`cabal check` should recommend against using the `stability` field

[sjakobi]

It seems to be consensus that the stability field is often misleading and shouldn't be used any more. See for example Simon Marlow's comment on this issue.

Therefore I propose that cabal check should discourage any explicit settings in the stability field.

As the stability field can also be set on a per-module basis (see e.g. Data.Set), I'd like cabal check to check all the modules too. My questions are:

1. Is checking the modules within the scope of cabal check?
2. Is the machinery for checking modules already in place?

[ezyang]

The test for a stability field in the package description should be easy enough.

cabal check does not currently make any invocations to the compiler so it's definitely not in the scope of current cabal check. And, AFAICT, probably the code that understands how to tell if a module has a stability field is in Haddock itself. Maybe this counts as a Haddock feature request?

[ulysses4ever]

@ffaf1 you did a ton of work on cabal check -- do you have an opinion on this?

[ffaf1]

As a first step we can definitely discourage it from .cabal.

If there relevant haddock parsing is available to us via a function, we could also check each file (I don't see how compiler calls are needed).