Installation is 'slow'

[maleadt]

Now that there's support for nightlies, I see juliaup downloading updates much more than before, which has led me to realize that the process is relatively slow:

❯ time juliaup add nightly
Installing Julia latest-linux-x86_64
juliaup add nightly  2.90s user 2.36s system 25% cpu 20.490 total

vs.

❯ time wget https://julialangnightlies-s3.julialang.org/bin/linux/x86_64/julia-latest-linux-x86_64.tar.gz -O - | tar -xzf - -C $HOME
2024-02-27 13:59:40 (25.9 MB/s) - written to stdout [245088202/245088202]

wget  -O -  0.72s user 0.21s system 10% cpu 9.214 total
tar -xzf - -C $HOME  4.70s user 2.09s system 73% cpu 9.218 total

This is of course a minor issue, 20 seconds isn't terrible, but it's curious that it's twice as slow as wget + tar.

I'm using the official released version of juliaup. When working on the nightly support, I was using a debug build which was even slower, so maybe this is simply a matter of missing compiler optimizations?

[KristofferC]

Here is some sort of profile:

Maybe

juliaup/src/operations.rs

Lines 35 to 51 in 5967916

	fn unpack_sans_parent<R, P>(mut archive: Archive<R>, dst: P, levels_to_skip: usize) -> Result<()>
	where
	R: Read,
	P: AsRef<Path>,
	{
	for entry in archive.entries()? {
	let mut entry = entry?;
	let path: PathBuf = entry
	.path()?
	.components()
	.skip(levels_to_skip) // strip top-level directory
	.filter(|c| matches!(c, Normal(_))) // prevent traversal attacks TODO We should actually abort if we come across a non-standard path element
	.collect();
	entry.unpack(dst.as_ref().join(path))?;
	}
	Ok(())
	}

is slow?

[maleadt]

Interesting; I was just collecting flamegraphs too:

Here it seems like much more time is spent writing data.