3.22.2 release tracker

[shonfeder]

Fixes

Regressions requiring fixes in main and backports to 3.22.2-rc:

• Revert the behavior of --diff-command wrt to non-existent files #13891
  • reverted via fix: revert --diff-command behavior #14098
  • backported in [3.22.2] Backport: revert --diff-command behavior (#14098) #14124

For each PR needing a backport

• Run VERSION=3.22.2 PR=<pr-number> ./doc/dev/releases/backport.sh
• Wait for the PR into the rc branch to pass CI.
• Merge the PR, and record it above, completing the checklist item.

Prepare

• Mirage CI test: https://github.com/ocaml/dune/actions/runs/24251492552
• Revdeps for platform tools: https://github.com/ocaml/dune/actions/runs/24251695846/job/70812526312

The other new CI tests we've added are not yet available to run on the last minor branch.

Release

• Work thru the point release process.
• opam repo publication PR : [new release] dune (18 packages) (3.22.2) opam-repository#29704

Post-release

• Merge release branch into main [link to dune PR]
• Write a post about the release on Discuss: https://discuss.ocaml.org/t/ann-dune-3-22/17912/3?u=shonfeder
• Announce release on ocaml.org: Announce dune 3.22.2 ocaml.org#3608
• Store the revdeps error file in the logs as HTML

[anmonteiro]

Thanks for opening this!

I would love to add:

• fix(melange): stage .cmj files in sandbox when --mel-cross-module-opt #14018
• still working on the fix for test(melange): show underspecified dependencies during emission #14077, will have something until Saturday if that works?

[shonfeder]

Thanks, @anmonteiro :)

This is only a patch release, so only for backporting regressions. Neither of those appear to be regressions introduced in 3.22. Unless I'm mistaken (please correct me if I am mistaken), those will get picked up in 3.23, which we will start ASAP once this patch is out.

If all goes smoothly, we'll start on 3.23 early next week, so you will have time to get the fix for #14077.

[anmonteiro]

This is only a patch release, so only for backporting regressions.

What's the reason for this policy?

[shonfeder]

@anmonteiro we are currently trying to improve dune's release cadence and reliability (see #13771).
The ultimate motivation for these improvements is to make it possible for us to have a substantially increased stability and frequency of releases (when we want it). But, in particular, we want to increase the frequency at which we can cut minor releases because (1) it will reduce the latency between adding substantive improvements and sharing them with users and (2) it will increase the frequency of feedback on those improvements, which we can incorporate into our development cycles.

One of the feedback loops that is creating drag on the regularity and responsiveness of our releases is this: the greater the span between release, the more people want to delay the release to get their fixes or improvements included, because they don't want to wait weeks or months for their changes to be released. But this creates drag on the releases, reducing the cadence, which increases the need to request delays for inclusion, and so on...

In the particular case of patch releases, the more changes we load into a patch release, the greater the chance that we introduce new regressions, which will then require additional patch releases to repair, and cause the release of the minor version to be postponed. That means all the other additions that would otherwise get released in the minor are delayed. Furthermore, patches releases always have an increased risk, because it requires backporting changes, which means we may miss inclusion of some requisite change the patch depends on to work correctly.

Once we have achieved the target capability for our release cadence, I am happy to revisit the proposed policy on patch releases.

But, in any case, I wills start the minor release ASAP, so there shouldn't be long to wait.

Any objection?

[shonfeder]

I've initiated the release with the planned patch. As noted, we'll get the next minor in the works ASAP.

[anmonteiro]

@shonfeder thanks for trying to improve the release cadence! No objections and looking forward to more frequent releases.

Neither of those appear to be regressions introduced in 3.22.

This seems to be the root of the issue. They're both a regression from #13619, but I can wait until the next version given that you've already initiated the release.