Revset recipes

[arxanas]

git-branchless v0.4.0 supports revset expressions for many commands. Here's a few examples of useful things you can do with revsets. Feel free to add your own recipes in the comments.

Move all other draft commits on top of this commit

$ git move -x ‘draft() - stack()’

Hide commits generated by imerge

As per #380.

$ git hide ‘draft() & message(imerge)’

“Drop” commits rather than hide them

As per #48 (reply in thread).

$ git move -x “children($1)” -d “$1^” && git hide “$1”

Insert a commit into a stack:

As per #478.

$ git move -I -x “$1” -d “$2”

Separate all commits in this stack for a given ticket into their own stack

This is kind of like git-branchstack.

$ git move -x ‘stack() & message(JIRA-123)’ -d origin/master

Move all commits touching a class of files to the top of this stack

$ git move -I -x ‘stack() & paths.changed("glob:*.txt")' -d 'roots(stack())'

Force push all branches in the current stack

As per #45 (reply in thread).

$ git push --force-with-lease origin $(git query -b 'branches() & stack()')

Note that this can be accomplished with the git submit command, in master at the time of this writing.

Find out about recently-upgraded dependencies, excluding Dependabot commits

$ git query 'paths.changed("Cargo.lock") & author.date("after:1 month ago") - author.email(dependabot)'

Hide any commits in the current stack which aren’t on a branch

$ git hide 'stack() - ancestors(branches())'

Delete any draft branches which weren’t created by me

$ git hide -D ‘stack(draft() & branches() - author.name(Waleed))'

[rouge8]

Is it possible to query for commits that aren't on any branch? I tried git query 'draft() - branches()' but that looks like all commits that aren't the tip of branches. For context, what I'm trying to do is hide all commits that aren't on a branch so that they can be gc'd. I came up with this alias, but I wonder if there's a way to express it as a revset: hide-branchless = "!f() { git query --raw 'draft()' | while read commit; do if [ -z $(git branch --contains $commit) ]; then git hide $commit; fi; done; }; f".

[arxanas]

Yep, that should be like the "Hide any commits in the current stack which aren’t on a branch" example. You just need to write ancestors(branches()) instead of branches(). (Also, you can use :expr as a shorthand for ancestors(expr), which is more convenient for interactive use.)

[h4l]

This is a nice feature. I needed to run tests for each commit in my current WIP chain of commits to make sure none of the intermediate commits are in a broken state. Let me know if there's a more idiomatic way to do this, but I came up with this:

# xargs stops immediately if a command exits with 255
$ git query -r 'range(51d3d1f, HEAD)' | xargs -I{} sh -c 'git checkout {} && pytest -x || exit 255'

Could easily tweak the query to use this approach to run something for all draft commits rather than just the immediate ancestors.

[h4l]

@arxanas Is the order of commits from query defined/consistent? Seems like its ordered such that a commit appears before its parent in the in the output (like git-log, or the opposite of smartlog).

I'm wondering because with a workflow where I'm linting files in a stack for issues, I want to find the first commit that introduces a problem, so that I can fix it at source. So it seems like I can just do git query -r 'xxx' | tac to get the commits I want with "older"/ancestor commits first.

Also, something I've found useful is to use a revset expression with smartlog to filter its output. Doing that is a good way to experiment with expressions, as it's easy to compare the filtered output with the default output to work out what your query is doing.

[claytonrcarter]

Is the order of commits from query defined/consistent?

I asked a similar question in discord last month and @arxanas answer suggests that the order may be consistent but is not defined. I briefly played with adding a sort() fn to revsets but didn't really get anywhere and gave for lack of enthusiasm.

[h4l]

Thanks @claytonrcarter, that's useful. I guess I'll quote what @arxanas said here for anyone not in Discord (you have to join the server to view your link):

arxanas — 16/08/2022
Mercurial has a sort function as well, so maybe. I didn't add anything like first() because I didn't think it was well-defined what order the commits are in; like if you do a & b and some x appears in both sets, would you expect to be in the order as if a were traversed first or b? (Some sets are optimized for " contains" checks; at the moment, we evaluate expressions in order, but a smart query planner might evaluate them in some arbitrary other order)

[arxanas]

@h4l The intermediate results of the revsets are not sorted or guaranteed to be in a particular order, but git query always sorts its output topologically. I've updated https://github.com/arxanas/git-branchless/wiki/Command:-git-query#use-in-scripts and opened #527 to clarify this. If we added the ability for commit sets to be ordered/sorted as part of the revset language itself, then we would want to also add functionality to git query to preserve that ordering.

In principle, the results of evaluating the revset when passed to other functions (git move, git hide, etc.) is not defined, but it doesn't matter for any of those.

[h4l]

Perfect, thanks @arxanas!

[pokey]

Wow these are epic! A couple quick typos in Force push all branches in the current stack: it uses a bad quote and is missing a closing paren. It should be:

$ git push --force-with-lease origin $(git query -b 'branches() & stack()')

[arxanas]

Thanks for the correction. I've updated the post. By the way, you can use git submit on the latest master to do the same thing a little more conveniently.

[pokey]

Oh nice! I'll have a look at that one. The --create flag looks useful

[pokey]

For the git-imerge one, I think the following is a bit more precise, as your query could sometimes pick things up that end up in a branch, and misses some things as well:

$ git hide 'descendants(draft() & message("regex:^Merge commit|^imerge") - ancestors(branches()))'