Parcel2 v. Webpack - Build times are significantly slower #4566

astegmaier · 2020-05-04T19:15:40Z

🐛 bug report

I've started trying to use Parcel2 in a real-life situation, and I'm noticed that the parcel build times are significantly slower than webpack's, even with a warm cache, as long as there is at least one change since the last build. (It also makes significantly larger bundles - see #4565).

Like all real-life situations, it's complicated, so I tried to create a simplified repro environment in the fluentui-button-babel branch of this repo. It's a simple react app with a single Button component from the @fluentui/react-northstar library.

🎛 Configuration (.babelrc, package.json, cli command)

Webpack (full config here):

Using babel-loader with this config:

presets: [
   "@babel/preset-env",
   "@babel/preset-react",
   "@babel/preset-typescript",
]

production flag set to true, which enables minification through TerserWebpackPlugin and scope-hoisting/tree-shaking through ModuleConcatinationPlugin, among other optimizations.

Parcel configuration:

No .parcelrc file, which means that the default babel transformer is being used.
No custom babel configuration.
Ran parcel build src/index.html

src/index.html:

<!DOCTYPE html>
<html lang="en">
  <head>
    <meta charset="utf-8" />
    <title>Sample App</title>
  </head>
  <body>
    <noscript>You need to enable JavaScript to run this app.</noscript>
    <div id="root"></div>
    <script src="./index.tsx"></script>
  </body>
</html>

src/index.tsx:

import React from "react";
import ReactDOM from "react-dom";
import App from "./components/App";

ReactDOM.render(<App />, document.getElementById("root"));

src/components/App.ts:

import React from "react";
import { Provider, themes, Button } from "@fluentui/react-northstar";

const App = () => (
  <Provider theme={themes.teams}>
    <Button content="Hello from FluentUI" />
  </Provider>
);

export default App;

🤔 Expected Behavior

Parcel should be at least as fast as webpack, hopefully significantly faster if there is a warm cache.

😯 Current Behavior

Here's a comparison of Parcel v. Webpack build times in various scenarios.

Scenario	Trial 1	Trial 2	Trial 3	Average
parcel build - no cache	44.8s	43.7s	46.7s	45.1s
parcel build - with cache, no changes	2.2s	2.2s	2.1s	2.2s
parcel build - with cache, one string change	24.1s	24.3s	23.2s	23.9s
webpack build - after clearing Terser cache	13.7s	13.2s	13.8s	13.6s
webpack build - with Terser cache, no changes	7.4s	7.6s	7.9s	7.6s
webpack build - with Terser cache, one string change	13.3s	13.2s	13.4s	13.3s

All scenarios (both parcel and webpack) are:

doing production builds (i.e. including tree-shaking, scope-hoisting, and minification)
using babel for transpilation (with default config in parcel and babel-loader for webpack).
generating sourcemaps.
not doing any typechecking.
On a MacBook Pro (13-inch, Early 2015) with 2.7 GHz Dual-Core Intel Core i5 and 16 GB RAM.

💁 Possible Solution

There are probably multiple ways to make things faster. One thing I noticed is that a lot of the time each build (even with a warm parcel cache) is spent during the "optimization" phase. It looks like TerserWebpackPlugin might be leveraging some caching that we are not. It might also be related to #4565 - if we could successfully tree-shake the stuff that webpack can, all the later steps would be faster.

🔦 Context

Trying to build a real-world app that uses @fluentui/react-northstar and parcel2.

💻 Code Sample

See the fluentui-button-babel branch of this repo.

🌍 Your Environment

Software	Version(s)
Parcel	2.0.0-nightly.248
Node	12.16.3
Yarn	1.22.4
Operating System	OSX 10.15.4

The text was updated successfully, but these errors were encountered:

mischnic · 2020-05-04T19:18:11Z

One thing I noticed is that a lot of the time each build (even with a warm parcel cache) is spent during the "optimization" phase

If you mean that you see "optimizing foo...." for a while, that could also mean (because optimizing and packaging happens in parallel):
packaging for bundle A started, packaging for bundle B started, optimizing for bundle A started, (now we're actually waiting for the packaging of bundle B).

astegmaier · 2020-05-04T20:28:24Z

If you mean that you see "optimizing foo...." for a while, that could also mean (because optimizing and packaging happens in parallel):
packaging for bundle A started, packaging for bundle B started, optimizing for bundle A started, (now we're actually waiting for the packaging of bundle B).

