I use changesets to manage releases in the main repositories I use at work. (I’ve written before about how I prefer it over semantic commits)

I find that, if I’ve written a clear description of my change for the changeset, I probably want to use the same thing for my commit message. I had been doing this manually, then moved to a shell pipeline that I’d find in my history and re-use, but today I decided to turn it into something a little more polished. Here’s the shell function in its entirety, and then I’ll explain how it works:

ccs() {
    local changeset_files
    changeset_files=$(
      git status -s |
      awk '{ if ($1 == "A") print $2 }' |
      grep .changeset
    )

    if [[ -n "$changeset_files" ]]; then
        echo "$changeset_files" | \
        xargs -I {} sed '1,/^$/d' "{}" | \
        git commit -F - --edit
    else
        echo "No changeset files found to commit."
        echo "Perhaps you forgot to git add them?"
        exit 1
    fi
}

First, we use git status -s to find a list of files that are changed in our working tree. I use awk to filter only for files that are added “A”, as opposed to modified, deleted, or untracked, then print only the filename ($2). Those are further filtered to only the files whose path includes “.changeset”.

Next, we need to reformat the files a bit. Changeset files have yaml frontmatter that describes which packages were changed, and whether it was a major, minor, or patch change. I didn’t want that in the commit message. set '1,/^$/d' says “delete the range of lines starting at 1 and ending with the first empty line”. This removes the frontmatter, leaving just the description of the change.

The xargs business before the sed command is there to make sure we delete the frontmatter for each changeset file found, not just the first one.

Finally, we pipe that into git commit -F - --edit. The -F - flag tells git to read the commit message from standard input, and --edit opens the commit message in your editor so you can make any final adjustments before committing. This is especially useful if there were multiple changeset files.

I initially had this all in one big pipeline, but it didn’t work very well. git status and git commit both want to hold .git/index.lock, and I got an error if git commit started when git status was still running. Capturing the changeset files in a variable ensures that git status is done before we try to commit.

I’ve added this as an alias to my gitconfig. I used ccs to stand for “commit changeset”. Here’s what it looks like in the config file, with all the necessary escaping.

# ~/.gitconfig
[alias]
    ccs = "!f() { local changeset_files; changeset_files=$(git status -s | awk '{ if ($1 == \"A\") print $2 }' | grep .changeset); if [[ -n \"$changeset_files\" ]]; then echo \"$changeset_files\" | xargs -I {} sed '1,/^$/d' \"{}\" | git commit -F - --edit; else echo \"No changeset files found to commit. Perhaps you need to git add them?\" && exit 1; fi; }; f"