I think we should delegate the decision on what tag to make to the workflow. If we change the workflow to not run on tag, and instead run on workflow dispatch (like in v1) we can make the script decide what version to use by itself, and optionally have some controls for forcing version bumps if we really want different behavior to what the script decides.

The script will decide what version bump to use (based on tags like release-breaking or whatever other metric we want) and then create the tag by itself.

@deansheather are you simply referring to that ./release.sh would normally take no argument, but you can optionally give it --minor to bump the specific number?

Or are you referring to how ./release.sh is run. I.e. by someone locally or via a github action or some such?

Wrt to the latter, I think for the first step to improving this, it'd be fine if it's run manually by a user on their machine.

With regard to implementation, it seems quite straightforward how we can pull this information out during release. Some initial exploration:

# Compare HEAD to previous version
❯ git log --oneline HEAD...v0.13.1
81c39487 (HEAD -> main, origin/main, origin/HEAD) fix: Close tty first in `ptytest` cleanup (#5252)
ee4f0fc5 chore: enable debug logging for gotestsum (#5248)
8469dbc0 fix: add index to `provisioner_jobs.started_at` (#5245)
92c217bd fix: add index on `workspace_agents.auth_token` (#5244)

# For each hash, check associated PRs
❯ gh api -H "Accept: application/vnd.github+json" /repos/coder/coder/commits/81c39487/pulls | jq -r '.[].labels[].name'
release/test

This'll most likely require a GH API key so we don't run into any rate limitations, utilizing gh is optional but convenient.

It's very easy to add parameters to a workflow dispatch run so it can be run on GitHub. Check the release workflow in coder v1, it just asks for what version you want to mark and tags it for you. In this case, you would just have checkboxes for whether you want to "force" a major, minor or patch and then have the workflow determine the tag to use and create it, then build coder.

With #5366 merged, this issue is now 90% complete. We still need to disallow manual tag/push to prevent human error in deciding if it's a patch/minor release (#5382).