Looks like cat is slower than its GNU counterpart

[khaalidrafi]

I tried cat on the following file, prepended with time. And, according to the results, uu-cat is slower than than cat.

emacs.md

Results:

$ time cat emacs.md

________________________________________________________
Executed in  149.81 millis    fish           external
   usr time    0.28 millis    0.28 millis    0.00 millis
   sys time    5.85 millis    1.02 millis    4.82 millis

$ time uu-cat emacs.md

________________________________________________________
Executed in  454.49 millis    fish           external
usr time    4.02 millis    0.00 millis    4.02 millis
sys time   21.83 millis    1.85 millis   19.98 millis

[oech3]

#9609 (comment)

slightly worse performance, esp. on smaller files (arguably most common use case for cat?), esp. due to more syscalls. (Many more syscalls overall in uutils than in GNU coreutils btw.) Esp. on FUSE. It always uses its read() / write() userspace API anyway.

[sylvestre]

note: please run hyperfine for benchmarks. time is too limited

[sylvestre]

also, you didn't say which versions you have been using

[oech3]

The simple clap bypassif args.len() > 1 && args.iter().skip(1).all(|a| !a.to_string_lossy().starts_with('-') || a == "-")
caused 1.12 ± 0.21 times faster on hyperfine -N --warmup 5 (still slower than GNU).

[FidelSch]

Looking into this, seems to have to do with the overhead of using splice. For a small enough file, results in worse performance.

[oech3]

splice() was not overhead and it was actually faster than write for small input.
But we had pipe() even output is not pipe to use splice(). Removing it improved few micro sec at #11551 .
We don't avoid pipe() for file input and output to terminal which is most usecase.

[oech3]

@khaalidrafi Is this issue still active for you

I noticed that I was not avoiding fcntl() for small input, but it is bit difficult to avoid naturally.

[oech3]

after 49e7a6b

$ hyperfine -N "cat emacs.md" "gnu9.10cat emacs.md"
Benchmark 1: cat emacs.md
  Time (mean ± σ):       1.2 ms ±   0.3 ms    [User: 0.4 ms, System: 0.7 ms]
  Range (min … max):     0.9 ms …   7.0 ms    426 runs
 
Benchmark 2: gnu9.10cat emacs.md
  Time (mean ± σ):     797.6 µs ± 171.1 µs    [User: 305.9 µs, System: 376.7 µs]
  Range (min … max):   546.0 µs … 3055.6 µs    2135 runs
  
Summary
  gnu9.10cat emacs.md ran
    1.47 ± 0.51 times faster than cat emacs.md

[sylvestre]

given that we are around the millisecond / µs, on small file, we are close enough now IMHO

[oech3]

fcntl() and pipe() overhead were less than 1 µs as CodSpeed. So this can be closed in flavor of localization setup issue.