When I load the page, the search_data.json file is automatically downloaded. Could we make it so that it is only loaded when the user clicks the Search button?
That way, we avoid downloading that giant file for users on mobile that don't want to use Search at all (would often happen if they click on a link to a release news or something).

UI-wise, the interface is confusing. When I type in something, I have to switch to the mouse to go and click on a link. I would like to use up/down arrows to select things + Enter to go. Enter could default to the first item.

Also, the search "button" at the right of the input box appears as a link but does nothing. It should be removed if it doesn't do anything.

That's all, I think.

The search data download is deferred after the page is shown, so it doesn't hinder usage. It's also cached and gzipped, so the penalty is quite low. If it were downloaded only when the search button is clicked, it would be too late from UX point of view.

After gzip it's <100kB in size.

If it were downloaded only when the search button is clicked, it would be too late from UX point of view.

There's still the time to download it while the user types in something.

After gzip it's <100kB in size.

I guess that's acceptable, but it still feels like a big hit for all the mobile users who won't ever use the search box, if they hit it on Data usage.

Maybe I'm too paranoid, here.

The search button is indeed superfluous, but you can navigate with Tab+Enter. I'll make an update for this. Arrow key navigation would be nice, but probably requires a bit more work, so can do that as a separate update later.

On mobile the issue is latency, not bandwidth, so downloading search data JIT would not work that well.

Besides, comparing to the old scala-js.org site, I think the overall download size and request count is still way less :)

OK

That's all.

LGTM