[cfe-dev] TSan reports a race on vtable access (Clang/7)

[Kostya Serebryany via cfe-dev]

I think this case is described in
https://github.com/google/sanitizers/wiki/ThreadSanitizerPopularDataRaces#data-race-on-vptr

On Thu, Mar 14, 2019 at 9:23 AM Oleg Smolsky <oleg at cohesity.com> wrote:

> I've hit a funky TSan warning: "data race on vptr (ctor/dtor vs virtual
> call) in StartImpl()". I fail to see a race as the object is constructed
> in the main thread and then a worker thread is started:
>
> struct Client : public ClientInterface {
>    Client() {
>      auto op = new Op{this};
>      thread_ = std::thread([op] { op->StartImpl(); });
>    }
>    ...
> };
>
> void Op::StartImpl() {
>    std::cout << "This is a race: " << owner_->VirtualFunction() << "\n";
> }
>
> struct TestClient : public Client {};
>
> int main(int argc, char **argv) {
>    TestClient().Wait();
>    return 0;
> }
>
> The full report with a minimal/complete repro is here:
> https://github.com/google/sanitizers/issues/1063
>
> What am I missing? Does TSan miss the fact that all data stores happen
> before the worker thread is started? Or is this about the vtable pointer
> which is adjusted during the object's construction in the main thread?
>
> Thanks!
> Oleg.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20190314/9eaec66d/attachment.html>