I think it looks fine, but need base bumps for haskellari/lukko#30 and phadej/singleton-bool#20, which I opened now :)

(For anyone wondering it is possible to build cabal-install with ghc-9.4.1 using --allow-newer.)

Can a revision be done on Hackage now?

The only code change in this PR is addressing a -Werror in the test-suite. This isn't relevant for the released version.
So, a revision is likely possible, but someone should first make the experiment whether the released version builds & works with the relaxed bounds.

Yes, we'd need a manual (or in a special PR) test run with a constraint that forces these bounds. But that's on 3.8 branch, so only after this is backported. Let me backport...

@mergify backport 3.8

backport 3.8

✅ Backports have been created

Oh, actually the dev (default) version of cabal.project forces that, so it's enough to build and test with that one.

Oh, actually the dev (default) version of cabal.project forces that, so it's enough to build and test with that one.

Yeah, I couldn't get cabal to pick up the latest versions otherwise, so I forced it.

The version obtained with cabal get cabal-install and then patched with this PR builds:

cabal build -w ghc-9.2.4 -c "text > 2" -c "time >= 1.12"

The testsuites do not build:

$ cabal build --enable-tests -w ghc-9.2.4 -c "text > 2" -c "time >= 1.12"

Error: cabal: Could not resolve dependencies:
[__0] trying: cabal-install-3.8.1.0 (user goal)
[__1] trying: cabal-install:*test
[__2] unknown package: Cabal-described (dependency of cabal-install *test)

$ cabal info Cabal-described

Error: cabal: There is no package named 'Cabal-described'.

The dependency Cabal-described of test-suite long-tests does not exist.

Are the testsuites of the released versions supposed to build?

(If not we could delete them from the .cabal file before releasing.)

That's a bummer. If they are in the .cabal file, they should build. Agreed we should either remove them or publish enough little packages for them to build. I suppose the former is easier and we have more urgent issues.

Edit: ideally we'd remove them to yet another package so that they can stay there unmolested whenever we release cabal proper.

Ok, pushed the revision: https://hackage.haskell.org/package/cabal-install-3.8.1.0/revisions/

• Changed the library component's library dependency on 'text'
  from >=1.2.3 && <1.3
  to >=1.2.3 && <1.3 || >=2.0 && <2.1

• Changed the library component's library dependency on 'time'
  from >=1.5.0.1 && <1.12
  to >=1.5.0.1 && <1.13

@Mikolaj: Indeed, cabal-install-3.6.2.0 was released without any test suites.

Ok, pushed the revision

Oh, wouldn't it be better to test first? Not sure if the tests in the backport use the cabal.project or ignore it, but at least somebody could checkout 3.8 and test manually.

@Mikolaj: Indeed, cabal-install-3.6.2.0 was released without any test suites.

Let me open a ticket in that case.

Oh, wouldn't it be better to test first? Not sure if the tests in the backport use the cabal.project or ignore it, but at least somebody could checkout 3.8 and test manually.

Indeed, this would be safer. Branch 3.8 seems to be ahead of the released version, but there is likely a tag from which one can branch off, patch, and run the CI.
We might be lucky; I haven't heard of any serious regression in text-2.0 yet...

Unfortunately, the CI for the backport does not show whether text-2.0 is picked up:

• Validate CI should display the output of cabal-plan #8440

Unfortunately, the CI for the backport does not show whether text-2.0 is picked up

I've rummaged in the logs and it seems it did not build text 2 even the first time it started on GHC 9.2, so it probably ignores cabal.project. Too bad, otherwise we'd have a test for free. Anyway, it should get merged soon, so if a volunteer would like to test, that's a good moment. Otherwise, we will depend on our luck. :)

The backport is merged.