Thanks for the tip! In case it's helpful, I ran parcel with the --profile flag and uploaded the .trace files to dropbox:

Scenario 1: fresh parcel build with no cache - profile
Scenario 2: build parcel with warm cache and a single string change in the source files - profile.

mischnic · 2020-05-04T21:16:00Z

Does "webpack build - with Terser cache" correspond to "no changes" or "one string change"?

BTW, it doesn't matter what you change (a string or something else), the JS AST handling is by far not intelligent enough to cache anything in that case, the packager has to run again.

Is it possible that the two trace files are swapped? The cold cache one should be larger than the warm cache run.

astegmaier · 2020-05-05T02:53:25Z

Does "webpack build - with Terser cache" correspond to "no changes" or "one string change"?

"no changes"

I clarified the table above and also tested the other scenario - "webpack build - with Terser cahce, one string change", and I got results similar to the "no terser cache" case (~13s on my machine).

Is it possible that the two trace files are swapped? The cold cache one should be larger than the warm cache run.

I'm pretty sure I didn't accidentally swap them, although I agree it is weird that something that should take more work would be smaller. Maybe something went wrong writing the profile to disk? (In an earlier attempt, I ran out of heap memory, and I noticed that it would still write something to disk, but it would be incomplete - maybe that happened, and I didn't notice?). To be extra sure, I ran it again, and here are links to the full set.

#	Scenario	First attempt (also linked above)	Second attempt (new)
1	fresh parcel build with no cache	profile1 (204.4MB)	profile2 (544.2MB)
2	build parcel with warm cache and a change in the source files	profile1 (492.7MB)	profile2 (500.2MB)

Per @wbinnssmith's suggestion, I gziped all four traces for faster downloading. It's only 69MB combined (!).

astegmaier · 2020-05-05T19:34:49Z

Based on what I learned investigating #4565 about how to bypass the tree-shaking issue, I re-ran the perf benchmarks on an app that uses the more-granular import style.

In these tests, I'm importing @fluentui/react-northstar components using the following syntax:

import Box from "@fluentui/react-northstar/dist/es/components/Box/Box";
import Provider from "@fluentui/react-northstar/dist/es/components/Provider/Provider";
import teamsTheme from "@fluentui/react-northstar/dist/es/themes/teams";

Parcel's bundle size is significantly smaller (631KiB) than in the original tests, and more comparable to webpack's (622Kib).

Scenario	Trial 1	Trial 2	Trial 3	Average
parcel build - no cache	35.8s	33.5s	33.4s	34.2s
parcel build - with cache, no changes	2.1s	2.0s	2.1s	2.1s
parcel build - with cache, one string change	18.1s	19.3s	17.8s	18.4s
webpack build - after clearing Terser cache	12.1s	11.2s	11.3s	11.5s
webpack build - with Terser cache	6.2s	6.1s	6.2s	6.2s
webpack build - with Terser cache, one string change	11.5s	11.4s	11.2s	11.4s

So that sped things up significantly, but it's still about 50% slower than webpack in the "one-string change" case and 3x slower in the "no cache" case.

devongovett · 2021-06-23T06:49:21Z

I think this should be mainly solved now after the latest rewrite

astegmaier mentioned this issue May 4, 2020

Parcel2 v. Webpack - parcel bundle size is significantly larger #4565

Closed

mischnic added 🐛 Bug 🚄 Performance ✨ Parcel 2 labels May 4, 2020

devongovett closed this as completed Jun 23, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Parcel2 v. Webpack - Build times are significantly slower #4566

Parcel2 v. Webpack - Build times are significantly slower #4566

astegmaier commented May 4, 2020 •

edited

mischnic commented May 4, 2020

astegmaier commented May 4, 2020 •

edited

mischnic commented May 4, 2020 •

edited

astegmaier commented May 5, 2020 •

edited

astegmaier commented May 5, 2020

devongovett commented Jun 23, 2021

Parcel2 v. Webpack - Build times are significantly slower #4566

Parcel2 v. Webpack - Build times are significantly slower #4566

Comments

astegmaier commented May 4, 2020 • edited

🐛 bug report

🎛 Configuration (.babelrc, package.json, cli command)

🤔 Expected Behavior

😯 Current Behavior

💁 Possible Solution

🔦 Context

💻 Code Sample

🌍 Your Environment

mischnic commented May 4, 2020

astegmaier commented May 4, 2020 • edited

mischnic commented May 4, 2020 • edited

astegmaier commented May 5, 2020 • edited

astegmaier commented May 5, 2020

devongovett commented Jun 23, 2021

astegmaier commented May 4, 2020 •

edited

astegmaier commented May 4, 2020 •

edited

mischnic commented May 4, 2020 •

edited

astegmaier commented May 5, 2020 •

edited