GitLab

The connection request system keeps around the LevelDBDatabase object
when a delete happens to allow subsequent open on that same database.
To do this, it resets the IndexedDBMetadata back to a default state.

During a recent refactor, this state reset wasn't done properly, and we
don't have any tests to catch that error. This change fixes that
incorrect state reset.

A test isn't included here because:
1. It passes the third party test provided to us,
2. The current unittesting framework isn't great and needs to be
   replaced (don't want to create negative work)
3. It is pretty obvious this fixes the problem, and there are no other
   'incorrectly' or 'not' re-initialized state here,
4. Current layout tests should have caught this, and it is unclear why.
   Bug to create this test: https://crbug.com/1018444
5. IndexedDB work is currently on the backburner while Storage Service
   is pursued, and won't be re-prioritized until next year-ish. I don't
   want to spend time on this now when it's needed for Storage Service
   work.

If more P0 bugs crop up in IndexedDB over the next week, then we may
have to re-think this policy and divert engineering effort towards
making IndexedDB more stable. This is present in m78, which was just
promoted to stable. If there are any more issues, we expect them to
show up in the next week or so.

TESTED=Manually tested using example test case from field.
R=cmp@chromium.org

Bug: 1018406
Change-Id: I36e42777e07ef91bf8cf11b9a633a9171ca3755e
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1880320
Commit-Queue: Daniel Murphy <dmurph@chromium.org>
Reviewed-by: Chase Phillips <cmp@chromium.org>
Cr-Commit-Position: refs/heads/master@{#709702}