?

[vincenthz]

@kazu-yamamoto I don't think it change the fact that new-build is not the default way to build for any stack user and cabal user < ghc 8.2 (and cabal user on 8.2 not yet using the non default new-build).
A pragmatic travis script at this stage would not remove the most currently used build path and would test cabal build and cabal new-build, not just new-build.

[eborden]

I don't have any major complaints about using new-build. I think @vincenthz's stance is reasonable. My only real concern is building utilizing cabal-install-head. network's CI is meant to ensure that network builds for most users of the package. Most users of the package are not going to be using cabal-install-head. I'm fine with adding a matrix point to utilize head the same way we do with GHC, but that matrix point should be allowed to fail.

[kazu-yamamoto]

@phadej I created the travis branch. I cannot guarantee that this branch will be merged but you can modify .travis.yml as you like on top of this branch.

To merge this branch to master, we should add cabal build in addition to the current cabal new-build. I don't know how to implement this.

[eborden]

@phadej any objection to changing cabal-install-head to cabal-install-2.0? It would be lovely if @hvr could add a cabal-install-stable to the ppa collection that tracks releases and to prevent having change this script with every release.

[kazu-yamamoto]

travis now supports 6 versions of GHC and CI works very well.
I need this feature to tackle other issues.
Actually, the build system detects mkWeakMVar is missing in GHC 7.4.
So, I will merge travis to master anyway and open a new issue to describe our concerns and todo list.