Add example app with React Intl

[ericf]

Fixes #1022

[timneutkens]

cc @arunoda @nkzawa could you do a final review?

[ericf]

Let me know if there's anything you guys want changed. It would be great to get this merged.

[sedubois]

Where is window.ReactIntlLocaleData created?

[ericf]

From the locale data modules; e.g. https://unpkg.com/[email protected]/locale-data/en.js

The server selects one React Intl's locale data scripts and inlines it on the page. Since the locale data is packaged as a UMD module, in the browser it creates the window.ReactIntlLocaleData global.

More details here: https://github.com/yahoo/react-intl/wiki#locale-data-in-browsers

[sedubois]

Passing locale data around like this seems contrived 🤔 As I commented in formatjs/formatjs#731 (comment)

Is there no way for getInitialProps to return the localeData and simply pass that to addLocaleData?

[ericf]

There are several issues with Next at play here:

1. page/_document.js is bundled with Webpack, and errors our if you try to use fs or any other built-in Node modules even though it only runs on the client.

2. The __NEXT_DATA__ is serialized using JSON.stringify(). This means that getInitialProps() must return an object that's JSON-serializable, and React Intl's locale data isn't JSON-serializable since it contains functions for pluralization.

3. Calling addLocaleData() on every page transition wouldn't be ideal. It's best to only call this function one time.

It's possible to address 1 and 2 of these issues in Next, and I'd be happy to rework example once they are changed, but as it stands now there isn't a good way around it.

For 2, serialize-javascript can be used which does support serializing functions. Changing Next to use this instead of raw JSON.stringify() would remove the need for this example to use a custom pages/_document.js.

[sedubois]

Thanks for the details 🙂

1. IMHO worth to create a bug on Next.js
2. the localeDataScript itself (loaded by require.resolve + readFileSync) is just a string, which is serializable? Then the string could be evaluated where needed in the HOC (e.g with eval)? Right now the intl code is spread all around and that's hardly maintainable.
3. should work by returning a serverRendered: !process.browser in your getInitialProps and wrap the addLocaleData in if (props.serverRendered)?: https://github.com/relatenow/relate/blob/ee51f1dd6990177ff0924b1c0d78fddfd23aa564/hocs/withIntl.js#L43

[ericf]

how is using eval worse than dangerouslySetInnerHtml? Both sound ominous to me anyway. (but I'm no expert on this)

In this case dangerouslySetInnerHTML is being used only on the server and it's just creating an inline <script> in the HMTL response. Your idea of using eval() would be running on the client would conflict with CSP; discussion here: #256

I hope not to spread i18n logic in my custom server etc just to handle this.

How are you doing HTTP content negotiation for choosing the locale on both the server and client?

I encourage people to who have apps that render on both the server and client to do the content negotiation and choose the correct locale server-side, and have the client use the locale the server decided on.

The main issue I ran into when trying to do this type of content negotiation and locale data loading in the HOC is that Webpack would bundle all locale data, all translated messages, and npm packages that were only meant to run on the client. I think the only way to keep everything in the HOC would involve i18n related stuff in a custom webpack config file — and I don't know what that would look like, but maybe you or others do?

[sedubois]

Thanks, I'm not familiar with CSP, will try to read more on this.

For detecting the browser locale I used express-request-language, then store it in the session, and allow the user to customise it (server-side).

For co-locating the code a bit more, I just managed to load the message files from within the HOC (it does involve a slight webpack tweak): https://github.com/relatenow/relate/blob/master/hocs/withIntl.js#L25

[sedubois]

I just updated to use the accept-language module to better co-locate the code: sedubois/relate@5c9d95a

This means that I don't need a custom server at all to use react-intl at this point.

[ericf]

@sedubois do you want to send a PR to my branch with your suggested changes? Then I can merge your commits into this branch/PR.

[sedubois]

@ericf ok I can try to send something, maybe tomorrow (not sure if have time).

[ericf]

Are you guys interested in merging this?

[timneutkens]

@ericf Sure are, Will have a review tonight/ tomorrow. Thanks for this PR, it's Great 🙌🏻🙌🏻

[timneutkens]

All examples have

• name the same as the directory in this case with-react-intl
• No description
• Empty author ("author": "")
• an ISC license https://github.com/zeit/next.js/blob/master/examples/basic-css/package.json#L15

[timneutkens]

Looks good, only added a comment for the package.json to keep it consistent. 👍

[impronunciable]

Looks good to me, LGTM after this package.json change @ericf

[ericf]

@timneutkens @impronunciable okay I updated the package.json to make it consistent.

[timneutkens]

Awesome ❤️

[steida]

Was serialize-javascript issue fixed? Thank you.

[sedubois]

@steida I believe not?, but there is a pull request ongoing to remove eval in favor or JSONP or something (#1611). Can same principle be used here?