In headless tests it was possible for TestInstantLoad() to erroneously
fail. There were two scenarios in which LoadingSpinner could be null:
1. If the test runner was quick enough, the assert could end up running
even before Loader.OnEntering() had even had a chance to, meaning
that the spinner was never even actually assigned to or instantiated
at that point in time.
2. Even if Loader.OnEntering() had managed to run, there was also
a possibility that the spinner itself wasn't loaded at the point
of checking the assertion. As the spinner is accessed through
ChildrenOfType(), which only checks InternalChildren and ignores
all currently-loading drawables, it would therefore return null.
As null != 0, both of these cases would actually fail the test (this is
best seen running headless, preferably with a [Repeat] attribute
attached).
To resolve, allow the spinner to be null at the point of asserting and
duplicate the assertion step at the end. This weakens the test, as case
(1) should probably be waited for and case (2) could be solved with
exposition as protected in the base, but when attempting to wait for
the loader itself to be loaded there were also cases where the
appropriate until step would take so much time that the spinner would
actually become visible in line with the delayed display logic, so this
is a best-effort attempt to address both points without radical
changes.
Dan Balasescu ·
Bartłomiej Dach ·
Dean Herbert ·