0.4 test failure: Ambiguous module name ‘Data.HashSet’

[juhp]

See commercialhaskell/stackage#1010

wagger2-0.4/src/Data/Swagger/Schema/Internal.hs:18:8:
    Ambiguous module name ‘Data.HashSet’:
      it was found in multiple packages:
      unordered-containers-0.2.5.1@unord_7E41cj9RvqSJeZ4jSnbs1a hashmap-1.3.0.1@hashm_622M1jMBjyMIwDJT1SEKMF
Test suite doctest: FAIL

Test failed in the Stackage Nightly for swagger2-0.4

[dmjio]

hashmap is deprecated, use unordered-containers if possible

[fizruk]

I am not sure why this happens, given that hashmap is not in swagger2 dependencies.

Should we use PackageImports to explicitly specify unordered-containers?
I think that would fix the problem, but it seems weird to manually disambiguate a module that is not ambiguous when you consider dependencies in .cabal file.

[bergmark]

You are correct in that you shouldn't need PackageImports, strange error. I also queued swagger2 at http://matrix.hackage.haskell.org/package/swagger2

[bergmark]

Sorry I missed that this was the test-suite, so matrix won't help. But it did uncover other build errors

[fizruk]

@bergmark I am sorry, but where can I find the error? I see no logs there.

[bergmark]

By clicking the red (failure) cells

• http://matrix.hackage.haskell.org/package/swagger2#GHC-7.10/swagger2-0.4
• http://matrix.hackage.haskell.org/package/swagger2#GHC-7.8/swagger2-0.4
• http://matrix.hackage.haskell.org/package/swagger2#GHC-7.6/swagger2-0.4

[fizruk]

The GHC 7.10 build error is weird because it refers to deriveJSON's generated code.
I see that it uses latest aeson-0.10.0.0, so I can probably fix it by setting an upper bound on aeson.

@bergmark how can I improve my Travis builds to catch this kind of errors before the release?

[bergmark]

I use https://github.com/hvr/multi-ghc-travis to test multiple GHC versions. Matrix essentially just does cabal install swagger2 so since travis starts with a clean sandbox it should give the same results.

[fizruk]

I hoped for a stack-based version of multi-ghc-travis :)

Okay, will switch back to cabal-install then.

[fizruk]

@bergmark what should I do about tests failure?
Am I correct that the problem was introduced by this swagger2.cabal diff v0.3–v0.4?

[bergmark]

Yes that seems correct. From the little I know about doctest it sounds plausible that it doesn't use packages dependencies properly and instead uses the whole packagedb. If so, a PackageImport would solve this. But it may be wise to ask the doctest developers as well.

[fizruk]

The GHC 7.10 build failure in matrix is due to aeson-0.10.0.0 bug (see haskell/aeson#293).
I am surprised it hasn't been released yet!

I have added an upper version bound for aeson in 9b3470d until the fix is released.

[fizruk]

Closing this issue since both build problems should be fixed in v0.4.1.

[juhp]

Thanks!