gosh benchmarks?

[cilki]

Has anyone ever compared the performance of gosh to something like bash or dash?

[mvdan]

There is one benchmark inside the interp package:

name    time/op
Run-16  3.51ms ± 0%

name    alloc/op
Run-16   157kB ± 1%

name    allocs/op
Run-16     876 ± 0%

It essentially benchmarks a small program with a mix of variables, commands, expansions, and external programs (i.e. cat):

echo a b c d
echo ./$foo_interp_missing/etc $(echo foo_interp_missing bar_interp_missing)
foo_interp_missing="bar_interp_missing"
x=y :
fn() {
	local a=b
	for i in 1 2 3; do
		echo $i | cat
	done
}
[[ $foo_interp_missing == bar_interp_missing ]] && fn
echo a{b,c}d *.go
let i=(2 + 3)

The summary is that the native Go bits of the interpreter are fast enough for its use cases so far. Right now, the slowest part of the shell by far is using os/exec to call separate programs like cat above; each one of those costs milliseconds, when the shell parsing and interpreting in Go is in the order of microseconds.

But also, the performance will heavily depend on your use case. If you only need to run a small shell script, then it's likely that this interpreter will be faster than Bash for you from Go, as you'll avoid having to call Bash as a separate program - thus saving milliseconds. If you need to run a very complex and long shell script, then it's entirely possible that shells like Bash might be faster, simply because they might have optimized some edge cases that we haven't yet. You should measure and compare for your use case :)

[cilki]

Thanks, I will do. I work with quite a few fairly complex scripts on a resource constrained platform, so I'm sensitive to performance and, of course, correctness.

I'm really interested in what kind of impact garbage collection has if any. I'll try it out and report real numbers if I find anything worthwhile.

[mvdan]

Let me know how you get on! If you do find that a script runs surprisingly slowly, perhaps you can share it so I can grab some CPU or memory profiles and optimize as needed.

Correctness is generally decent, though sometimes there are edge cases that we haven't implemented exactly the same as Bash - for instance, see #886. People have been using this interpreter for years in projects like go-task though, so my best estimate is that your scripts will just work unless they hit some particularly rare edge cases.