[Lldb-commits] [lldb] [lldb][test] fix sporadic failures with NumberAllocatedModules assert (PR #176884)

[via lldb-commits]

daniilavdeev wrote:

@Michael137, thanks for providing the context. Actually, I am not sure about my solution.

> we shouldn't try tearing down the debugger unless all process have exited.

I was about to write more about my observations. In particular, the lack of synchronization does seem to be the source of the issue. The `tearDown()` function (`lldb/packages/Python/lldbsuite/test/lldbtest.py`) invokes `cleanupSubprocesses()`, `SBDebugger.Destroy()`, and immediately after that calls `GarbageCollectAllocatedModules()`. Thus, I suspect that `SBDebugger.Destroy()` or `cleanupSubprocesses()` may not have released the lock by the time `GarbageCollectAllocatedModules()` is called.

Given that, I tried to make `GarbageCollectAllocatedModules()` wait until the lock becomes available again, rather than giving up after a single attempt. This approach seems to fix the sporadic failures, at least, I haven’t been able to reproduce any of them. However, I’m not sure whether it is appropriate, from the python api perspective, to change the mandatory flag in such a straightforward way.

https://github.com/llvm/llvm-project/pull/176884