git branching workflow for use_release_issue()?

[kevinushey]

Using https://github.com/rstudio/renv/issues/1634 as a template, I wonder if we should also suggest starting with something like:

• [ ] Create and switch to release branch release/<version>

Then, follow the regular steps. After building and submitting to CRAN, if the submission is accepted, we can then:

• [ ] Merge release branch to main

And then run the regular steps. Does that make sense?

[gadenbuie]

This is also what we do for Shiny for R, although we follow the branch naming convention of rc-vX.Y.Z. I personally really like this workflow.

One thing to watch out for is that you have to merge the release branch into the main branch in a way that preserves the git history if you want to tag the commit that was sent to CRAN (i.e. use a merge commit). If you're used to a squash merge workflow or don't mind tagging a different commit than the exact one sent to CRAN, you might need to remove the CRAN-SUBMISSION file created by submit_cran() before calling use_github_release().

[gadenbuie]

A common thread between this and #1887 in my work is that I would appreciate the ability to customize the release issue in a more detailed way. Currently, release_bullets() adds steps to the end of "Prepare for release", but a branching workflow or a workflow that requires additional checks might benefit from customizing where the todo items appear in the list.

One option would be to allow release_bullets() to return a named list such that items can be added to specific places in the release checklist, e.g. "prepare-start", "pre-check", "post-check", etc. This could be backwards compatible in that an unnamed list would be assumed to be "prepare-end" (at the end of the prepare section).

If there's interest in this approach, I'd be happy to work up a PR.

[hadley]

I'd be happy to make this the default workflow, assuming we can automate the steps sufficiently that I can follow the checklist without having to think about it. Maybe it's sufficient to add pr_init(), pr_push(), and pr_finish() in the appropriate places? I think it would be useful with a concrete proposal there.

(@gadenbuie we're tracking the issue of adding bullets anywhere in #1818, but I do think it's important to standardise as much as possible on one process)

cc @davisvaughan since I know this is something he has also asked for, and @thomasp85 since I think we do generally use a branch for ggplot2 releases.

[DavisVaughan]

@hadley I have done that actually. It is probably a little out of date now but I did it here https://github.com/business-science/tibbletime/issues/104, so we could maybe pull that over?

[hadley]

Yeah, that seems like a good start. Want to do a PR? (I'd leave the change in revdeps for now since we should discuss that elsewhere)