Make Android app start faster by disabling JS bundle compression

[mrousavy]

Disables compression for the .bundle resource files in the .apk.

This makes the .apk a bit bigger, but improves startup time on Android - we measured a 400ms startup improvement (12%) in one of our apps.

Based on facebook/react-native#49449

[mrousavy]

Draft because I couldn't yet figure out how to make this work with Expo Config Plugins.

fixed

[mrousavy]

Okay now it works, I fixed it!

@gaearon do you guys have any TTI measurements in BlueSky? If yes, would be great if you could run those in release builds to see how much faster it got with my PR.

I can't seem to be able to build the app in release.

[surfdude29]

FYI @gaearon isn't working at Bluesky anymore:

https://bsky.app/profile/danabra.mov/post/3lhcrrc76s22i

[gaearon]

There's this:

social-app/src/Navigation.tsx

Line 860 in 40d05ac

console.log(`Time to first paint: ${initMs} ms`)

You can get it to show up (and see it in logcat) by commenting out the "remove console" plugin in the prod Babel config.

I can't seem to be able to build the app in release.

Not sure if anything is broken, last time I tried yarn android --variant release was working fine.

[mozzius]

Ok results are in:

Current (mean of 10 runs): 805ms

With change (mean of 10 runs): 773ms

Improvement: 4% faster!

That's not nothing, and this was on a Pixel 7A. I'd imagine that lower-powered phones might see more benefit.

[mozzius]

🔥

[mrousavy]

I haven't dug too deeply, but it looks like initMs is calculated from __BUNDLE_START_TIME__:

social-app/src/Navigation.tsx

Lines 856 to 859 in 40d05ac

	const initMs = Math.round(
	// @ts-ignore Emitted by Metro in the bundle prelude
	performance.now() - global.__BUNDLE_START_TIME__,
	)

This sounds wrong to measure this specific improvement, since the performance gains already start a bit earlier than that; when Android starts to load the actual .bundle from resources (see _CompressedAsset::getBuffer).

So if you measured 4% with this change, you only measured the mmap difference. The real improvements might be much bigger because we didn't measure the uncompression part of this whole thing. Anyways; it's faster now 😄

[gaearon]

Seems like you’d want to measure the time from the first native app log to when this console log comes in.

[mrousavy]

yep