feat(es/minifier): Implement minifier

[kdy1]

Postponed

To simplify management, I'll merge this PR as incomplete state and implement more rules with follow-up PRs.

swc_bundler:

• Add option for minification.

swc_ecma_minifier:

• Implement.
• Provide js api.

[soilSpoon]

Are you thinking of implementing it using UglifyJS?

[kdy1]

@soilSpoon Yes, but mainly because I don't know well about internals of minifiers.
Is there something better than uglifyjs?

Or I can add some passes after implementing it using uglifyjs.

[soilSpoon]

uglifyjs doesn't seem to be able to minify ECMAScript 2015 or above. So, to use this, you need to use it after transpile it to ECMAScript 2015 or lower. I think this can adversely affect performance. So I look better to use terser that can minify ECMAScript 2015 or above.

It seems good to understand minify in the future and implement it yourself.

Reference

terser vs uglifyjs
Benchmarks
terser ESTree/SpiderMonkey AST

[kdy1]

@soilSpoon Oh. I got it. Thanks!
I'll port terser instead.

[divmain]

You may also want to look at babel-minify for inspiration. From what I recall, it doesn't achieve quite the same size wins as Uglify or Terser. But the port may be significantly more straightforward - the minifier is split into discrete operations for the different types of transformations that it applies.

[pickfire]

Typo

Suggested change

	pub(super) fn optimize_in_fn_termiation(&mut self, e: &mut Expr) {
	pub(super) fn optimize_in_fn_termination(&mut self, e: &mut Expr) {

[kdy1]

Oh, thanks!

[fabiosantoscode]

Hey, I maintain Terser, and it's helplessly slow. Recently I started a port to rust (in a very early stage) but I don't think it's useful to keep two separate projects that are essentially the same. How can I help?

[kdy1]

@fabiosantoscode I implemented many rules but there are many rules left.
If you don't mind implementing it as a part of swc, I will add lots of comment how the minifier works and about some design decisions.

Note that being in-tree would make version management much simpler, as bundler depends on minifier and minifier depends on ast / visitor while those crates live in a momorepo.

[fabiosantoscode]

Ok, I'll go on your fork and send PRs when I have the time.

I'm interested in how to implement reduce_vars, unused, and inline, because in Terser they're kind of the same for speed reasons. They're also very relevant for tree shaking.

[Austaras]

Any possible to expose as webpack plugin like terser

[kdy1]

Eventually I'd like to test this against every single Terser test. Not all of them require exact output. Some of them run the resulting code using node's VM module and compare the result.

👍 That's also my goal.
Terser has strength in the output and I want to make minification faster without degrading output. (while migrating)

[SoloJiang]

However, I see this PR had beed merged, but how can I try it out? At now, it only removes comment and space in my code.

[kdy1]

@SoloJiang It's not published yet because it's not done

[fabiosantoscode]

So much excitement for this port, I love it 💖

I need to try to reduce my hours in my side gig so I can give this more attention.

There's some transforms Terser itself doesn't do that are feasible here (such as a dictionary for common immutable constants or cross-module intelligence used in combination with spack) that can reduce this port's output bytes to a point that makes it a viable alternative to Terser.

[kdy1]

I'm glad to hear that and thank you!

There's some transforms Terser itself doesn't do that are feasible here (such as a dictionary for common immutable constants or cross-module intelligence used in combination with spack) that can reduce this port's output bytes to a point that makes it a viable alternative to Terser.

Sounds interesting. I expect such cross-module transforms to reduce final output by a margin because libraries used a lot along with UI frameworks like react encourages such code.