On 2015-10-22 at 05:32, quoth ericbav...@openmailbox.org: > Testing for and with doctest is tricky. I'm not sure your comment about > missing test frameworks is accurate; it looks, from what I can tell, like all > the necessary inputs are present. > > I believe the issue is that tests that call the 'doctest' function may not > (probably do not) include the "-package-db" argument in the parameter list > given, but since haskell-build-system relies on the fact that cabal passes > that argument to GHC, it is essential that that information get there somehow. > I would say it's a deficiency of most package's build systems that they do not > take this situation into account.
I see, i finally understand the issue. That sounds exactly right, indeed. > So, either the tests files can be patched to include "-package-db" in the GHC > arguments given to the 'doctest' function, or GHC_PACKAGE_PATH can be defined > appropriately before the 'check' phase (at that point in the build cabal will > not complain, as it would if GHC_PACKAGE_PATH were defined during the > 'configure' phase). I think this is unnecessarily complex, especially given the upcoming native-search-path patch. > It looks to me like tests are disabled for all packages that use doctest in > this patch series: > > ghc-trifecta > ghc-lens > ghc-wai-logger > ghc-parsers > ghc-semigroupoids > ghc-comonad > ghc-distributive > ghc-http-types > ghc-unix-time > ghc-iproute > > I would suggest that tests for these packages simply be disabled (and the > corresponding test-only inputs be removed) for the time being. The pending > GHC native-search-path patches should trivially resolve this issue, at which > time tests for these packages can be revisited. I have indeed simply removed the test-only inputs on these packages, and fixed the comments to correspond to the real problem. Thanks, p.
