3.21.1 release tracker

[shonfeder]

Release branch: https://github.com/ocaml/dune/tree/3.21.1-rc

Preparation

• Need backport:
  • chore: bump ocaml-lmdb #13074
  • fix: add eta-expansions for OxCaml compatibility #13342 (for OxCaml)
    • has conflicts needing resolution, only needs to be partially backported
  • ox: keep compat with oxcaml/oxcaml#main #12846 (for OxCaml)
  • chore(test): make test menhir version agnostic #13534 (for 3.21 CI)
  • fix(melange): make melange.emit respect -p <PKG> #13522
    • depends on backport of test(melange): show (melange.emit ..) doesn't respect -p <PKG> #13521
  • fix: avoid killing process group when not applicable  #12360
    • 3.19 regression, consider skipping (will appear in 3.22)
    • too entangled with 3.22 at the moment
  • fix: stop starting rpc with dune promote #13428
    • has conflicts needing resolution
• Backports:
  • [3.21.1] chore: bump ocaml-lmdb #13560
  • chore(test): make test menhir version agnostic #13561
  • fix: add eta-expansions for OxCaml compatibility (#13342) #13562
  • backport: test(melange): show (melange.emit ..) doesn't respect -p <PKG> #13563
  • fix: stop starting rpc with dune promote (#13428) #13564

Release

• Update dune changelog to X.Y.Z on X.Y branch [3.21.1] Merge release branch #13602
• Open then pull request on opam-repository [new release] dune (17 packages) (3.21.1) opam-repository#29374
• Triage (ensure it does not break anything)

Post-release

• [3.21.1] Merge in release branch #13627
• Update ocaml.org changelog: https://github.com/ocaml/ocaml.org/pull/3498/changes#diff-2c49d3178f92df89fb598a76a5074dd46bccf3ecd7204935ea8ce7d709873d41
• Write a post about the release on Discuss: https://discuss.ocaml.org/t/ann-dune-3-21-0/17700/2
• Store the revdeps error file in the logs

[shonfeder]

FYI @Alizter

[anmonteiro]

@shonfeder would love to get this one in: #13522

[Leonidas-from-XIV]

I've created a backport of #13074 by cherry-picking and updated the description.

[Leonidas-from-XIV]

@anmonteiro Can you submit a backport to the 3.21.1-rc branch?

[Alizter]

@Leonidas-from-XIV I've handled it, just needed the test PR before.

[shonfeder]

We discussed this in our team meeting today. The last remaining work needed is to add a changelog entry for #13074 (The other's either have them or don't need them)

IMO, we should not merge the release branch here back into main:

• there are merge conflicts
• the only reason to backport patchs to a previous release branch is to make fixes available for earlier versions, for users who can't wait (or for some other reason cannot) update to the next linear release. backporting in this way effectively creates an LTS branch (even if the time span is limited) and merging these divergent branches back into main is not advised, confuses our history, and could introduce bugs

This means a manual followup will be needed to the main branch to reflect these changes into the changelog.

I've run out of time for this today, however. Will resume dealing with this tomorrow.

[Alizter]

I don't understand why merging back into main is difficult? In every conflict, main should be preferred. The only thing that will change in main is that some changelog entries get removed and the main changelog is updated.

[shonfeder]

We discussed merging back in our meeting:

• Merging back with conflicts is risky insofar as it introduces the needless chance of bugs (even if low, it is still needless).
• Since we are backporting fixes to the previous release branch, it is also incorrect, because the tag for the patch release should not end up on a commit that includes the history of everything we are not including in the backports. If were going to do this kind of thing, then we should just cut the next minor release.

There's a decent explanation of the dynamics at https://softwareengineering.stackexchange.com/a/361817

[rgrinberg]

because the tag for the patch release should not end up on a commit that includes the history of everything we are not including in the backport

Except that the tag is not on a commit that includes the history we are not including. The commit that includes the history of everything that we are not including has an ancestor with said tag. Which is exactly the thing that we are trying to accomplish.

[shonfeder]

Ah, I see. I had a defective mental model, now remedied. Sorry for the confusion @Alizter and @rgrinberg. The required PR is at #13627