The completion tests are fragile and too sensitive to task model
changes. I have a change that inserts a new analysis result between
RESOLVED_UNIT3 and RESOLVED_UNIT4, and by doing this the next task to
perform resolves types in imported libraries. But in CC test we
continue performing analysis tasks and incidentally resolve something
what the 'null' checks don't expect.
So, I think we need to stop doing these checks.
If needed, we can check something directly by asking the state of some
analysis results from AnalysisContext.
R=jwren@google.com, danrubel@google.com
BUG=
Review URL: https://codereview.chromium.org/1932003002 .
Konstantin Shcheglov
470c5de12e