Prompt-induced command lag in large repos like https://github.com/nixos/nixpkgs is annoying.
Something like gitstatusd is an improvement, but it still felt slow and returns a lot of detail I don't use in my prompt. I wrote lilgit to see if I could trade detail for speed. (TL;DR: yes. See Performance for more.)
If you have Nix installed you can open a lilgit ~demo shell by running:
# if you have nix-command and flakes enabled:
nix develop github:abathur/lilgit
# otherwise, you can use either of the below:
nix develop github:abathur/lilgit --extra-experimental-features 'flakes nix-command'
nix-shell -E 'import (fetchGit { url="https://github.com/abathur/lilgit"; ref="main"; } + "/shell.nix")'Each time you run a command, it'll clearly indicate lilgit's output, and how long it took to generate.
Some of the speed comes from what I've left out, so I know it won't work for everyone. I'm happy to discuss cases where you think it is wrong or misleading (especially if we can make it more accurate without a large performance penalty).
It prints a "name", which is:
- blank if not in repo
- branch name if on branch
detached @ 11_chars_of_hashif detached
That name is:
- plain/uncolored if "clean"
- red if it meets my idiomatic sense of "dirty":
- latest local commit != latest upstream
- working copy differs from upstream
- working copy differs from HEAD
I assume it just works in bash for now, because the MVP depends on bash coproc.
It should work in other shells, and I'm open to help improving the daemonization model to pave the way :)
You'll two main things to get started:
- get the lilgit package
- prepare your rc/profile script:
source lilgit.bash- add
$__lilgitto yourPS1 - add
__go_off_now_lilgitto an EXIT trap
Let's look at each of these in a little more detail.
If you're using flakes, you can see how I pull lilgit into my [bashrc flake] (https://github.com/abathur/bashrc.nix/blob/main/flake.nix).
If you're using traditional Nix, you can use something like:
lilgit = (import (pkgs.fetchFromGitHub {
owner = "abathur";
repo = "lilgit";
rev = "v0.3.1";
hash = "sha256-vnK9LbhJHeSEiGA4sdWYrqTPvMEf9x0ykGulSyf2hYs=";
})).default;If you use resholve for your rc/profile script, you can just drop lilgit in the inputs as below:
pkgs.resholve.writeScript "trivial-test" {
# interpreter can be "none" after nixos/nixpkgs#210761
interpreter = "${pkgs.bash}/bin/bash";
inputs = [ lilgit ];
} ''
source lilgit.bash
export PS1="$__lilgit"
trap __go_off_now_lilgit EXIT
''You can find a complete real-world example of how I do this in https://github.com/abathur/bashrc.nix
If you don't use resholve, you'll want to do one of the following:
- add
lilgitto your system/user packages, and let your profile find lilgit viaPATHlookup - explicitly write/substitute the correct path to lilgit into your profile
It isn't a perfect picture, but you can find a benchmark table in each CI run's "performance" job. Here's an example:
test status-provider time footprint
---- ---- ---- ----
'1x clean @ master' lilgit.bash 315ms 2276KB
'1x clean @ master' gitstatus.bash 865ms 30MB
'1x clean @ master' gitparse.bash 1.0s 2924KB
---- ---- ---- ----
'1x clean @ detached' lilgit.bash 101ms 2276KB
'1x clean @ detached' gitstatus.bash 925ms 30MB
'1x clean @ detached' gitparse.bash 984ms 2888KB
---- ---- ---- ----
'1x clean @ new branch' lilgit.bash 218ms 2272KB
'1x clean @ new branch' gitstatus.bash 872ms 29MB
'1x clean @ new branch' gitparse.bash 352ms 2900KB
---- ---- ---- ----
'1x dirty after rm' lilgit.bash 259ms 2272KB
'1x dirty after rm' gitstatus.bash 803ms 30MB
'1x dirty after rm' gitparse.bash 334ms 2880KB
---- ---- ---- ----
