Reimplement cabal check

[ffaf1]

Reimplementing cabal check, closes #8211, #652, #2065.

For a detailed description of the problem at hand, the proposed solution, benefits and future work check the explanatory article.

Here I will only highlight bits to ease reviews. In brief:

• cabal check goes from “make a soup with flattenPackageDescription, check it” to “start from GenericPackageDescription and walk down, perform the appropriate checks near the relevant datatype”.
• Same goes for conditional blocks: the various ifs inside a .cabal are merged in a single ball of mud no more. Rather we visit the CondTree in a principled manner, building the target from slices and then validating it in its appropriate context.
• Checks happen inside CheckM m, a wrapped RSTW transformer where m abstracts over some IO/.tar archive/VCS filesystem.
• the only API changes are one function removal (function which is not used nor in this repo, nor in hackage-server, nor in any haskell repo, nor in stack) and a signature change for checkPackage, eliminating a useless parameter. I tackled the rewrite with the idea to minimise disruption.

To the reviewer: a good way of approaching this is to to avoid the two “large diffs” files at first. Check first how other modules were impacted and the two preliminary patches. Then you can dive in PackageDescription/Check.hs:

-- ☞ N.B.
--
-- Part of the tools/scaffold used to perform check is found in
-- Distribution.PackageDescription.Check.Prim. Summary of that module (for
-- how we use it here):
-- 1. we work inside a 'Check m a' monad (where `m` is an abstraction to
--    run non-pure checks);
-- 2. 'checkP', 'checkPre' functions perform checks (respectively pure and
--    non-pure);
-- 3. 'PackageCheck' and 'CheckExplanation' are types for warning severity
--    and description.


-- ------------------------------------------------------------
-- * Checking interface
-- ------------------------------------------------------------

-- | 'checkPackagePrim' is the most general way to invoke package checks.
-- We pass to it two interfaces (one to check contents of packages, the
-- other to inspect working tree for orphan files) and before that a
-- Boolean to indicate whether we want pure checks or not. Based on these
-- parameters, some checks will be performed, some omitted.
-- Generality over @m@ means we could do non pure checks in monads other
-- than IO (e.g. a virtual filesystem, like a zip file, a VCS filesystem,
-- etc).
checkPackagePrim :: Monad m => Bool -> Maybe (CheckPackageContentOps m) ->
                    Maybe (CheckPreDistributionOps m) ->
                    GenericPackageDescription -> m [PackageCheck]
⁝

and you should find your way.

Testsuite added in #8248 and #8250 passes. As a proof of what can be done after the rewrite, this PR also:

• closes cabal check warns about -O2 even if it is just a flag #7423.

If not merged by the time this is approved, #8441 and #8361 might need to be updated.

This patch is big, the logic of checks sometimes tricky, I expect breakage. I took care to minimise API change, this should make bugs pop up faster and allow me to fix them faster. Once everything is stable, features can be accomodated appropriately.

• Patches conform to the coding conventions.
• Any changes that could be relevant to users have been recorded in the changelog.
• The documentation has been updated, if necessary.
• QA notes: calling cabal check in your project folder should warn you about common mistakes, exit with 1 on more serious ones.

[jappeace]

no need to wait for me, I'm good at rebasing 😉

[ffaf1]

Contributing to cabal is a fullfilling experience! You can have an immediate impact on the ecosystem, people are helpful, thanks everyone for their support!

[andreasabel]

@ffaf1 Congrats on the successful merge, and also on your persistence! I am glad to see this project completed.
It was more than a Summer of Code, more like a Summer Plus Another Year of Code (SpaYoC). It is great you lived it out.

[ulysses4ever]

Reimplementing cabal check, closes #8211, #652, #2065.

#652 wasn't auto-closed. Should it be?

[ffaf1]

Thanks, I closed that manually!