NetworkManager/MAINTAINERS.md

Triaging issues

Issue tracker: https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues

Help other maintainers with the triage following these guidelines. This way, it will be easier to find issues that require attention.

• Workflow: indicate the current stage with the corresponding label:

  • workflow::triage: needs triage to determine whether is a bug, an RFE or we should close it.

  • workflow::investigation: has been triaged but more info is needed to start working on a fix. This step can be skipped if enough information is provided during triage.

  • workflow::devel: on development (or waiting for it).

  • Closed: the issue is already solved, either via a code fix or via providing the required info. Also if the request is clearly incorrect or doesn't fit at all in the project.

• Assignee: assign the issue to the person that is working on it. This will remove the unassigned and help-needed::* labels.

• Manual labels: add suitable labels to indicate the state of open issues:

  • need-info: waiting for info or feedback from anyone.

  • good-first-issue: the task is simple and well suited for a first time contributor.

  • triaged::bug / triaged::RFE: type of issue.

  • need-discussion: something is not clear about what to do, or about if something has to be done at all. The problem should be discussed by the maintainers and/or with the reporter and/or other interested parts.

  • blocked: the issue is waiting for something that blocks its progress

  • close-proposed: there are good reasons to reject the request (explain those reasons when adding the label) but a chance to others to provide reasons against it is desired.
    It is not mandatory to always use this tag.

  • Spam: mark as spam, blocking the author. See damspam.

• Automatic labels: some labels are automatically added and removed by a bot and the maintainers don't need to care much about them. The label stale is used for issues without activity for a long time. The labels unassigned and help-needed::{triage, investigation, devel} are used for issues without an assignee.

Merging Merge Requests

• almost all new code, gets merged to main branch only. Stable branches only receive backports via git cherry-pick -x.

• almost always, make sure that the merge request is rebased against current main branch.

• if the merge request contains multiple patches, create a --no-ff merge request that envelop the patches. The merge commit can have a trivial commit message like "area: merge branch 'xy/topic'" and refer to all relevant resources. At least the full URL to the gitlab merge request should be there.

• for single patches, the merge commit can be skipped. In that case, add the full URL to the commit message before merging.

• before merging the result to main, make again sure that the merge request is up-to date (in particular, if you just rebased the branch or amended the commit message). So usually first do a git push origin -f -o ci.skip to update the merge request one more time, and the push the merge request to main. The result is that the merge request in gitlab is shown as "Merged".

• always refer to relevant URLs (bugzilla, gitlab issues, gitlab merge request). Do so via full URLs, not abbreviations like "!XYZ", so that the URL is clickable in the browser. Note that while the merge request is still under review and being reworked, we will frequently force push the branch. Gitlab and github will create backlinks to full URLs, so we want not to specify those URLs while development, but the moment before merging, we will add them. This means, usually when we decide that a merge request is ready to be merged, we still need to rebase it to latest main and amend the commit messages. Then we usually need to push once more to the merge-request, before pushing the final result to main.

The purpose of this elaborate scheme is to get a clean history that is easy to review and links to relevant resources.

If you forget to mention an URL, you can do so afterwards via git-notes. See CONTRIBUTING.md.

Upstream backports

There are situations where it is necessary to backport a patch to an earlier version of NetworkManager.

In order to do the backport, use git cherry-pick -x. Please use the commit from the next stable branch. If the commit is not on that branch then it is also necessary to backport to that branch.

Example:

We want to backport commit 323e182768 from main (1.33) branch to nm-1-30 branch. In order to do that, we must search if this bug has been backported to 1.32.

git log --all --grep "323e18276894591712a5e29f6e907562c79c5216"

In case the backport to 1.32 is missing it would not show anything so please do the backport to 1.32 first.

If the backport is done, the output should be similar to:

commit c94b1c43d4b5c5b88d67d7966d23a005028e78d8
Author: Thomas Haller <thaller@redhat.com>
Date:   Wed Sep 1 09:30:29 2021 +0200

    cloud-setup: return structure for get_config() result instead of generic hash table

    Returning a struct seems easier to understand, because then the result
    is typed.

    Also, we might return additional results, which are system wide and not
    per-interface.

    (cherry picked from commit 323e18276894591712a5e29f6e907562c79c5216)

In this case, the commit that should be backported is c94b1c43d4.

Resolving conflicts

To find conflicts when doing a backporting in NetworkManager is very common but we do not resolve the conflicts manually. Instead, we abort the current cherry-pick and search for the commit that introduced the changes that are causing the conflict and backport it too.

We only resolve the conflict manually if the extra commit introduces a lot of unnecessary changes or excesive code changes which is not common.

