[Lldb-commits] [lldb] [lldb] Fix scripted frame providers deadlock caused by command & event-handler threads (PR #208242)

[Med Ismail Bennani via lldb-commits]

medismailben wrote:

> The change itself looks good, but the out-of-line fix belongs in its own PR. The comments in this PR also need to be cleaned up to be (1) less verbose and (2) explain the generic principle rather than one particular application.
> 
> Finally, please ensure that TSan agrees that this doesn't introduce any races. While I think the general logic here is sound, I want to make sure there were no latent races that happened to be synchronized by the API lock that now become real. While TSan does have some scenarios where it reports false positives (i.e. non-instrumented code, atomic fences, inline assembly) there's nothing in this PR it shouldn't understand.

@JDevlieghere I ran the test suite under TSAN and this doesn't introduce newer failures.

https://github.com/llvm/llvm-project/pull/208242