Proposed resolution:

This wording is relative to N4917.

1. Modify [thread.jthread.cons] as indicated:

   jthread& operator=(jthread&& x) noexcept;

   -12- Effects: If &x == this is true, there are no effects. Otherwise, if joinable() is true, calls request_stop() and then join(). Assigns, then assigns the state of x to *this and sets x to a default constructed state.

   -13- Postconditions: x.get_id() == id() and get_id() returns the value of x.get_id() prior to the assignment. ssource has the value of x.ssource prior to the assignment and x.ssource.stop_possible() is false.

   -14- Returns: *this.

[ 2022-11-12 Approved at November 2022 meeting in Kona. Status changed: Voting → WP. ]

[ 2022-09-29; Reflector poll ]

Set status to Tentatively Ready after five votes in favour during reflector poll.

[ 2022-09-28; Jonathan provides wording ]

[ 2022-09-28; Reflector poll ]

Set priority to 3 after reflector poll.

In the Postconditions element of jthread& jthread::operator=(jthread&&) ([thread.jthread.cons] p13), we have the following:

Postconditions: x.get_id() == id(), and get_id() returns the value of x.get_id() prior to the assignment. ssource has the value of x.ssource prior to the assignment and x.ssource.stop_possible() is false.

Assume j is a joinable jthread. Then, j = std::move(j); results in the following post-conditions:

• Let old_id = j.get_id() (and since j is joinable, old_id != id())

• Since x.get_id() == id(), j.get_id() == id()

• Since get_id() == old_id, j.get_id() == old_id

• Thus, id() == j.get_id() == old_id, and old_id != id(), which is a contradiction.

One can see that these postconditions are therefore unimplementable.

Two standard libraries – the MSVC STL and libstdc++ – currently implement jthread.

The MSVC STL chooses to follow the letter of the Effects element, which results in unfortunate behavior. It first request_stop()s, then join()s; then, it assigns the values over. This results in j.get_id() == id() – this means that std::swap(j, j) stops and joins j.

libstdc++ chooses instead to implement this move assignment operator via the move-swap idiom. This results in j.get_id() == old_id, and std::swap(j, j) is a no-op.

It is the opinion of the issue writer that libstdc++'s choice is the correct one, and should be taken into the standard.