Simplify git dirty check using git status

[jesses-canva]

In large repositories, checking for untracked files is very slow, so status.showUntrackedFiles=false is typically set to improve git status performance.

_pure_prompt_git_dirty doesn't use git status however, but three separate git commands to check for staged changes, unstaged changes and untracked files. The command for untracked files is git ls-files --others which doesn't respect status.showUntrackedFiles=false.

Instead we can check for non-empty output from git status --porcelain, which checks all three of these things in one command and respects status.showUntrackedFiles=false.

This reduces _pure_prompt_git_dirtry runtime from 9.3s to 0.8s on our repository.

[edouard-lopez]

Thanks,
the git commands were pretty old, and we haven't reviewed them in a long time.

Performance increase sound huge!
How does that impact smaller repos when used without status.showUntrackedFiles=false?

Could you update the doc by adding:
1.a quick reference to the Git section about big repo to send back to the Troubleshooting section, e.g. using Mkdocs admonition.
2. add an entry to the Troubleshooting component to add your remark about large repo

[jesses-canva]

How does that impact smaller repos when used without status.showUntrackedFiles=false?

Using this repo as an example, it is a bit faster, probably just because its 1 command instead of (up to) 3, but the data is noisy:

$ git config --global status.showUntrackedFiles true

$ cat test.fish
#!/usr/bin/env fish
set --prepend fish_function_path (path dirname (status --current-filename))/functions
_pure_prompt_git_dirty

$ hyperfine \
  --warmup 10 \
  -n master \
  --prepare 'git checkout master' \
  'fish ./test.fish' \
  -n patch-1 \
  --prepare 'git checkout patch-1' \
  'fish ./test.fish'

Benchmark 1: master
  Time (mean ± σ):     253.5 ms ± 197.9 ms    [User: 52.4 ms, System: 60.4 ms]
  Range (min … max):   154.2 ms … 718.7 ms    14 runs

  Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet system without any interferences from other programs.

Benchmark 2: patch-1
  Time (mean ± σ):     163.3 ms ± 163.8 ms    [User: 40.2 ms, System: 46.0 ms]
  Range (min … max):   109.2 ms … 776.8 ms    16 runs

  Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet system without any interferences from other programs.

Summary
  patch-1 ran
    1.55 ± 1.97 times faster than master