cabal-install 3.6.2 release plan and post tasks

[jneira]

• Preserving a @Mikolaj comment in irc:

---

We've hit a milestone, so let me summarize the state of 3.6. The three main issues we know of has been solved by our esteemed contributors (in the order of disastrousness, only PRs to master listed):

• partial regression: broken install of packages with autoconf: Revert PR #7510 (which was "Set PATH_SEPARATOR=; when calling ./configure") #7651 Windows: rewrite paths to configure #7652
• regression: broken compilation of Paths_ with non-standard Prelude: Generate Paths_ module with qualified Data.List.last import #7650
• infra: broken CI on master and 3.6 (due to GHA changes last week): various PRs

Loose ends that we probably can't afford to (all) tie before 3.6.2:

• add changelogs for the fixes
• add tests that reproduce the two regressions:
  • extend the new dogfooding CI job (compiling cabal deps and code with cabal from master) that catches the autoconf regression to Macos and Linux: Extend new dogfooding ci job to linux and macos #7742
  • add focused, synthetic, future-proof autoconf scripts test with many sample generated files and/or add tests of compiling newest versions of network, process, time, old-time and any other important autoconf-using packages. splitted in Add focused, synthetic, future-proof autoconf scripts test #7762
  • add a test of compiling Paths_ with completely disabled Prelude: splitted in Stop using Prelude at all in autogenerated Paths_<packageName>.hs modules #7761
  • stop using Prelude at all in Paths_ (related: cabal install policeman fails using cabal version 3.6 kowainik/policeman#65, fixed with Generate Paths_ module with qualified Data.List.last import #7650) Splitted in Stop using Prelude at all in autogenerated Paths_<packageName>.hs modules #7761
• investigate the non-reproducible 3.6 Macos failure ticket cabal 3.6 does not accept a cabal-version of 3.4, 3.0 #7644

Backporting of all the firefighting PR is complete, as follows:
PRs on master (from old to new):

1 #7651 v
2 #7653 v
3 #7650 v
4 #7658 v forwardport of #7657
5 #7659 v
6 #7652 v

corresponding PR backports (some extended) on 3.6:

2 #7653 backported directly in the branch
4 #7657 forward-ported to #7658
5 #7659 backported with #7662
3 #7650 backported with #7656
#7665 3.6-exclusive CI fix
1 #7651 backported with #7664
6 #7652 backported with #7668

[fgaz]

add changelogs for the fixes

This must be resolved before the release, or the person that does the release should add those entries themselves. Being able to easily tell when a bug was fixed is very important imo.

[emilypi]

Please resolve before release, so that I don't have to question whether I fucked up the aggregation step and go hunting through the backlogs :)

@jneira, thanks for posting, and thank you all for working to get these fixes out

[jneira]

Ok, i am not familiar with the changelog process, could it be done manually?
@Mikolaj are the actually listed backports the source of the changelog?

I am afraid that 3.6 is unusable in windows in practice due to the network bug

[Mikolaj]

I think probably none of the PRs have any changelog so I'd propose to add some manually. I don't know if it can be done directly to CHANGELOG.md or if it really needs to be done as separate files.

I think there are 3 changes to changelog:

• change the partial regression of the autoconf fix to one that actually works
• make Paths_* work again with non-standard Preludes
• update boostrapping infrastructure to new versions of packages and to GHC 8.10.7

The rest is our internal dev plumbing and its fixes (bootstrapping is external, as well, and requested by Ben Gamari).

I'm afraid I'm still out of the picture. Thank you for taking care of poor 3.6.

[Mikolaj]

here we go, please review: #7687

[jneira]

nice, the rest of tasks could be spawned in dedicated issues I guess

[Mikolaj]

I doubly agree (I'm currently busy elsewhere).

[jneira]

this only needs split off issues about test improvements