Benchmark results of the fastest way to check if a git branch is dirty

git-dirty-checks.md

Tested against the WebKit git repo by entering the repo with 1 file dirty.

---

git diff --quiet --ignore-submodules HEAD # Will tell if there are any uncomitted changes, staged or not.
0.6 sec

git diff-index --quiet HEAD # Only tracked
2 sec

git diff --shortstat
8.2 sec

git status --porcelain
42 sec

zstyle ':vcs_info:*' check-for-changes true
50 sec

git diff --quiet --ignore-submodules HEAD doesn't show the untracked files so it might cause some undesired accidents.

True, but the only git status --percelain will do that, and it's sloooow.

@sindresorhus Doesn’t git status without --porcelain do it too, kind of? That’s what I am using at the moment in my dotfiles: https://github.com/mathiasbynens/dotfiles/blob/9aecb166d67e0119472aceb8bba71b06256929d8/.bash_prompt#L46-L48 (credit to that code goes to @gf3; I stole it from him)

It might be slow, though.

Because it's related... this appears to be the fastest way to check the current branch (used by vcs_info and others)

git symbolic-ref -q HEAD | sed -e 's|^refs/heads/||'

@paulirish My (non-scientific) tests show that not using the -q or --quiet option is the fastest, but then you'll probably want to redirect error messages to /dev/null/.

@paulirish Add the --short option and you can get rid of sed:

git symbolic-ref --short HEAD

Did you try git status -suno? It's working faster for me and have the same limitations.
Full version is here: https://github.com/artjock/dotfiles/blob/master/bash/config#L20

Unless you want to take untracked files into account too. The best I've come up with is:

git diff-files --no-ext-diff --quiet || git diff-index --no-ext-diff --quiet --cached HEAD

This is the benchmark I got in a NFS filesystem with the WebKit repo clean:

Command	Real time (secs)
git ls-files --other --exclude-standard (lists untracked files)	2.756
git status --porcelain (lists indexed, unindexed and untracked files)	2.313
git diff-index --no-ext-diff --quiet --cached HEAD (checks indexed files)	0.317
git diff-files --no-ext-diff --quiet (checks unindexed files)	0.022

git diff-index --quiet HEAD -- can return outdated information unless you run git update-index -q --refresh (or a non-plumbing git status) first, (see https://stackoverflow.com/questions/2657935/checking-for-a-dirty-index-or-untracked-files-with-git/2659808#comment28747279_2659808) so it's not a fair comparison vs git status --porcelain without that.

Also any use of git diff-index --no-ext-diff --quiet --cached HEAD breaks if there's a file called HEAD in the repo; you need to use git diff-index --no-ext-diff --quiet --cached HEAD --, see here.

Finally, I found plumbing commands to be slower than the porcelain commands when the goal is to ignore untracked files in a repo with many large untracked files (not WebKit in this case) with git 2.29.2:

• time (git diff-files --quiet --no-ext-diff || (git update-index -q --refresh; git diff-index --quiet --no-ext-diff --cached HEAD --))
  • 255 ms mean
• time git status --porcelain --untracked-files=no > /dev/null
  • 110 ms mean

This was also observed here.

Do there exist any plumbing commands that are as fast and as correct as the porcelain commands? (Also asked here.)