windows - Spurious timing test failures on windows.

[garyfurnish]

Related to Trac 9856 in test_9856

The fact that Thread doesn't return timeout up to 10% of the time on Windows is wontfix.

Can we at least change the threshold for the test on windows so that we don't get spurious failures?

This was tested in 3222938.

[viboes]

Could you tell me on which platform and compiler you got the issue? are those run in http://www.boost.org/development/tests/develop/developer/thread.html?

[garyfurnish]

This is on Windows 10 insiders with Visual Studio 17.4 Preview. I think it was also failing on 17.3 on windows 10, but I can't guarantee that. I could rerun with regular 17.3. This is a 4/8 sandy bridge machine. (More detailed specs in other bugs). This persisted seemingly independent of system load.

[viboes]

Could you tell me which % threshold is good for Windows?

I don't know, maybe it isn't worth running this test.

[garyfurnish]

If I run it alone the highest I've gotten was 13.8% and 15% might be a reasonable target.
If I use b2 -j8 (num hyperthreads) to run the tests then it can go at least 38.2%, maybe higher. I suspect this test will always fail if run in parallel with other tests. Run alone its probably reasonable.

[viboes]

I guess that the testers are using -jx

I will set the threshold to 40% :(

[viboes]

3ee73c8

[garyfurnish]

Tests clean under load, closing as fixed.