Update benchmark versions

[rtsao]

Parcel v2.10.0 was just released and appears to significantly improve build performance.

This PR updates all bundler versions to latest. Notably, Parcel v2.10.0 now runs the three benchmark in around 7 seconds on my machine.

⏱️ `make bench-three` log on my M1 Max 64GB MacBook Pro:

❯ make bench-three
node scripts/esbuild.js --update-version-go
CGO_ENABLED=0 go build "-ldflags=-s -w" -trimpath ./cmd/esbuild
rm -fr bench/three/esbuild
time -p ./esbuild --bundle --global-name=THREE --sourcemap --minify bench/three/src/entry.js --outfile=bench/three/esbuild/entry.esbuild.js --timing
▶ [INFO] Timing information (times may not nest hierarchically due to parallelism)

  Scan phase: 241ms
    On-start callbacks: 0ms
    Preprocess injected files: 0ms
    Add entry points: 2ms
    Scan all dependencies: 237ms
    Process scanned files: 1ms
  Compile phase: 114ms
    Spawn source map tasks: 6ms
    Link: 106ms
      Clone linker graph: 12ms
      Scan imports and exports: 16ms
        Step 1: 0ms
        Step 2: 0ms
        Step 3: 0ms
        Step 4: 5ms
        Step 5: 4ms
        Step 6: 4ms
      Tree shaking: 1ms
      Code splitting: 1ms
      Compute chunks: 4ms
      Compute cross-chunk dependencies: 0ms
      Waiting for mangle cache: 0ms
      Mangle props: 0ms
      Mangle local CSS: 0ms
      Generate chunks: 69ms
        Generate chunk "entry.esbuild.js": 67ms
          Minify symbols: 12ms
            Compute reserved names: 1ms
            Accumulate symbol counts: 7ms
              Parallel phase: 4ms
              Serial phase: 3ms
            Assign names by frequency: 2ms
          Print JavaScript files: 38ms
          Join JavaScript files: 2ms
          Generate source map: 13ms
        Generate final output files: 1ms
  Write output files: 3ms
  On-end callbacks: 0ms


  bench/three/esbuild/entry.esbuild.js       5.8mb ⚠️
  bench/three/esbuild/entry.esbuild.js.map  19.7mb

⚡ Done in 362ms
real 0.78
user 1.18
sys 0.42
du -h bench/three/esbuild/entry.esbuild.js*
5.8M	bench/three/esbuild/entry.esbuild.js
 20M	bench/three/esbuild/entry.esbuild.js.map
shasum bench/three/esbuild/entry.esbuild.js*
1fab8f5660f49850edec482120604ce8645433b6  bench/three/esbuild/entry.esbuild.js
15282d14da3af1d8cc1038f2d18bd555e9322658  bench/three/esbuild/entry.esbuild.js.map
rm -fr require/rollup/bench/three bench/three/rollup
mkdir -p require/rollup/bench/three bench/three/rollup
echo "import terser from '@rollup/plugin-terser'; export default { output: { format: 'iife', name: 'THREE', sourcemap: true }, plugins: [terser()], }" > require/rollup/bench/three/config.mjs
ln -s ../../../../bench/three/src require/rollup/bench/three/src
ln -s ../../../../bench/three/rollup require/rollup/bench/three/out
cd require/rollup/bench/three && time -p ../../node_modules/.bin/rollup src/entry.js -o out/entry.rollup.js -c config.mjs

src/entry.js → out/entry.rollup.js...
(!) Circular dependencies
src/copy1/renderers/WebGLRenderTargetCube.js -> src/copy1/cameras/CubeCamera.js -> src/copy1/renderers/WebGLRenderTargetCube.js
src/copy2/renderers/WebGLRenderTargetCube.js -> src/copy2/cameras/CubeCamera.js -> src/copy2/renderers/WebGLRenderTargetCube.js
src/copy3/renderers/WebGLRenderTargetCube.js -> src/copy3/cameras/CubeCamera.js -> src/copy3/renderers/WebGLRenderTargetCube.js
...and 7 more
created out/entry.rollup.js in 22s
real 22.38
user 33.41
sys 4.46
du -h bench/three/rollup/entry.rollup.js*
5.8M	bench/three/rollup/entry.rollup.js
 19M	bench/three/rollup/entry.rollup.js.map
rm -fr require/webpack5/bench/three bench/three/webpack5
mkdir -p require/webpack5/bench/three bench/three/webpack5
ln -s ../../../../bench/three/src require/webpack5/bench/three/src
ln -s ../../../../bench/three/webpack5 require/webpack5/bench/three/out
cd require/webpack5/bench/three && time -p ../../node_modules/.bin/webpack --entry ./src/entry.js --devtool=source-map --mode=production --output-library THREE -o out/entry.webpack5.js
asset main.js 5.84 MiB [emitted] [minimized] [big] (name: main) 2 related assets
orphan modules 11.4 MiB [orphan] 3680 modules
runtime modules 670 bytes 3 modules
cacheable modules 11.5 MiB
  ../../../../bench/three/src/entry.js + 3680 modules 11.4 MiB [built] [code generated]
  ../../../../bench/three/src/copy1/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy2/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy3/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy4/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy5/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy6/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy7/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy8/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy9/polyfills.js 1.65 KiB [built] [code generated]
  ../../../../bench/three/src/copy10/polyfills.js 1.65 KiB [built] [code generated]

