[clang] [compiler-rt] [llvm] [tsan] Add simulation to TSAN (PR #183200)
Chris Cotter via cfe-commits
cfe-commits at lists.llvm.org
Sun Nov 8 21:59:43 PST 2026
================
----------------
ccotter wrote:
Coming back to this, I don't think the scheduler can be cleanly separated from the TSAN runtime. For example, implementing the simulation for something like `pthread_mutex_lock` in simulation mode must take a different code path to use `pthread_mutex_trylock` to avoid deadlock. Although I do see the parallels to this PR being a fuzzer, at the moment there is no feedback sent back to the scheduler. I don't think the choice of `__tsan_simulate` (the lone public API) locks us in too much if we find a way to separate the scheduler out from the runtime in the future. The bigger API lock-in risk is the new TSAN_OPTIONS, though a theoretic evolution into a distinct fuzz/schedule engine could read from those options in a transition state.
@vitalybuka - thoughts on merging this as-is given the above? I don't see a path forward for separating the scheduler given how intertwined the logic is.
https://github.com/llvm/llvm-project/pull/183200
More information about the cfe-commits
mailing list