FR: Make prioritization in jj log configurable

[jakobhellermann]

When rendering the commit graph in jj log, jj will prioritize rendering the working copy first, and then the remaining branches. This is a good starting point for a heuristic, but it's doesn't always result in the most readable graph.

Example 1: jj prev/next

When traversing a branch with these commands, prioritizing just @ will display the otherwise linear list of commits out of order, instead of 5 4 3 @ 2 1 it will be @ 5 4 3 2 1

@   ooss jakob.hellermann 2m ago main wip
◆  lmxr jakob.hellermann 2m ago commit 4
◆  lvzk jakob.hellermann 2m ago commit 3
◆  romu jakob.hellermann 2m ago commit 2
◆  pruv jakob.hellermann 3m ago commit 1
◆  zzzz root()

@  vywr jakob.hellermann 1s ago (empty) (no description set)
│ ◆  ooss jakob.hellermann 5m ago main wip
├─╯
◆  lmxr jakob.hellermann 5m ago commit 4
◆  lvzk jakob.hellermann 5m ago commit 3
◆  romu jakob.hellermann 5m ago commit 2
◆  pruv jakob.hellermann 5m ago commit 1
◆  zzzz root()

@  wrsz jakob.hellermann 27s ago (empty) (no description set)
│ ◆  ooss jakob.hellermann 5m ago main wip
│ ◆  lmxr jakob.hellermann 5m ago commit 4
├─╯
◆  lvzk jakob.hellermann 5m ago commit 3
◆  romu jakob.hellermann 5m ago commit 2
◆  pruv jakob.hellermann 5m ago commit 1
◆  zzzz root()

vs jj log --config 'revsets.graph-prioritize=main':

◆  ooss jakob.hellermann 7m ago main wip
◆  lmxr jakob.hellermann 7m ago commit 4
│ @  rvzp jakob.hellermann 15s ago (empty) (no description set)
├─╯
◆  lvzk jakob.hellermann 7m ago commit 3
◆  romu jakob.hellermann 7m ago commit 2
◆  pruv jakob.hellermann 7m ago commit 1
◆  zzzz root()

Example 2: Megamerge

Another situation where I noticed the graph being clearer with nonstandard priorization is when you have a "megamerge" commit of some local branches, and temporarily go work on another branch that's not

jj log

@  qmkx jakob.hellermann 3m ago push-mokuookxrmnu* cli: add dynamic completions for -T template aliases
│ ○  pqmm jakob.hellermann 4m ago push-pqmmuqxxzvmt* cli: completion: derive `clap::ValueEnum` for `ColorChoice`
├─╯
│ ○  squk jakob.hellermann 4m ago add templater size_hint
├─╯
│ ○  qlvu jakob.hellermann 4m ago push-qlvuxttmqltm* templater: add String.replace("from", "to") method
├─╯
│ ○    utmp jakob.hellermann 21m ago (empty) dev
│ ├─╮
│ │ ○  yomz jakob.hellermann 55m ago jiff-duration use jiff for duration formatting
├───╯
│ ○  qxnv jakob.hellermann 55m ago revset-prioritize* add revsets.graph-prioritize
├─╯
◆  xrnl bsdinis 2h ago main git_head() git: describe codepaths that require git2
◆  mvvt bsdinis 2h ago cli: relocate git fetch code to cli/src/commands/git/fetch.rs
◆  zypt bsdinis 2h ago git: update jj git clone|fetch to use new GitFetch api directly.
│
~

jj log --config 'revsets.graph-prioritize="description(\"dev\")"'

○    utmp jakob.hellermann 22m ago (empty) dev
├─╮
│ ○  yomz jakob.hellermann 56m ago jiff-duration use jiff for duration formatting
○ │  qxnv jakob.hellermann 56m ago revset-prioritize* add revsets.graph-prioritize
├─╯
│ @  qmkx jakob.hellermann 4m ago push-mokuookxrmnu* cli: add dynamic completions for -T template aliases
├─╯
│ ○  pqmm jakob.hellermann 5m ago push-pqmmuqxxzvmt* cli: completion: derive `clap::ValueEnum` for `ColorChoice`
├─╯
│ ○  squk jakob.hellermann 5m ago add templater size_hint
├─╯
│ ○  qlvu jakob.hellermann 5m ago push-qlvuxttmqltm* templater: add String.replace("from", "to") method
├─╯
◆  xrnl bsdinis 2h ago main git_head() git: describe codepaths that require git2
◆  mvvt bsdinis 2h ago cli: relocate git fetch code to cli/src/commands/git/fetch.rs
◆  zypt bsdinis 2h ago git: update jj git clone|fetch to use new GitFetch api directly.
│
~

Solution

I've experimented with adding a graph-prioritize revset expression and changing the priorization logic to

for prio in prio_revset {
  if log_revset.contains(prio) {
    graph.prioritize_branch(prio)
  }
}
graph.prioritize_branch(@)

which is already usable with

[revsets]
graph-prioritize = "main | description('dev')"

The one thing I don't like about this solution is that it exposes an ordering of revsets, which I don't think is meant to be relied on.