'1x clean after reset --hard' lilgit.bash 251ms 2268KB
'1x clean after reset --hard' gitstatus.bash 845ms 30MB
'1x clean after reset --hard' gitparse.bash 350ms 2828KB
---- ---- ---- ----
'1x dirty after append' lilgit.bash 228ms 2284KB
'1x dirty after append' gitstatus.bash 824ms 30MB
'1x dirty after append' gitparse.bash 359ms 2916KB
---- ---- ---- ----
'1x dirty after stage' lilgit.bash 224ms 2276KB
'1x dirty after stage' gitstatus.bash 885ms 30MB
'1x dirty after stage' gitparse.bash 325ms 2908KB
---- ---- ---- ----
'1x clean after commit' lilgit.bash 193ms 2256KB
'1x clean after commit' gitstatus.bash 830ms 30MB
'1x clean after commit' gitparse.bash 285ms 2864KB
---- ---- ---- ----
'1x dirty after reset --soft w/o upstream' lilgit.bash 229ms 2276KB
'1x dirty after reset --soft w/o upstream' gitstatus.bash 800ms 30MB
'1x dirty after reset --soft w/o upstream' gitparse.bash 322ms 2912KB
---- ---- ---- ----
'1x clean after reset --hard w/o upstream' lilgit.bash 222ms 2264KB
'1x clean after reset --hard w/o upstream' gitstatus.bash 888ms 30MB
'1x clean after reset --hard w/o upstream' gitparse.bash 330ms 2940KB
---- ---- ---- ----
'1x dirty after reset --soft w/ upstream' lilgit.bash 102ms 2296KB
'1x dirty after reset --soft w/ upstream' gitstatus.bash 846ms 30MB
'1x dirty after reset --soft w/ upstream' gitparse.bash 339ms 2904KB
---- ---- ---- ----
'1x dirty after reset --hard w/ upstream' lilgit.bash 108ms 2288KB
'1x dirty after reset --hard w/ upstream' gitstatus.bash 872ms 30MB
'1x dirty after reset --hard w/ upstream' gitparse.bash 368ms 2884KB
test status-provider time footprint
---- ---- ---- ----
'10x clean @ master' lilgit.bash 1.28s 2428KB
'10x clean @ master' gitstatus.bash 2.62s 30MB
'10x clean @ master' gitparse.bash 8.59s 2900KB
---- ---- ---- ----
'10x clean @ detached' lilgit.bash 110ms 2288KB
'10x clean @ detached' gitstatus.bash 2.75s 30MB
'10x clean @ detached' gitparse.bash 2.77s 2916KB
---- ---- ---- ----
'10x clean @ new branch' lilgit.bash 1.33s 2368KB
'10x clean @ new branch' gitstatus.bash 2.71s 30MB
'10x clean @ new branch' gitparse.bash 2.23s 2964KB
---- ---- ---- ----
'10x dirty after rm' lilgit.bash 1.38s 2372KB
'10x dirty after rm' gitstatus.bash 2.69s 30MB
'10x dirty after rm' gitparse.bash 2.19s 2940KB
---- ---- ---- ----
'10x clean after reset --hard' lilgit.bash 1.38s 2384KB
'10x clean after reset --hard' gitstatus.bash 2.75s 30MB
'10x clean after reset --hard' gitparse.bash 2.19s 2948KB
---- ---- ---- ----
'10x dirty after append' lilgit.bash 1.33s 2380KB
'10x dirty after append' gitstatus.bash 2.57s 30MB
'10x dirty after append' gitparse.bash 1.98s 2892KB
---- ---- ---- ----
'10x dirty after stage' lilgit.bash 1.33s 2376KB
'10x dirty after stage' gitstatus.bash 2.52s 30MB
'10x dirty after stage' gitparse.bash 1.96s 2936KB
---- ---- ---- ----
'10x clean after commit' lilgit.bash 1.9s 2364KB
'10x clean after commit' gitstatus.bash 2.54s 30MB
'10x clean after commit' gitparse.bash 1.82s 2928KB
---- ---- ---- ----
'10x dirty after reset --soft w/o upstream' lilgit.bash 1.26s 2344KB
'10x dirty after reset --soft w/o upstream' gitstatus.bash 2.49s 30MB
'10x dirty after reset --soft w/o upstream' gitparse.bash 2.2s 2928KB
---- ---- ---- ----
'10x clean after reset --hard w/o upstream' lilgit.bash 1.29s 2356KB
'10x clean after reset --hard w/o upstream' gitstatus.bash 2.67s 29MB
'10x clean after reset --hard w/o upstream' gitparse.bash 2.13s 2928KB
---- ---- ---- ----
'10x dirty after reset --soft w/ upstream' lilgit.bash 112ms 2408KB
'10x dirty after reset --soft w/ upstream' gitstatus.bash 2.56s 30MB
'10x dirty after reset --soft w/ upstream' gitparse.bash 2.3s 2900KB
