[lmi] Sporadic error in 'timer_test'

[Greg Chicares]

A few weeks ago I saw this failure:

**** test failed:   '0.975153' < '0.032'
[file /lmi/src/lmi/timer_test.cpp, line 126]

and something similar occurred just the other day:

**** test failed:   '0.974785' < '0.032'
[file /lmi/src/lmi/timer_test.cpp, line 126]

so I'm wondering what could be wrong with TimerTest::TestResolution().

I made this local change, to get more information...

Index: timer_test.cpp
===================================================================
--- timer_test.cpp      (revision 6402)
+++ timer_test.cpp      (working copy)
@@ -124,6 +124,15 @@
     double relative_error = std::fabs(observed - interval) / interval;

     BOOST_TEST_RELATION(relative_error,<,2.0*clock_resolution);
+
+    std::cout
+        << CLOCKS_PER_SEC << " CLOCKS_PER_SEC\n"
+        << clock_resolution << " clock_resolution\n"
+        << observed << " observed\n"
+        << interval << " interval\n"
+        << relative_error << " relative_error\n"
+        << std::endl
+        ;
 }

 void TimerTest::TestExceptions()

...and changing the code made the problem disappear. Now I see, e.g.:

1000 CLOCKS_PER_SEC
0.015 clock_resolution
1.00005 observed
1 interval
4.94609e-05 relative_error

AIUI, the msw-xp clock ticks 64 times per second, so 'clock_resolution'
looks correct, and so do the failing tests cited above--they report
       BOOST_TEST_RELATION(relative_error,<,2.0*clock_resolution);
                                            ^^^^^^^^^^^^^^^^^^^^
two times 'clock_resolution', which should be between .030 and .032 .

I run it a hundred times:
  for a in {1..100}; do ./timer_test --accept ; done 2>&1 |less
and the test never fails.

As a last resort, I looked at the code, but it uses the C RTL, which
I don't remember well. Now I'm staring at this and wondering how it
can work:

    std::clock_t first = std::clock();
    std::clock_t last;
    double clock_resolution;
    for(;;)
        {
        last = std::clock();
        clock_resolution = double(last - first) / CLOCKS_PER_SEC;
        if(0.0 != clock_resolution)
            {
            break;
            }
        }

The way I read that, it starts at a random moment between clock ticks,
and measures the time until the next tick. But if that's correct,
then 'clock_resolution' should vary randomly between 0 and .015625,
whereas it always seems to return .015 or .016 . Vadim, help?

Stepping back, I question whether this test is even useful in any way.
I think TimerTest::TestResolution() should just be eliminated.