Backporting API

NetworkManager allows the users to build their application against the latest stable release and then run it against a newer release without relinking. To allow this, we need to guarantee that after we release a version that includes a new libnm linker version, then any release done after that point with a higher version number contains that linker version with the same symbols.

In practice when we want to backport new API from main we have two options:

• if the new API hasn't been included in a stable release of NetworkManager, then we can just backport the API to the old branch and pretend it was introduced there. For example, 8763e6da9c introduces new API on main during the 1.21 development cycle; 1.22 is not released yet. Then the symbol is backported to nm-1-20 before 1.20.6 with commit 90671a30b7. The symbol on main branch is then adjusted with 551fd3e28f. Note that at this point 1.20.6 must be released before 1.22.0.

• if the new API is already included in a stable release, we backport the API to the old branch and then duplicate the symbol on main with both versions. For example, 2e2ff6f27a introduces new API on main, which is released as 1.12.0. The API is backported to 1.10.14 in commit 19d7e66099. Then, on main we add duplicate symbols with commit 5eade4da11.

NetworkManager release process

It's mostly automated by release.sh.

Before running the script:

• For stable releases, remember to backport all commits with "Fixes:" tag that are applicable. Use the find-backports script to find them.
• Start all the jobs in the latest Gitlab pipeline of the right branch. The script checks that they ran successfully.
  Tiers 1 and 2 must pass, failed Tier 3 jobs can be fixed after the release.

The script also takes care of choosing the right version number depending on the release type that you specify, like devel, rc1, rc, major, major-post, etc. Run the script with --help to see all options.

Notes:

• You need access to master.gnome.org, see here.
• The GPG key used to sign the tags must be exported to a keyserver.

Versioning scheme, automatically handled by the script (version numbers are called MAJOR.MINOR.MICRO):

• Development releases has an odd MINOR version number (i.e. 1.47.2).
• Stable releases has an even MINOR version number (i.e. 1.48.1).
• Release candidates (RC) are tagged like 1.48-rc1, 1.48-rc2, etc. But in NM's internal code they looks like 1.47.90, 1.47.91, etc. (MINOR is one number less, and MICRO is >= 90).

The main differences between the different kind of releases are:

• Development releases: for depelopment and testing purposes only.
• Release candidates (RC): stabilization phase before a stable release. Normally there are one or two RCs with ~2 weeks cadence. More RCs can be releases if they are needed.
• Stable releases: Releases within the same stable branch should remain very stable while fixing important bugs, backported from main. New features are added very rarely.

Stable branches are branched out from main to prepare the first release candidate (RC) of the next stable branch. These branches are called nm-MAJOR-MINOR (i.e. nm-1-48). As they are used to release stable versions, the last number is always even.

There are some additional tasks that the script doesn't handle:

• For RC releases:
  • The NEWS file should reflect a curated summary of the changes that the new stable release will include.
  • The release should be announced on the mailing list.
• For stable releases:
  • The official documentation must be updated on the website when there is a new stable release. Use the import-docs.sh script from the website's repo.
  • The release should be announced on the mailing list.

VPN plugins and nm-applet release process

The same versioning scheme and release process is used for the VPN plugins, nm-applet (including nm-connection-editor) and libnma.

Note that each of them is hosted in its own repository, but this is documented here to avoid duplication, as the process is the same for all (at least for those that we maintain).

Also note that there are no stable branches or development versions. Everything is developed on main, and releases are done on main.

Versioning scheme (version numbers are called MAJOR.MINOR.MICRO):

• Small changes increments only the MICRO number.
• Bigger changes or new features increments the MINOR number.
• There is no strict criteria to define what change is small or big, but try to adhere mostly to semantic versioning.
• Use only even numbers for MINOR, skipping odd ones. That way we use the same versioning scheme than the main NM project despite there are no development versions here.

When doing a release, follow this process:

1. Ensure that NEWS file is up to date.
2. Increment the version in configure.ac, commit and tag the commit. Example: git tag -s 1.2.8 -m 'Tag 1.2.8'.
3. Ensure that you are on the right commit and create the tarball: git clean -fdx && ./autogen.sh && make distcheck
4. Upload the tarball: scp ./*-*.tar.xz "$user@master.gnome.org:"
5. Login to master.gnome.org and run ftpadmin install.
   Ensure the new tarballs show up at https://download.gnome.org/sources/ (happens after a short delay)
6. Announce the release on the mailing list.

Notes:

• You need access to master.gnome.org, see here.
• The GPG key used to sign the tags must be exported to a keyserver.

Reimporting systemd

See here.

Copr repository

See here.

Gitlab-ci Pipelines

See here.