Bump typelit plugins

[kleinreact]

The PR bumps ghc-typelits-natnormalise, ghc-typelits-knownnat, and ghc-typelits-extra to the latest versions and introduces some required code changes in clash-prelude to make it compile.

Still TODO:

• Blocked by Type error after plugin update ghc-typelits-natnormalise#108
• Requires Bump typelit plugins (preparations) #3062 obsolete due to Derive new givens ghc-typelits-natnormalise#110
• Blocked by Plugin causes either Winaccessible-code or Wincomplete-patterns ghc-typelits-natnormalise#109
• Requires Derive new givens ghc-typelits-natnormalise#110
• Write a changelog entry (see changelog/README.md)
• Check copyright notices are up to date in edited files

Nice to have:

• Add missing rewrite rules ghc-typelits-extra#64
• Check whether it fixes CLog type family reduction #3078

[DigitalBrains1]

Could you also locally run a check with --prefer-oldest at least on our oldest supported GHC, to see that the older plugins also still allow the changed code to type-check?

[kleinreact]

Could you also locally run a check with --prefer-oldest at least on our oldest supported GHC, to see that the older plugins also still allow the changed code to type-check?

Every build is currently failing due to warnings caused by clash-lang/ghc-typelits-natnormalise#105 (comment). I created another branch for just checking the changes to Clash.Sized.Vector: #3062.

[leonschoorl]

If the new code in Vector.hs doesn't typecheck with the old version of the plugin(s), the lower bound increases should be permanant.
But possibly conditional on the GHC version.

[christiaanb]

I’m adding some features and fixes in clash-lang/ghc-typelits-natnormalise#110 which will hopefully only require haddock/doctest changes w.r.t. error messages for clash-prelude.

[christiaanb]

With the changes in clash-lang/ghc-typelits-natnormalise#110, the following is all that's needed to compile clash-prelude warning-free and pass the unit-tests and doctests: 9b4ca8a

[DigitalBrains1]

[...] all that's needed to compile clash-prelude warning-free and pass the unit-tests and doctests: 9b4ca8a

I note you don't increase the lower bounds on the dependencies. So does it work with the whole range of versions in the interval then? I'm having trouble understanding how you fix a compilation error by merely allowing more versions of a dependency to be used...

[edit]
Oh! Wait, was the problem that only the new plugin version(s) caused an issue to begin with? If so, shouldn't we exclude ghc-typelits-natnormalise 0.9.1 specifically? Would that be enough for people not to construct a build plan that doesn't compile with -Werror and doesn't pass unit tests and doctests?
[/edit]

[christiaanb]

I have released version 0.9.3 of the natnormalise plugin, which requires no changes to any of our clash-* packages. See also c6f2ed3

[kleinreact]

I have released version 0.9.3 of the natnormalise plugin, which requires no changes to any of our clash-* packages.

That's great. Thanks.

I'd wait a bit more to also include the next ghc-typelits-extra release (due to the most recent features), and then this PR should be quite easy to land.