feat: add benchmark for displaying all programs in PATH

[vegerot]

Summary:
This is the first regression we've seen for fd vs. find.

$ PATH="/opt/homebrew/Cellar/findutils/4.10.0/libexec/gnubin/:$PATH" ./warm-cache-exe-paths.sh
Benchmark 1: FIND_PROG=find FIND_ARGS='-maxdepth 1 -executable -type f,l -printf %f\n' get_programs_in_path
  Time (mean ± σ):     142.0 ms ±   6.0 ms    [User: 32.7 ms, System: 72.0 ms]
  Range (min … max):   129.7 ms … 154.6 ms    20 runs

Benchmark 2: FIND_PROG='fd .' FIND_ARGS='--hidden --max-depth=1 --type=executable --follow --format {/} ' get_programs_in_path
  Time (mean ± σ):     272.1 ms ±   6.0 ms    [User: 88.8 ms, System: 142.1 ms]
  Range (min … max):   258.2 ms … 279.9 ms    10 runs

Summary
  FIND_PROG=find FIND_ARGS='-maxdepth 1 -executable -type f,l -printf %f\n' get_programs_in_path ran
    1.92 ± 0.09 times faster than FIND_PROG='fd .' FIND_ARGS='--hidden --max-depth=1 --type=executable --follow --format {/} ' get_programs_in_path
Both fd and find found the same 2943 results

[vegerot]

The commit description was measured on macOS. On my GNU+Linux machine it's closer to 7x slower

$ ./warm-cache-exe-paths.sh
Benchmark 1: FIND_PROG='find -L' FIND_ARGS='-maxdepth 1 -executable -type f,l -printf %f\n' get_programs_in_path
  Time (mean ± σ):      40.6 ms ±   0.5 ms    [User: 20.5 ms, System: 24.7 ms]
  Range (min … max):    39.4 ms …  42.1 ms    71 runs

Benchmark 2: FIND_PROG='fd .' FIND_ARGS='--hidden --max-depth=1 --type=executable --follow --format {/} ' get_programs_in_path
  Time (mean ± σ):     281.2 ms ±   4.3 ms    [User: 54.0 ms, System: 397.8 ms]
  Range (min … max):   276.5 ms … 288.7 ms    10 runs

Summary
  FIND_PROG='find -L' FIND_ARGS='-maxdepth 1 -executable -type f,l -printf %f\n' get_programs_in_path ran
    6.93 ± 0.13 times faster than FIND_PROG='fd .' FIND_ARGS='--hidden --max-depth=1 --type=executable --follow --format {/} ' get_programs_in_path
Both fd and find found the same 3806 results

[tavianator]

See sharkdp/fd#1614 (comment) and below for some analysis. I don't think this benchmark is useful as-is, but it may be interesting to add benchmarks for wide, shallow directories as well as empty directories to measure startup overhead.