Improve the run time of base_unittests by about 20s by reducing timeouts in

Improve the run time of base_unittests by about 20s by reducing timeouts in
the following tests:

ObserverListThreadSafeTest.CrossThreadObserver
ObserverListThreadSafeTest.CrossThreadNotifications
ThreadCollisionTest.MTSynchedScopedBookCriticalSectionTest
ThreadCollisionTest.MTSynchedScopedRecursiveBookCriticalSectionTest

TEST=Covered by base_unittests.
BUG=none

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=31112

[Paweł Hajdan Jr., 2009-11-05 08:45:52]

This is a part of GTTF, based on the slowness report at
http://build.chromium.org/buildbot/slowness-report/

I think that now we have tools like ThreadSanitizer etc. which allow us to
reduce the timeouts in these tests that are there to try triggering a race
condition. Timur, could you comment on that (whether the tools can replace a
longer loops and timeouts here)?

[Nicolas Sylvain, 2009-11-05 15:15:44]

the change LGTM, make sure to watch for new flakiness

Nicolas

[Paweł Hajdan Jr., 2009-11-05 16:03:04]

On 2009/11/05 15:15:44, Nicolas Sylvain wrote:
> the change LGTM, make sure to watch for new flakiness

Just to clarify: this change cannot introduce new flakiness (new test failures
without reason). It may give more false negatives though, but it should be safe
as we use better tools to detect threading issues (tsan etc).

[Mike Belshe, 2009-11-05 16:30:03]

lgtm

[Timur Iskhodzhanov, 2009-11-05 18:14:18]

LGTM

[Timur Iskhodzhanov, 2009-11-05 18:17:29]

Regarding ThreadSanitizer,
Sleeps don't affect its race detection algorithm most of the time.
When sleeps do affect the algorithm, having shorter sleeps is even better for
the accuracy.

On 2009/11/05 18:14:18, Timur Iskhodzhanov wrote:
> LGTM