It should, but due to a GitHub security limitation (preventing modifications to another repo from a different repo's GitHub Actions), it doesn't for external contributors.

However, I ran it for you here: #666

You can find the raw stats in the data.json file.

I'll give this more thought. I previously rejected config changes from:

• OXC: fix: align ecma target version to es2020 for esbuild, oxc, swc and terser #660
• All minifiers with target: Set minified output target for all tools, fix questionable results #417
• Google Closure: Fixing the google-closure-compiler advanced mode #14

But I've previously made exceptions for Terser and Uglify since their compression modes produce significantly different results.

The key issue is that there are endless configurations optimized for different inputs and we can only bench/compare so much. That said, most users just stick with the default settings, and I want these benchmarks to reflect the typical out-of-the-box experience.

I'll investigate the reason for the difference. Actually, the minifier is what I expected to be the default or the out-of-the-box thing for the SWC minifier, so there's a bug somewhere in the minify() of @swc/core.

I understand that you want to prevent the minifiers to game this benchmark by fine tuning the configs too much.

But you should definitely make sure that every minifier removes unused top-level declarations, and also removes comments. Otherwise it's either incorrectly configured or there is a bug in the minifier (as kdy was saying).

Disabling comments is definitely fine.

Looking into the top-level declarations one. Some minifiers like esbuild simply doesn't support that: evanw/esbuild#3570

Closing to cleanup

Were you able to reconcile the differences with https://github.com/swc-project/swc/blob/e862c329fc9af61414b664e81030976bde313983/crates/swc_ecma_minifier/scripts/size.sh ?

Things I found:

• There was a bug with the option handling in SWC. fix(es/minifier): Fix propagation of top level option swc-project/swc#10227
• zlib.gzipSync of node.js has a different output size than

fn gzip_size(data: &[u8]) -> usize {
    use flate2::{write::GzEncoder, Compression};
	use std::io::Write;

    let mut encoder = GzEncoder::new(Vec::new(), Compression::best());
    encoder.write_all(data).unwrap();
    encoder.finish().unwrap().len()
}

Guessing from the API of https://docs.rs/flate2/latest/flate2/write/index.html, I think zlib and gzip format is a different?
There are both GzEncoder and ZlibEncoder.

As a side note, GzEncoder matches the size of gzip -c -9