Missing upper bounds on `ghc` on hackage

[andreasabel]

trafficstars

Leaving through the releases of apply-refact on hackage, I see no upper bounds on ghc starting with release 0.7.0.0. This may not reflect buildability correctly, see e.g. #114. What would be the correct upper bound for the released versions?

[zliu41]

@andreasabel Thanks for catching this. I've updated the upper bounds of base for the latest several versions on Hackage. Since ghc versions map 1:1 to base versions, will remove bounds on ghc in future releases.

[andreasabel]

@zliu41 wrote:

Since ghc versions map 1:1 to base versions, will remove bounds on ghc in future releases.

Are you sure about this? Aren't there two GHCs to consider?

• The GHC you are compiling apply-refact with (here, a bound on base can be used to restrict compilability---although I do not see why one would restrict this way).
• The GHC library you are using (to parse, print, manipulate Haskell code). This is the ghc package, I suppose.

[zliu41]

My understanding is that you can't depend on a different version of the GHC library than the GHC compiler version you are using. That is the whole reason why ghc-lib exists.

So there's no need to have bounds on both base and ghc. While I don't have a preference on which one to put the bounds on, almost all Hackage packages have bounds on base so I chose base.

[andreasabel]

I think because build failures (e.g., https://github.com/ndmitchell/hlint/runs/4691040973?check_suite_focus=true#step:8:286) happen due to changes in the ghc module rather than changes in base, it would make more sense to put the relevant bounds on ghc.