At some point, the matching call to reporter.uninit got removed from this test.
The result is that the reporter still exists and is still collecting errors
during the rest of the tasks in the file. In most tasks, this isn't an issue
since we use reporter.handleMessage to test message handling at a per-reporter
level.
But the telemetry measures are shared between multiple reporters, thus they are
susceptible to interference from other running reporter instances.
The error that is being logged when this test fails is from the test add-on
created in testAddonIDMangle. My best guess is that the error logged by the
add-on is being processed during an idle moment in another task, since we
schedule processing to be delayed until the browser is idle. It seems this
scheduling is pretty consistent on certain Linux platforms.
Differential Revision: https://phabricator.services.mozilla.com/D1836
Telemetry data suggests between 13%-40% of errors being collected are from
builds older than a week. Since Nightly updates twice a day, errors from builds
that old don't reflect the current state of Nightly, so we can ignore them and
save some bandwidth.
A build is considered "recent" if the date encoded at the start of the
appBuildId (YYYYMMDD) is within 7 days of the current date. Since this is mostly
for preventing high load on the collection service, the check does not handle
problems with the local clock being inaccurate in order to simplify the
implementation.
MozReview-Commit-ID: BbCO4kaBprL
testFetchArguments and the add-on ID mangling tests already cover error
collection end-to-end, so no new tests are needed to cover testing the idle
callback.
MozReview-Commit-ID: DJrqT5jCq44
This also removes any redundant Ci.nsISupports elements in the interface
lists.
This was done using the following script:
acecb401b7/processors/chromeutils-generateQI.jsm
MozReview-Commit-ID: AIx10P8GpZY
Tests are also added here for the legacy `getTopWindow` method to guard against
basic regressions.
We now start tracking browser windows right after the DOMContentLoaded event, which
is earlier than before. We now also assume that any newly tracked window has the
focus initially, which is closer to the nsIWindowMediator semantics.
MozReview-Commit-ID: 6QYJqA1tMPC
Also adds resource://devtools to the whitelist of reported paths for the
scalar.
Differential Revision: https://phabricator.services.mozilla.com/D831
MozReview-Commit-ID: BiAyoTQsWxx
The transforms for turning an nsIScriptError into a payload that Sentry
understands were getting a bit complex for a single function, so they're
refactored into a list of transform functions that are applied in sequence to
produce the payload. This will make it easier to manage adding new transforms to
the list.
Refactoring this revaled a problem with the test code: it assumed that listeners
for console messages were notified in order of registration (since it used a
temporary listener to determine when the rest of the listeners had been notified
of a message). Changing the async evaluation of the code broke the tests, so
they had to be refactored as well.
Without a way to know when all console listeners have been notified, we can't
assert that a message will not be received by BrowserErrorReporter. We do two
things to get around this:
- Where possible, call `observe` directly on the reporter instance.
- Add constructor params for registering and unregistering listeners so we can
test that logic without relying on messages being received or not.
MozReview-Commit-ID: EEH6IROOuHD
When compartment-per-addon is disabled, browser mochitests will no longer
automatically run in an implicit Sandbox scope, which means that things like
Cu.importGlobalProperties will stop working.
MozReview-Commit-ID: AWloQ7gasEf
