1401744 - Opt-out Enrollment Study breaks about:addons after failsafe date

Status: RESOLVED WONTFIX

(Shield :: Shield Study, defect)

Description

[Osmose [:osmose, :mkelly]]

The add-on used in bug 1394962 contains a failsafe that automatically uninstalls the study on startup if the local time is past Tuesday, September 12, 2017 12:00:00 AM. However, when that failsafe is executed, the following error is logged to the browser console:

1505942118428	Toolkit.Telemetry	ERROR	TelemetryEnvironment::_checkForChanges: Error collecting addons: TypeError: promiseCallProvider is not a function (resource://gre/modules/AddonManager.jsm:2513:32) JS Stack trace: [e-mail]:2513:32 < async*[e-mail]:3609:7 < [e-mail]:655:27 < async*[e-mail]:627:27 < async*[e-mail]:581:25 < [e-mail]:566:5 < [e-mail]:561:5 < [e-mail]:1667:11 < [e-mail]:3162:5 < [e-mail]:4738:7 < [e-mail]:5699:5 < removeAddon/</<@resource://gre/modules/addons/XPIProvider.jsm -> jar:file:///Users/mkelly/Library/Application%20Support/Firefox/Profiles/yy0fzwly.default-1477538586801/extensions/[e-mail]:125:62 < [e-mail]:187:5 < makeSafe/<@AddonManager.jsm:202:25 < promise callback*[e-mail]:228:3 < [e-mail]:3578:12 < removeAddon/<@resource://gre/modules/addons/XPIProvider.jsm -> jar:file:///Users/mkelly/Library/Application%20Support/Firefox/Profiles/yy0fzwly.default-1477538586801/extensions/[e-mail]:125:5 < removeAddon@resource://gre/modules/addons/XPIProvider.jsm -> jar:file:///Users/mkelly/Library/Application%20Support/Firefox/Profiles/yy0fzwly.default-1477538586801/extensions/[e-mail]:124:9 < async*startup@resource://gre/modules/addons/XPIProvider.jsm -> jar:file:///Users/mkelly/Library/Application%20Support/Firefox/Profiles/yy0fzwly.default-1477538586801/extensions/[e-mail]:24:11 < async*[e-mail]:4494:20 < startInstall/<@XPIInstall.jsm:1878:13 < async*[e-mail]:1778:6 < [e-mail]:1450:7 < [e-mail]:2121:5 < [e-mail]:1736:5 < [e-mail]:1447:7 < [e-mail]:2121:5 < checkPrompt/<@XPIInstall.jsm:1711:7 < async*[e-mail]:1691:6 < [e-mail]:1444:7 < [e-mail]:2121:5 < [e-mail]:2644:5 < [e-mail]:2062:5 < [e-mail]:2132:9 < [e-mail]:2139:9 < [e-mail]:3644:5 < handleUriInChrome/<@browser.js:1032:11 < [e-mail]:187:5 < makeSafe/<@AddonManager.jsm:202:25 < promise callback*[e-mail]:228:3 < [e-mail]:3552:12 < [e-mail]:1031:9 < [e-mail]:1075:9 < [e-mail]:8505:13 < [e-mail]:443:5 < [e-mail]:207:3 < [e-mail]:2412:9  Log.jsm:752
TypeError: logger is undefined[Learn More]  AddonManager.jsm:229:5
TypeError: AddonManagerInternal is undefined[Learn More]

After this, about:addons will not load until the browser is restarted. The study add-on is still successfully uninstalled; after a browser restart, Firefox will behave normally.

It's unclear yet if this is only a problem with the enrollment study or with other Shield studies that are using the study utils.

Comment 1

[Osmose [:osmose, :mkelly]]

The cause of the bug is these lines in the shutdown handler (they're only run when the add-on is being disabled or uninstalled):

Cu.unload('resource://gre/modules/Console.jsm');
Cu.unload('resource://gre/modules/AddonManager.jsm');
Cu.unload('resource://gre/modules/TelemetryEnvironment.jsm');

Unloading these modules while they're in use by other parts of the browser causes lots of weird issues, such as about:addons not being able to load properly. We shouldn't be unloading them from an add-on.

I cannot find any other studies that have this issue. The example study in the study utils library also does not have these lines: https://github.com/mozilla/shield-studies-addon-utils/blob/master/examples/stage-2-shield-study/addon/bootstrap.js

We don't intend to run this study again, and I don't think there is a repo for the enrollment study anyway, so I'm going to mark this as WONTFIX. I've filed https://github.com/mozilla/normandy/issues/1069 for adding the ability to disable extensions uploaded to Normandy, so that we can mark this extension as unusable.

I'm also going to send a an email to dev-shield to follow up on the requirements/standard test plans we should have in place for opt-out studies, as a way to address why we missed this bug in the first place.