Strip PropTypes from production builds

[ziir]

Though we use Flow almost everywhere, there are still quite a lot of PropTypes left in our production bundles.

In a typical production bundle, .propTypes search gives ~150 occurences.

We'd gain precious bytes by using a plugin such as babel-plugin-transform-react-remove-prop-types in production builds while we progressively get rid of propTypes alltogether.

[pascalduez]

This could be easily done in our Apps webpack config. Just add that plugin for the production build.
But I'm afraid the size gain might be negligible.
Especially if we envision that they should progressively be removed anyway. (Flow, new context API).

For libs it's not really worst it.
node_modules are excluded from the babel-loader (for good reasons I guess).
So if we add it here, there won't be proptypes for dev, which is the main goal of proptypes.
As for libs shipping a dev and prod versions that feels like a lot of complexity ahead for again small gains.

To me the most significant size gain we can have are:

• React 16
• Webpack 4
• Babel 7 (smaller produced code)
• ESM everywhere
• Maybe Rollup (at least on large libs like ui) (already tested here)
• https://github.com/GoogleChromeLabs/webpack-libs-optimizations
• ...

[ziir]

Yeah, I forgot about our libs...

But, about:

node_modules are excluded from the babel-loader (for good reasons I guess).

If we move forward with #5

Then we will apply babel-loader to all our internal libraries in our bundles compilation.
So applying this plugin only in bundle compilation will result in propTypes being stripped:

• from the app code
• from our libraries code

So we can use this plugin in our webpack config today, and it'll be ~really worth it the day we tackle #5

Thanks for sharing your insights, we definitely to get on these

[pascalduez]

Then we will apply babel-loader to all our internal libraries in our bundles compilation.

I'm concerned about build time, especially in dev, but everything's open.

Food for thoughts:
facebook/create-react-app#1125
gatsbyjs/gatsby#2114

[ziir]

I think there's no reason for us to compile the legacy bundle in development, so build time increase should not be an issue.

The main thing I'm gathering from these discussions is that applying babel-loader to the entire node_modules might not be safe, which I would agree.
We will only need to apply it to @gandi scoped packages, over which we have control.

[ziir]

Closing this as no longer relevant.

We are applying babel-plugin-transform-react-remove-prop-types on a limited set of node_modules packages, during the production client-side bundle compilation, with little gains.