Sometimes the tests are run under high load, e.g. immediately
after boot while all apps that listen for the boot completed
broadcast run their listeners. Under these circumstances,
40ms of leeway is tight ; on recent Android devices HZ is
commonly 100 too, meaning a general time unit of 60 will give
a little bit of air.
The drawback in this is that the test is less stringent on
TrackRecord, which as a multi-thread primitive should have
really tight tests to make sure that a non-threading primitive
(e.g. a polling implementation) doesn't pass. This patch makes
the testing on that front a little bit less reliable.
Note that this patch won't actually fix all flakes in that it
isn't really possible to guarantee the device will have any
CPU time to run the test at all anyway, but this should make
most failures go away.
Also, the argument for the time unit is never used, and can
be removed ; at least, the possibility of the timeArg() func
not using the same value as the sleep directive is scary, so
that possibility should be removed. If the time unit argument
is ever needed it can be added back, but considering the arg
has been available for years and never used, it's unlikely.
Test: NetworkStaticLibTests
Fixes: 254188131
Change-Id: Ia7a4fe775942484dcd7dc223c7710e82a8329beb
347dd8ded4