---- ---- ---- ----
'10x dirty after reset --hard w/ upstream' lilgit.bash 115ms 2408KB
'10x dirty after reset --hard w/ upstream' gitstatus.bash 2.48s 30MB
'10x dirty after reset --hard w/ upstream' gitparse.bash 2.7s 2932KB
test status-provider time footprint
---- ---- ---- ----
'100x clean @ master' lilgit.bash 10.9s 2436KB
'100x clean @ master' gitstatus.bash 19.0s 30MB
'100x clean @ master' gitparse.bash 1m22s 2980KB
---- ---- ---- ----
'100x clean @ detached' lilgit.bash 180ms 2296KB
'100x clean @ detached' gitstatus.bash 19.5s 30MB
'100x clean @ detached' gitparse.bash 19.5s 2928KB
---- ---- ---- ----
'100x clean @ new branch' lilgit.bash 11.7s 2404KB
'100x clean @ new branch' gitstatus.bash 14.0s 30MB
'100x clean @ new branch' gitparse.bash 18.5s 2980KB
---- ---- ---- ----
'100x dirty after rm' lilgit.bash 11.4s 2376KB
'100x dirty after rm' gitstatus.bash 13.9s 30MB
'100x dirty after rm' gitparse.bash 19.2s 2980KB
---- ---- ---- ----
'100x clean after reset --hard' lilgit.bash 11.7s 2392KB
'100x clean after reset --hard' gitstatus.bash 13.8s 30MB
'100x clean after reset --hard' gitparse.bash 18.6s 2964KB
---- ---- ---- ----
'100x dirty after append' lilgit.bash 11.5s 2376KB
'100x dirty after append' gitstatus.bash 13.9s 30MB
'100x dirty after append' gitparse.bash 18.4s 2988KB
---- ---- ---- ----
'100x dirty after stage' lilgit.bash 11.5s 2372KB
'100x dirty after stage' gitstatus.bash 13.8s 30MB
'100x dirty after stage' gitparse.bash 18.7s 2972KB
---- ---- ---- ----
'100x clean after commit' lilgit.bash 9.39s 2384KB
'100x clean after commit' gitstatus.bash 14.0s 30MB
'100x clean after commit' gitparse.bash 15.8s 2956KB
---- ---- ---- ----
'100x dirty after reset --soft w/o upstream' lilgit.bash 11.1s 2396KB
'100x dirty after reset --soft w/o upstream' gitstatus.bash 13.8s 30MB
'100x dirty after reset --soft w/o upstream' gitparse.bash 18.4s 2976KB
---- ---- ---- ----
'100x clean after reset --hard w/o upstream' lilgit.bash 11.3s 2400KB
'100x clean after reset --hard w/o upstream' gitstatus.bash 13.9s 30MB
'100x clean after reset --hard w/o upstream' gitparse.bash 18.6s 2972KB
---- ---- ---- ----
'100x dirty after reset --soft w/ upstream' lilgit.bash 208ms 2356KB
'100x dirty after reset --soft w/ upstream' gitstatus.bash 14.0s 30MB
'100x dirty after reset --soft w/ upstream' gitparse.bash 18.8s 2980KB
---- ---- ---- ----
'100x dirty after reset --hard w/ upstream' lilgit.bash 211ms 2376KB
'100x dirty after reset --hard w/ upstream' gitstatus.bash 13.9s 30MB
'100x dirty after reset --hard w/ upstream' gitparse.bash 18.7s 2972KB
Notes:
- Each test pays startup overhead once and runs the prompt-printing function the specified number of times. I used a few different counts so that it's easier to get a sense of the fixed startup costs and per-call performance.
- Measures memory use with the bsd
footprintcommand because it seems to do a good job of measuring the persistent marginal overhead of the daemons (which GNU time was missing). Both gitparse.bash and lilgit.bash can trigger git invocations likegit diff --quietandgit status, which take nearly 15MB against my local nixpkgs checkout according to GNU time. I'll be happy if anyone contributes a more accurate accounting. - The current difference in daemonization models for lilgit and gitstatus mean the reported numbers for both are a bit of a fiction relative to real world uses.
- lilgitd's per-instance use should be fairly stable--but it will currently run one instance per terminal. (See #2 for more on ~fixing this.)
- the single instance of gitstatusd can take up more memory as you use it against different repos.