fix: windows download incorrectly selects 32bit on 64bit machine

[styfle]

I just installed Windows 11 64 bit and opened Edge to install Node.js and it incorrectly installed the 32 bit version.

User Agent:

Mozilla/5.0 (Windows NT 10.0) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/109.0.0.0 Safari/537.36 Edg/109.0.1518.55

Also navigator.cpuClass is undefined so we need a new way to detect the the bitness.

• Fixes #5028

[ovflowd]

I appreciate the PR, but we're on a code-freeze right now, not to mention all the current "javascript/jquery" code is going to be trashed, as we transition to the codebase from nodejs.dev.

Would you mind opening an issue instead of a PR? So we can work on that when we start migrating the layouts?

Thank you!

[styfle]

I created issue https://github.com/nodejs/nodejs.dev/issues/3162

I also created PR https://github.com/nodejs/nodejs.dev/pull/3163

[nschonni]

Closing this, as the error is on the .dev site, and not the .org which this PR is against. .org doesn't have this problem. Thanks for creating the issue for the other site!

[styfle]

@nschonni The .org site does have this problem too

[styfle]

I created https://github.com/nodejs/nodejs.org/issues/5028 so you don't lose track of the .org bug either.

However, I suggest you reopen my PR here since it is still relevant.

[nschonni]

No, the download URL comes from .org for both, but only https://nodejs.dev/en/ has this problem, not https://nodejs.org/en/

[styfle]

@nschonni Both have the same issue. Did you try it?

Here's a screenshot with proof:

[nschonni]

It's possible something in the Insider build is broken, I can reproduce your issue on .dev with both Firefox and Edge, but not on .org.

PS: The solution isn't to wipe out the architecture detection and always serve x64

[styfle]

It's possible something in the Insider build is broken

Could be. Or it could point to changes coming in the future where the user agent no longer includes that information.

The solution isn't to wipe out the architecture detection and always serve x64

How about flipping the if statement so that it checks for 32bit first and then falls back to 64bit if the string didn't match? I think its safe to say most Windows installations are 64bit, right?

[nschonni]

Could be. Or it could point to changes coming in the future where the user agent no longer includes that information.

Definitely possible, but could also just a be a regression in a particular build. Here is the Edge user agent for Win 11 Edge Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/109.0.0.0 Safari/537.36 Edg/109.0.1518.55 The spec says it should be there https://html.spec.whatwg.org/multipage/system-state.html#dom-navigator, but I'm wondering if there is some privacy masking be tried out in the Insider build

How about flipping the if statement so that it checks for 32bit first and then falls back to 64bit if the string didn't match? I think its safe to say most Windows installations are 64bit, right?

If you can find a way to do that, but I think only 64 had indications in the user agent, so that's why it is done this originally.

I think most of the actual x86 windows versions are approaching EOL, so there may be value in visiting the approach as part of the redesign

[styfle]

I investigated some more and found out Windows 11 is only 64 bit, there is no 32 bit version.

Furthermore, there is no way to detect Windows 11 based on the user agent string.

Are you looking for Windows 11 user agents? Surprisingly, they don't exist and never will. Recently, User Agent strings have started to become less important for browser detection online and some browser developers - notably the Chrome and Firefox development teams - have "frozen" the Windows version number in their user agent strings at "Windows 10". So even as new versions of Windows are released, the user agents for those browsers won't reflect this and will continue to report Windows 10.

https://www.whatismybrowser.com/guides/the-latest-user-agent/windows

So then I found the official microsoft docs which say to use client hints or this JS script:

navigator.userAgentData.getHighEntropyValues(["platformVersion"])
 .then(ua => {
   if (navigator.userAgentData.platform === "Windows") {
     const majorPlatformVersion = parseInt(ua.platformVersion.split('.')[0]);
     if (majorPlatformVersion >= 13) {
       console.log("Windows 11 or later");
      }
   }
 });

https://learn.microsoft.com/en-us/microsoft-edge/web-platform/how-to-detect-win11

I confirmed that this works on the insiders windows 11 build because platformVersion is 16.0.0 which is above the 13.

But then I looked at MDN and saw this can be simplified even further to check for bitness.

navigator.userAgentData.getHighEntropyValues(["bitness"])
 .then(ua => {
   console.log(ua.bitness) // "64"
 });

https://developer.mozilla.org/en-US/docs/Web/API/NavigatorUAData/getHighEntropyValues

I think adding that script should solve the problem since it works on my build of windows 11 😄

Thoughts?

[nschonni]

That might make sense, but I'd suggest keeping the existing checks, and adding as the other ||. Note that the site is currently mostly frozen while they try and merge the layout from the .dev site, but if you fix it there it might come down to this repo eventually

[ovflowd]

I wanted to mention that UA's are being sunset (frozen) (through phasing). It will be tough to predict future OS's User-Agents in the future, as browsers will stop updating them. This is already reflected in Windows 11 by giving less information.

• https://groups.google.com/a/chromium.org/g/blink-dev/c/-2JIRNMWJ7s/m/yHe4tQNLCgAJ?pli=1
• https://developer.chrome.com/en/blog/user-agent-reduction-oct-2022-updates/

[styfle]

@nschonni @ovflowd I pushed some changes to this branch to use the new API. Can we reopen this PR?

(function () {
  'use strict';
  if (navigator.userAgentData && navigator.userAgentData.getHighEntropyValues) {
    navigator.userAgentData
      .getHighEntropyValues(['bitness'])
      .then(function (ua) {
        initDownloadHyperlink(ua.bitness);
      })
      .catch(function () {
        // Ignore errors since not every browser supports this API
      });
  } else {
    initDownloadHyperlink('unknown');
  }
})();

[ovflowd]

I'm not sure how adopted this API is. What is the caniuse.com of this API?

Anyways, I would highly recommend not making PRs to nodejs.org, as this will all go to waste. This main.js is going to be discarded very soon, so I recommend a PR against nodejs.dev to improve its detection mechanism there :)

[styfle]

I did create one here https://github.com/nodejs/nodejs.dev/pull/3163

[ovflowd]

Does that PR contain similar changes? (Leveraging the new API?)

[styfle]

@ovflowd Yep, both PRs are very similar using the same navigator.userAgentData.getHighEntropyValues API.

[styfle]

@nschonni @ovflowd @shanpriyan Anything else we need to land this?

[ovflowd]

@styfle we're on a code freeze for #4991

[ovflowd]

@styfle could you rebase your PR so we can get this merged?

[SEWeiTung]

@styfle : The file is totally changed with the new tech for Nodejs.org. So yours is useless and I've closed your change and re-open a new PR for you.

Any suggestions welcomed!