I regularly work in a few repos — most especially the one for The Rust Programming Language — which have a gh-pages branch for deploying to GitHub Pages.¹ I also basically never want to see the history of the gh-pages branch when working in the repo, because in most cases, it is utterly uninteresting. Here’s how I set up Jujutsu to

First, I created a revset which grabs all changes which are ancestors of a local or remote gh-pages bookmark (which jj maps to Git branches; more on the difference in some other post). I put this in my user configuration for Jujutsu,² because I do not want to repeat it everywhere:

[revset-aliases]
ghp = 'ancestors(bookmarks(gh-pages) | remote_bookmarks(gh-pages))'
"immutable_heads()" = "builtin_immutable_heads() | ghp"

The second of those definitions defines the set of changes Jujutsu treats as “immutable”, i.e., the changes it will not let you alter without explicitly opting into it with --ignore-immutable. This is a handy bit of protection to avoid accidentally rebasing or otherwise editing main or a tag (something I have definitely missed in Git). The buildin_immutable_heads() revset is just what it says: the default value for immutable_heads() that comes with Jujutsu. Presently, that is defined as:

buildin_immutable_heads() = "present(trunk()) | tags() | untracked_remote_bookmarks()"

That means by default, Jujutsu will not let you edit the trunk(), if it exists, or any tags, or any untracked remote bookmarks. Meanwhile trunk() will use main, master, or trunk bookmarks if they exist, or the default branch on the default remote if you are using a Git-backed repo.

Shipping buildin_immutable_heads() as a revset rather than defining immutable_heads() directly in terms of definition of buildin_immutable_heads() lets us refer to it this way and stay in sync if the project is able to land improvements to the definition of buildin_immutable_heads().

I defined the immutable_heads() this way because that in turn lets me use ~immutable() without having to remember to additionally exclude ghp — so if I want to rebase all my open, mutable branches, I can now just write this (see my post on rebasing with all: from last week):

jj rebase -b "all:(mine() ~ immutable())"

That, and anything else which uses immutable() will inherently exclude ghp… which is what I want, because I never, ever want to edit those changes!

I also have a couple aliases for checking out the log of my history, which I usually reach for instead of jj log:

[aliases]
l = ["log", "-r", "@ | ancestors(immutable_heads().., 2) ~ ghp | trunk()"]
lall = ["log", "-r", "..all() ~ ghp"]

The first of these, l, is the same as the normal revset used by jj log, except that it excludes my ghp revset. The second is the “show me everything” version of jj log… again, except automatically excluding ghp. I do this often enough that it was worth making an alias instead of typing jj log -r "all() ~ ghp".

Between those couple of tweaks, I can now completely ignore the gh-pages branch/bookmark in all my normal use — but I can always get it back by asking for it directly, e.g. jj log -r ghp. Revsets are great!

---

Notes

1. We sometimes use GitHub Pages for deploying previews to get feedback. For example, I did that when getting early technical reviews for the new chapter I wrote on async and await. ↩︎

2. Right next to the wip alias I mentioned in my post about rebasing all branches matching a revset. ↩︎