Only build-depend on fail and semigroups with ancient GHC

[clinty]

No description provided.

[jacobstanley]

Is this causing you a problem? fail and semigroups do that switch internally to make sure the right version is exported afaik

[clinty]

We don't have empty packages in Debian so any uses of them (fail, nats, bytestring-builder, etc.) need to be patched out which leads to a lot of wasted labor that needs to be repeated on every package upgrade. Things like semigroups are okay because they're not completely empty, but since you're not actually using Data.Semigroup.Generic there's no real point in including it either.

[moodmosaic]

We don't have empty packages in Debian so any uses of them (fail, nats, bytestring-builder, etc.) need to be patched out

What's the current procedure for working around this issue? I don't have Debian around to try this, but both builds are green (just had to re-run the AppVeyor one).

---

If we decide to merge this, at least the dependencies should be aligned with all the others 🔎, like so:

[clinty]

Here's an example of how we work around fail in the repline package: https://sources.debian.org/src/haskell-repline/0.2.1.0-2/debian/patches/no-fail/

(cf. sdiehl/repline#14)

[sol]

@moodmosaic from what I see in the current version of the Cabal file, I think this can be closed.

[moodmosaic]

Thank you, @sol 🙌 Closing this, though @clinty absolutely feel free to re-open if needed. Thank you for your work on this. 👍