WARNING in asset size limit: The following asset(s) exceed the recommended size limit (244 KiB).
This can impact web performance.
Assets:
  main.js (5.84 MiB)

WARNING in entrypoint size limit: The following entrypoint(s) combined asset size exceeds the recommended limit (244 KiB). This can impact web performance.
Entrypoints:
  main (5.84 MiB)
      main.js


WARNING in webpack performance recommendations:
You can limit the size of your bundles by using import() or require.ensure to lazy load some parts of your application.
For more info visit https://webpack.js.org/guides/code-splitting/

webpack 5.88.2 compiled with 3 warnings in 26977 ms
real 27.54
user 38.12
sys 2.49
du -h bench/three/webpack5/entry.webpack5.js*
 25M	bench/three/webpack5/entry.webpack5.js
rm -fr require/parcel2/bench/three bench/three/parcel2
mkdir -p require/parcel2/bench/three bench/three/parcel2
# Copy the whole source tree since symlinks mess up Parcel's internal package lookup for "@babel/core"
cp -r bench/three/src require/parcel2/bench/three/src
echo 'import * as THREE from "./src/entry.js"; window.THREE = THREE' > require/parcel2/bench/three/entry.parcel2.js
cd require/parcel2/bench/three && time -p node ../../node_modules/.bin/parcel build \
		entry.parcel2.js --dist-dir ../../../../bench/three/parcel2 --cache-dir .cache
✨ Built in 7.06s

../../../../bench/three/parcel2/entry.parcel2.js    ⚠️  6.64 MB    3.75s
real 7.51
user 14.19
sys 2.05
du -h bench/three/parcel2/entry.parcel2.js*
6.6M	bench/three/parcel2/entry.parcel2.js
 34M	bench/three/parcel2/entry.parcel2.js.map

[evanw]

Cool! It's great to see everyone's hard work on performance paying off. I'm a little too busy right now but I'm going to try to get some time to re-run the benchmarks this weekend (it takes a good bit of time because I restart my computer in between runs to make sure the file system cache is clean).

[evanw]

I took a look at this. My findings:

1. Webpack 5 and Rollup 4 and esbuild output sizes and build times don't change significantly.

2. Parcel 2 is faster but I get build times around 15 seconds, not 7 seconds. Presumably because the benchmark machine isn't a M1 Max 64GB MacBook Pro.

3. Parcel 2's output is much larger because the new minifier strangely preserves all comments (not just license-related comments). Disabling this requires an extra config file, which I can add to the benchmark. This seems like a Parcel 2 bug to me.

4. Parcel 2 can handle the JavaScript benchmark but not the TypeScript benchmark, which is kind of a deal-breaker. It looks like bundling for node (specifically engines: node in package.json) disables interpretation of aliases (specifically alias in package.json) which is required for bundling this code base. I see this regression as a bug with Parcel 2's new resolver. This is a bug that esbuild had too, before I fixed it in esbuild: The module imported with the path alias specified in paths in tsconfig.json is not being bundled #3160.

   Edit: I found Parcel fails to bundle require('node:stream/web') used in node-fetch: gets transformed to import "8a78abd0b0f7c0d4:stream/web"; parcel-bundler/parcel#7387 (comment). This scenario requires includeNodeModules: true to be present for it to work. Now the TypeScript benchmark builds and runs successfully with Parcel 2. Next up I need to figure out why Parcel 2 is no longer minifying the TypeScript benchmark output...

   Edit 2: I discovered that a production build with engines: node needs optimize: true in package.json or the result is not minified. Plus I need the same minifier config file as before or the minifier preserves all comments in the source code. So this benchmark now works with Parcel 2.

[evanw]

Here are the updated benchmark results:

JavaScript benchmark

Old:

Bundler	Time	Relative slowdown	Absolute speed	Output size
esbuild	0.37s	1x	1479.6 kloc/s	5.80mb
parcel 2	30.50s	80x	17.9 kloc/s	5.87mb
rollup 3 + terser	32.07s	84x	17.1 kloc/s	5.81mb
webpack 5	39.70s	104x	13.8 kloc/s	5.84mb

New:

Bundler	Time	Relative slowdown	Absolute speed	Output size
esbuild	0.39ѕ	1x	1403.7 kloc/ѕ	5.80mb
parcel 2	14.91ѕ	38x	36.7 kloc/ѕ	5.78mb
rollup 4 + terser	34.10ѕ	87x	16.1 kloc/ѕ	5.82mb
webpack 5	41.21ѕ	106x	13.3 kloc/ѕ	5.84mb

So Parcel 2 is roughly twice as fast than before on this benchmark.

TypeScript benchmark

Old:

Bundler	Time	Relative slowdown	Absolute speed	Output size
esbuild	0.10s	1x	1318.4 kloc/s	0.97mb
parcel 2	8.18s	82x	16.1 kloc/s	0.97mb
webpack 5	15.96s	160x	8.3 kloc/s	1.27mb

New:

Bundler	Time	Relative slowdown	Absolute speed	Output size
esbuild	0.10ѕ	1x	1318.4 kloc/ѕ	0.97mb
parcel 2	6.91ѕ	69x	19.1 kloc/ѕ	0.96mb
webpack 5	16.69ѕ	167x	7.9 kloc/ѕ	1.27mb

So Parcel 2 is roughly 20% faster than before on this benchmark.