[PATCH] D51170: [libc++] Remove race condition in std::async

[Louis Dionne via Phabricator via cfe-commits]

ldionne added inline comments.


================
Comment at: libcxx/include/future:556
     bool __has_value() const
         {return (__state_ & __constructed) || (__exception_ != nullptr);}
 
----------------
jfb wrote:
> I'm not auditing everything, but it seems like code above can still access __state_ without holding __mut_? Like in the dtor.
> 
> Generally this patch lgtm because it's a step forward, but maybe we should separately refactor the code to make it so that accesses to __state_ require passing in a reference to lock_guard to show we actually hold __mut_. It would ignore that reference, but that's a way to enforce, in the type system, that __state_ is only touched when the lock is held.
> 
> WDYT?
I think you're right, and I filed this bug to keep track of the issue: https://bugs.llvm.org/show_bug.cgi?id=38688

Not all of them need a lock (some are in the constructor where only one thread has a reference to the data, for example), but most of them probably do.


Repository:
  rCXX libc++

https://reviews.llvm.org/D51170