[libc++abi] Fix backtrace_test.pass.cpp failure seemingly caused by inlining differences.

[EricWF]

Summary

The modified assertion fails when the test is compiled at various optimization levels.
The value of nothrow_traced and throw_traced at various optimization levels are:

• -O0: nothrow_traced = 7, throw_traced = 7
• -O1: nothrow_traced = 4, throw_traced = 5
• -O2: nothrow_traced = 4, throw_traced = 4
• -O3: nothrow_traced = 4, throw_traced = 4

I'm not sure exactly what this test is doing, so I would like somebody else to sign off on this change.

[jroelofs]

I'm not sure that's guaranteed behavior either. That being said, I don't see a more robust way to write this test.

[danalbert]

Maybe __attribute__((noinline)) every function instead?

[EricWF]

Another fix was committed.

test/backtrace_test.pass.cpp

try {
     call1(&throw_ntraced, true);
   } catch (...) {
     assert(false);
   }
 
   // Different platforms (and different runtimes) will unwind a different number
   // of times, so we can't make any better assumptions than this.
   assert(nothrow_ntraced > 1);
-  assert(throw_ntraced == nothrow_ntraced); // Make sure we unwind through catch
+  assert(throw_ntraced >= nothrow_ntraced); // Make sure we unwind through catch
   return 0;
 }