Android x86 7.0 reftests (crashtests, jsreftests) running tier 3 on mozilla-central are currently permafailing and have been since https://treeherder.mozilla.org/#/jobs?repo=mozilla-central&searchStr=android%2Cx86%2C7.0&revision=423bdf7a802b0d302244492b423609187de39f56 https://treeherder.mozilla.org/#/jobs?repo=mozilla-central&searchStr=android,x86,7.0&tochange=1169e8a4ca2b7f3cbffdaf70f6d18a5142ed32d7&fromchange=b0bf04a1cbbf633497cca6cf047b3573e53d484b

In that push, bug 1451503 stands out as potentially related. Of course this reminds me of bug 1426822.

This is easy to reproduce locally in the Android 7.0 x86 emulator: $ ./mach reftest layout/reftests/reftest-sanity --app=org.mozilla.geckoview.test The same tests run fine without the --app argument (against Firefox for Android).

Consider https://dxr.mozilla.org/mozilla-central/rev/ac9f1219d11bf1a56ec1ace8e3ba9ff113b5cacb/layout/tools/reftest/api.js#70-77 if (Services.appinfo.OS == "Android") { ChromeUtils.import("resource://reftest/reftest.jsm"); if (win) { startAndroid(win); } else { Services.wm.addListener(WindowListener); } return; } In Firefox for Android, startAndroid(win) is called and the tests run fine; in org.mozilla.geckoview.test, addListener() is called, but the listener's onOpenWindow is never called. Andrew, Jim -- Any insight/suggestions?

I meant to also note the win setting just before that: let win = Services.wm.getMostRecentWindow("navigator:browser"); Do we expect that to work in org.mozilla.geckoview.test?

I assume this was working before bug 1451503? The code at that point was: https://searchfox.org/mozilla-central/rev/62bcb25a7c5dda799e294af67a93ddf660c38c0a/layout/tools/reftest/bootstrap.js#41-45 But note that the bootstrap startup() method ran earlier in browser startup than a webextension api onStartup() runs. If onOpenWindow is never called, that suggests the browser window has already been opened when onStartup() is called, but then the window we get here should be non-null: https://searchfox.org/mozilla-central/rev/0b8ed772d24605d7cb44c1af6d59e4ca023bd5f5/layout/tools/reftest/api.js#68 Unless geckoview uses a different windowType? Sorry, I don't really know anything about geckoview...

(In reply to Andrew Swan [:aswan] from comment #5) > I assume this was working before bug 1451503? Yes. > Unless geckoview uses a different windowType? I think that's it maybe -- I see references to "navigator:geckoview", and allowing for that works well locally. Testing more...

This is working fine on try.

Comment on attachment 9010741 [details] [diff] [review] allow for "navigator:geckoview" Review of attachment 9010741 [details] [diff] [review]: ----------------------------------------------------------------- I'm okay with this but it would be nice to either: 1. have a comment explaining what's going on, or 2. Just query once and use something like Services.appinfo.name to decide what type to ask for

Comment on attachment 9010741 [details] [diff] [review] allow for "navigator:geckoview" Review of attachment 9010741 [details] [diff] [review]: ----------------------------------------------------------------- Yeah GV windows use the "navigator:geckoview" window type.

(In reply to Andrew Swan [:aswan] from comment #9) OK - have added a comment.

Pushed by gbrown@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/0ed8c2fb2cd6 Allow reftests to start in geckoview TestRunnerActivity; r=aswan,jchen