Not sure if anyone has had a chance to look at this yet, but it is somewhat mission-critical that this work.

I have a client that needs to be able to detect a dozen or so switches being triggered, and we have a board deployed that uses the gamepad API on a chromebox in kiosk mode to accomplish this. It has been working perfectly up to this point, but broke at the aforementioned base position. I have redeployed the app as a web app as a temporary patch for the issue, but I need to be able to guarantee function when the internet is out and a chrome app seems to be the best way to provide that guarantee.

Any help with this issue would be greatly appreciated.

Best guess:

https://chromium.googlesource.com/chromium/src/+/9f6cd0eb83b8d3166071dd5d5780b688034e0a61

It is the only commit in that range that touches gamepad related code. I looked at the diffs, but I don't understand the function call changes well enough to submit a patch myself.

Thanks for narrowing down the regression range. I'll try to repro.

Just to confirm (since it wasn't clear to me), it works fine as a plain page (i.e. using the repro via `python -m SimpleHTTPServer`, but fails when loaded as an extension).

I can repro per #c0 in 59.0.3046.0 (Official Build) canary (64-bit). 

yukishiino@, would you be able to investigate based on https://codereview.chromium.org/2341723002?

#5 - That is correct.

I'm wondering if that bisect is a bit misleading now. It changed behaviour so that the unconnected gamepads are "null" rather than "undefined", but in the extension case it just doesn't show up.

I'll try to bisect with a modified repro that checks for any truthy Gamepad object showing up instead.

(I don't recall if the spec specifies if we must have undefined or null there, but that's sort of separate from this issue.)

I got this bisect

You are probably looking for a change made after 404931 (known good), but no later than 404965 (first known bad).
CHANGELOG URL:
  https://chromium.googlesource.com/chromium/src/+log/8373bd3fbd078372e714edd73ab95a23b1d1d603..254dc66bc04a9eeef14440dfc7fe79b8cc9ec2a5


using

setInterval(() => {
  var gamepads = navigator.getGamepads();
  console.log(gamepads)
  if(gamepads[0] ||
     gamepads[1] ||
     gamepads[2] ||
     gamepads[3]){
    document.getElementsByTagName("body")[0].style.backgroundColor = "green";
  }
}, 30);

instead.

But I don't see a CL that looks promising there. Hmm.

I did a commit level git bisect and landed on

[f18ffeb...]c:\src\cr2\src>git bisect bad
f18ffebd279ad21196d988f763313e4a460db316 is the first bad commit
commit f18ffebd279ad21196d988f763313e4a460db316
Author: haraken <haraken@chromium.org>
Date:   Tue Jul 12 20:28:34 2016 -0700

    Remove DOMWindowProperty::willDetachGlobalObjectFromFrame

    DOMWindowProperty::willDetachGlobalObjectFromFrame is dispatched
    at the same timing as ContextLifecycleObserver::contextDestroyed
    (via Document::detach). Hence this CL replaces DOMWindowProperty::willDetachGlobalObjectFromFrame
    with ContextLifecycleObserver::contextDestroyed.

    BUG=610176

    Review-Url: https://codereview.chromium.org/2138003002
    Cr-Commit-Position: refs/heads/master@{#404961}

:040000 040000 c85dca6812f180beded7719df5aa79bdc49ce7a6 f48e13db6a89a80c9ff354245ec1d3099d84680f M      third_party

[f18ffeb...]c:\src\cr2\src>git bisect log
git bisect start
# good: [8373bd3fbd078372e714edd73ab95a23b1d1d603] Removed histogram merge code that is no longer actively used.
git bisect good 8373bd3fbd078372e714edd73ab95a23b1d1d603
# bad: [254dc66bc04a9eeef14440dfc7fe79b8cc9ec2a5] Auto-rebaseline for r404891
git bisect bad 254dc66bc04a9eeef14440dfc7fe79b8cc9ec2a5
# good: [bdbc56ba8839e6bb32a3677bf199234641f36806] [Remoting Android] Fix GetWeakPtr() calls on multiple threads
git bisect good bdbc56ba8839e6bb32a3677bf199234641f36806
# good: [59d3718ac1328d46fac147720f078c53f63070cf] Implement use_cfi_cast to optionally enable cast checks.
git bisect good 59d3718ac1328d46fac147720f078c53f63070cf
# good: [b7f49840eb4c80f98c30e599fb39ae1a41a6eefc] Show empty bell icon in notification button initially
git bisect good b7f49840eb4c80f98c30e599fb39ae1a41a6eefc
# bad: [6aa36b6c741ab188448a6673c393af135ee6572e] Implement SignedCertificateTimestamp::isolatedCopy
git bisect bad 6aa36b6c741ab188448a6673c393af135ee6572e
# bad: [f18ffebd279ad21196d988f763313e4a460db316] Remove DOMWindowProperty::willDetachGlobalObjectFromFrame
git bisect bad f18ffebd279ad21196d988f763313e4a460db316
# first bad commit: [f18ffebd279ad21196d988f763313e4a460db316] Remove DOMWindowProperty::willDetachGlobalObjectFromFrame


I guess this indicates some change in lifetime of the gamepad object that somehow only affects extensions/apps.

haraken, do you understand what's happening here?

re #8: Just FYI (not related to this issue itself),

https://w3c.github.io/gamepad/#navigator-interface-extension
sequence<Gamepad?> getGamepads();

where "Gamepad?" means an Object of interface Gamepad or null.  "?" stands for "nullable" (https://heycam.github.io/webidl/#idl-nullable-type).

Just filed  bug 711252  which may be related: it's not to do with Chrome apps, it's that gamepad input doesn't work at the top level in popup windows. In other words gamepad input only works when the full tabbed browser UI is showing, or if it's in an iframe.