Closed shayanhabibi closed 2 years ago
See my comment here to show the tsan/asan analysis pre & post this PR
We need tests that actually do work and TRY to run into data races and undefined behaviour.
The deallocations and what not are more reasonable and are unlikely to actually cause undefined behaviour in our program despite thread sanitizers whinging about them.
Will have to fix that but I'm really struggling to think of how I can develop a stringent test for memory coherance that doesn't rely on thread sanitizers since they will raise problems either way.
In regards to #9
Unfortunately tsan/asan/valgrind will continue to not agree with nim in general since deallocshared on thread destructions will consistently triggered global memory writes