CI validate-old-ghcs broken after haskell-actions/setup upgraded to node20

[andreasabel]

The container for the "old ghcs" workflow uses a container phadej/ghc:8.8.4-xenial that does not support node20.

cabal/.github/workflows/validate.yml

Lines 223 to 232 in 519f2f7

	validate-old-ghcs:
	name: Validate old ghcs ${{ matrix.extra-ghc }}
	runs-on: ubuntu-latest
	needs: validate
	# This job needs an older ubuntu (16.04) cause
	# the required old ghcs using the `-dyn` flavour
	# are not installable from ppa/hvr in newer ones
	# see https://github.com/haskell/cabal/issues/8011
	container:
	image: phadej/ghc:8.8.4-xenial

Since I bumped haskell-actions/setup to node20 the workflow fails, missing a recent glibc:
https://github.com/haskell/cabal/actions/runs/8496523229/job/23275396876#step:5:16

/__e/node20/bin/node: /lib/x86_64-linux-gnu/libm.so.6: version `GLIBC_2.27' not found (required by /__e/node20/bin/node)

A short term solution would be to pin to haskell-actions/[email protected], however, GH will axe node16 in a couple of weeks, and then the problem will be back.

I'd rather axe GHC 7 validation.
E.g., for 7.0.4, less than 20% of the tests are actually run. Not much of a point, if you ask me.
https://github.com/haskell/cabal/actions/runs/8392859676/job/22990941034#step:8:455

532 tests, 436 skipped, 0 unexpected passes, 0 unexpected fails.

Or someone has to make a new container.

[andreasabel]

There seem to be other issues with the validate-old-ghcs workflow:

It builds cabal-install but does not run the cabal-install tests. Why?

• https://github.com/haskell/cabal/actions/runs/8392859676/job/22990941034#step:7:135

  cabal-install-3.11.0.0 (exe:cabal) (first run)

• https://github.com/haskell/cabal/actions/runs/8392859676/job/22990941034#step:8:16

  PackageTests/TestSuiteTests/ExeV10/cabal-with-hpc.multitest.hs SKIP no cabal-install

The validate.sh script calls jq which is not installed: https://github.com/haskell/cabal/actions/runs/8392859676/job/22990941034#step:7:156

validate.sh: 379: validate.sh: jq: not found

Also, something I don't understand: Why is it essential to install the -dyn version of the old ghc?

cabal/.github/workflows/validate.yml

Lines 256 to 259 in 519f2f7

	- name: Install extra compiler
	run: |
	apt-get update
	apt-get install -y ghc-${{ matrix.extra-ghc }}-dyn

Is the -dyn version actually run?

cabal/.github/workflows/validate.yml

Lines 282 to 285 in 519f2f7

	- name: "Validate lib-suite-extras --extra-hc ghc-${{ matrix.extra-ghc }}"
	env:
	EXTRA_GHC: "/opt/ghc/${{ matrix.extra-ghc }}/bin/ghc-${{ matrix.extra-ghc }}"
	run: sh validate.sh ${{ env.COMMON_FLAGS }} --lib-only -s lib-suite-extras --extra-hc ${{ env.EXTRA_GHC }}

[Mikolaj]

Ouch. I'm afraid the answers to most of your questions, Andreas, are lost in time, like tears in rain.

The "old ghcs" workflow also tests against a few old GHC 8 versions. Are they affected as well?

Could we just use a newer phadej/* container or do all of them have the same problem?

Can we use ghcup or gihub default instead of the containers?

[Mikolaj]

Copied from https://hackmd.io/c1yOln1fR_K5WiEhhPsUlQ?both:

there seems to be a consensus to drop old ghc (old: not on ghcup) and there's hope the others can relatively easily be handled by switching from containers to ghcup (if they are not already set up like that). Volunteers welcome, but in the worst case, we'll react to the CI outage once it happens.

[andreasabel]

I give it a shot:

• CI: stop validating GHC 7 #9886