[Bug libstdc++/54352] New: relaxed data race rules for ~condition_variable_any

[Bug libstdc++/54352] New: relaxed data race rules for ~condition_variable_any

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54352

             Bug #: 54352
           Summary: relaxed data race rules for ~condition_variable_any
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Keywords: ABI
          Severity: normal
          Priority: P3
         Component: libstdc++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: redi@gcc.gnu.org


(related to PR 54185, but for condition_variable_any)

As Howard pointed out in c++std-lib-32966 this should work:

#include <list>
#include <mutex>
#include <condition_variable>
#include <functional>
#include <thread>
#include <chrono>
#include <cassert>
#include <algorithm>

template <class T>
class locked_list
{
    std::mutex   mut_;
    std::list<T> list_;
public:
    typedef typename std::list<T>::iterator iterator;
    typedef typename T::key key;

    template <class ...Args>
        void emplace_back(Args&& ...args)
            {list_.emplace_back(std::forward<Args>(args)...);}

    iterator find(const key& k)
    {
        std::unique_lock<std::mutex> lk(mut_);
        while (true)
        {
            iterator ep = std::find(list_.begin(), list_.end(), k);
            if (ep == list_.end())
                return ep;
            if (!ep->busy())
            {
                ep->set_busy();
                return ep;
            }
            ep->wait(lk);
        }
    }

    void erase(iterator i)
    {
        std::lock_guard<std::mutex> _(mut_);
        assert(i->busy());
        i->notify_all();
        list_.erase(i);
    }

    iterator end() {return list_.end();}
};

template <class Key>
class elt
{
    Key key_;
    std::condition_variable_any notbusy_;
    bool busy_;
public:
    typedef Key key;

    explicit elt(const Key& k) : key_(k), busy_(false) {}

    bool busy() const {return busy_;}
    void set_busy() {busy_ = true;}
    void unset_busy() {busy_ = false;}
    template <class Lock>
        void wait(Lock& lk) {notbusy_.wait(lk);}
    void notify_all() {notbusy_.notify_all();}

    bool operator==(const Key& k) const {return key_ == k;}
};

void
f1(locked_list<elt<int>>& list)
{
    auto i = list.find(1);
    assert(i != list.end());
    std::this_thread::sleep_for(std::chrono::milliseconds(500));
    list.erase(i);
}

void
f2(locked_list<elt<int>>& list)
{
    auto i = list.find(1);
    assert(i == list.end());
}

int main()
{
    locked_list<elt<int>> list;
    list.emplace_back(1);
    std::thread t1 = std::thread(f1, std::ref(list));
    std::this_thread::sleep_for(std::chrono::milliseconds(250));
    std::thread t2 = std::thread(f2, std::ref(list));
    t1.join();
    t2.join();
}

This test doesn't actually crash with libstdc++, but valgrind shows it's
faulty.

Fixing this involves replacing std::condition_variable_any::_M_mutex with a 
std::shared_ptr<std::mutex>

[Bug libstdc++/54352] relaxed data race rules for ~condition_variable_any

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54352

--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-08-22 20:43:34 UTC ---
Created attachment 28066
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28066
patch against 4.8 trunk

[Bug libstdc++/54352] relaxed data race rules for ~condition_variable_any

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54352

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> 2012-08-22 20:58:10 UTC ---
(In reply to comment #1)
> Created attachment 28066 [details]
> patch against 4.8 trunk

That patch needs some additional exports:

    _ZNSt16_Sp_counted_baseILN9__gnu_cxx12_Lock_policy*;
    _ZNKSt16_Sp_counted_baseILN9__gnu_cxx12_Lock_policy*;
    _ZT[ISV]St16_Sp_counted_baseILN9__gnu_cxx12_Lock_policyE2EE;

[Bug libstdc++/54352] relaxed data race rules for ~condition_variable_any

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54352

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |FIXED
   Target Milestone|---                         |4.9.0

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Fixed for 4.9