Support GHC 9.12

[halostatue]

GHC 9.12 uses base 4.21 and ghc-prim 0.13; I also

When compiling with GHC 9.12.1, I got GHC-90584. I'm not a Haskell
programmer, so I made the simplest changes that could work — I don't
know how one would make the deriving Typeable conditional based on
GHC version or if there would be some other way to do that.

[halostatue]

I noticed that pandoc specifies base >= 4.12 && < 5; is there a good reason to not go with < 5 for unicode-transforms?

[harendra-kumar]

Sorry for the delay in reviewing this PR. Can you please make the suggested change and we will merge it ASAP.

There are also some CI errors in GHC 8.4/8.6, we need to either drop support of these compilers or fix the errors. We can open another issue/PR for that.

[harendra-kumar]

Since Typeable issue is only a warning and the rest of the changes are only bounds change we can just revise the package on hackage and everything should work. It won't be necessary to upload the new changes.

[harendra-kumar]

While we are at it, we can also update these:

$ cabal outdated
Outdated dependencies:
QuickCheck >=2.1 && <2.15 (latest: 2.15.0.1)
tasty-bench >=0.2.5 && <0.4 (latest: 0.4.1)

[halostatue]

While we are at it, we can also update these:

$ cabal outdated
Outdated dependencies:
QuickCheck >=2.1 && <2.15 (latest: 2.15.0.1)
tasty-bench >=0.2.5 && <0.4 (latest: 0.4.1)

Just saw this. Can do.

Should base be changed to < 5?

[halostatue]

I ask in part because MacPorts committed a change to its pandoc port which uses cabal build --allow-newer as it appears that this happens periodically and because there's few people who use Haskell and have to rediscover these changes to get them upstreamed, so…

[harendra-kumar]

Should base be changed to < 5?

I am neutral on it, we can ask @wismill and @Bodigrim as well. There is a risk if there is some breakage in base, though that is a very minimal risk. Hackage trustees advise to use the stricter major version constraint on base. Also, if we have to change bounds on other packages anyway (e.g. ghc-prim in this case) then there is no advantage of relaxing base.

[harendra-kumar]

I have revised the package on hackage, updating the bounds, so we do not have to upload a new release.

[halostatue]

OK. Let me push the version updates up.

[harendra-kumar]

There are also some CI errors in GHC 8.4/8.6, we need to either drop support of these compilers or fix the errors.

This should be easy to workaround. Only benchmark build is failing so we can disable that on these CIs:

See https://github.com/composewell/unicode-transforms/blob/master/.github/workflows/packcheck.yml#L155 . In the CI stanza we can add :

            pack_options: >-
              DISABLE_BENCH=y

same for 8.4.4 CI as well.

[halostatue]

That should get it working, except the s390x test, which appear to be failing for reasons completely out of the code's control.

[harendra-kumar]

Yeah, I do not know what needs to be done for s390 - I raised #101 for this.

[harendra-kumar]

Ok, looks like all done, I am merging it as soon as the CIs complete unless you have something else in mind.

[harendra-kumar]

We can optionally update the tested-with field in the cabal file to include 9.10 and 9.12 .

[halostatue]

I'm fine either way, but it will be ~18h before I get a chance to update anything.