MOZ_ASSERT is only defined in debug builds. All the automatic tests run in debug builds, but there are obviously corner cases which are not covered by our tests. So the only chance how to better identify these corner cases is to have assertions in optimized Nightly builds which are used daily by many users.
It adds some overhead, but I think we don't have any other options.

This is primarly intended to identify remaining deadlocks and shutdown hangs.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=73ac0759c5bf43e95dd25fe38f0ed441fd202e5f

Basically, all MOZ_ASSERT should be converted to MOZ_DIAGNOSTIC_ASSERT, except the ones for thread checks.
Also, all #ifdef DEBUG should be converted to #ifdef MOZ_DIAGNOSTIC_ASSERT_ENABLED, except the ones for NS_WARNING or LS_WARNING.

https://hg.mozilla.org/mozilla-central/rev/97d88752079f

Ok this was expected to happen, sorry for the trouble.
We obtained really good data and now we can fix remaining bugs.

I plan to land only one part/patch a day.

https://hg.mozilla.org/mozilla-central/rev/d64e4d9400ec

So, the incorrect usage is still stored in some usage files. We need to change the "cookie" of the usage file so all usage files get recalculated value.

There are 13 crashes with signature 'mozilla::dom::(anonymous namespace)::QuotaClient::InitOrigin' and 4 with 'mozilla::dom::`anonymous namespace'::QuotaClient::InitOrigin'.
For all these crashes the moz_crash_reason is: MOZ_RELEASE_ASSERT(usage >= 0).

Yeah, the patch has been backed out.

Jan, next time can you please ask sheriffs to back out the crashing patch from central directly? That way we avoid an extra unnecessary crashy build. Thanks :)

There were no issues on my local machine and all the tests passed on try. I needed to verify with bigger audience.
Sorry about that.

Backout of part 1: https://hg.mozilla.org/mozilla-central/rev/3dcc6eec59f4

https://hg.mozilla.org/mozilla-central/rev/d134fc3b9e29

https://hg.mozilla.org/mozilla-central/rev/729f59649b3f

There are 8 crashes with signatures 'mozilla::dom::(anonymous namespace)::Snapshot::RecvCheckpoint' and 'mozilla::dom::`anonymous namespace'::Snapshot::RecvCheckpoint'.
They all hit the assertion: MOZ_RELEASE_ASSERT(mPeakUsage >= mUsage).

https://hg.mozilla.org/mozilla-central/rev/b02147ae4804

https://hg.mozilla.org/mozilla-central/rev/dad8b92996d5

https://hg.mozilla.org/mozilla-central/rev/29bf13f9b35a

Setting checkin-needed-beta to get the latest backout (from comment 32) grafted to 68 to stop crashing DevEdition builds.

https://hg.mozilla.org/releases/mozilla-beta/rev/298f7d8bb0c0

https://hg.mozilla.org/mozilla-central/rev/288c3317b2fc

https://hg.mozilla.org/mozilla-central/rev/1d394d2c86db

https://hg.mozilla.org/mozilla-central/rev/585544d6e2cd

Julien, do we also want to uplift patches from comment 36, comment 38 and comment 40 ?
All those crashes will automatically stop once early beta ends, not sure if it will happen in beta 9.

https://hg.mozilla.org/releases/mozilla-beta/rev/8a44e86052bb5439b6f71ee0c616ba2f6251a655
https://hg.mozilla.org/releases/mozilla-beta/rev/8814bb1352784e68fc0944c17104d308166a93b7
https://hg.mozilla.org/releases/mozilla-beta/rev/927b23f07645dd3f1ff4b63d8cf477bec77b9e22

Hi :janv, is the leave-open still needed?

Yes, the ultimate goal is to have most of the assertions defined as diagnostic asserts, but it's not a high priority right now.

The leave-open keyword is there and there is no activity for 6 months.
:jstutte, maybe it's time to close this bug?

See comment 44.

Closing because no crashes reported for 12 weeks.