websocket close/async_close doc wrong

[vinniefalco]

The javadocs are outdated, they state that the close frame is only written. The current implementation also reads and waits to receive a close frame, then performs a teardown.

[jpramosi]

I think the documentation for async_close should also state:

-when it makes sense to call this function respectively when it does has an effect -where it is allowed to use it -are there any caveats (for e.g. check if stream is open or an error_code check is required from read/write handler)

Currently it looks like the examples for session / server related code doesn't make use of async_close. After i have tried to implement a close function for server sessions, i came up with this but not bullet-proof solution which doesn't fire an assert:

void on_async_read()
{
    // ...

    if (m_read_async_close)
    {
        // Maybe it needs additional checks
        // with a wrapper function
        m_stream.async_close(
            websocket::close_code::normal,
            boost::beast::bind_front_handler(&session::on_async_close,
                                             this->shared_from_this()));
        return;
    }

    //...

    // Next read
    m_stream.async_read(
        m_read_buffer,
        boost::beast::bind_front_handler(&session::on_async_read,
                                            this->shared_from_this()));
}

void on_async_close(boost::beast::error_code ec)
{
}

void handle_close()
{
    m_read_async_close = true;
    m_stream.next_layer().cancel();
    // respectively 
    // m_stream.next_layer().next_layer().cancel();
}

void close()
{
    boost::asio::post(m_stream.get_executor(),
                        boost::beast::bind_front_handler(
                            &session::handle_close,
                            this->shared_from_this()));
}

Using async_close just with post and possibly other handlers (on_async_write) triggers this assert in my stress tests: https://github.com/boostorg/beast/blob/79e6c61db5d8d9ba12ad9d2371d34e4bc5a4b9ff/include/boost/beast/websocket/detail/impl_base.hpp#L328

BOOST_BEAST_VERSION 277 a.k.a Boost 1.72.0
/usr/local/include/boost/beast/websocket/detail/impl_base.hpp:328: 
std::size_t boost::beast::websocket::detail::impl_base<true>::read_size_hint_pmd(
std::size_t, bool, std::uint64_t, const detail::frame_header &) const: 
Assertion `rd_remain > 0' failed.

However this happens only for plain websocket sessions (dunno if this is relevant).

A few references: https://github.com/boostorg/beast/blob/1e5b4f531d620879e152979351df920ce339d018/example/websocket/client/async/websocket_client_async.cpp#L171 https://github.com/boostorg/beast/blob/ff2c0ca91e445812d49de68beeed7942fc8d10e2/test/beast/websocket/test.hpp#L138

[madmongo1]

Hi @reapler ,

Would you be in a position to share a complete test program that demonstrates the assert?

If your program has only one of each async operation outstanding at one time we are surprised that this happens.

I would like to investigate ASAP.

[vinniefalco]

That's asking a lot, but I guess it can't hurt to ask :)

[madmongo1]

@reapler Actually you may have uncovered a bug. I am investigating.

[jpramosi]

@madmongo1 I will try to boil this down to an example if i can. At the first sight it doesn't seem to be an issue with concurrent async operations. async_close can only be called once in write/read/timer handler and further async operations concerning read, write, ping or other are stopped with a variable in its handlers. I will add later more information to aid here. Thank you

[jpramosi]

I have found the problem at my side.. moved a part of the code which returns if close was initiated to a separate function, the check was ignored and caused additional async_read after async_close. Now async_close does work also on other handlers without this cancel trickery. I guess this is what the triggered assert actually tried to tell me :) I'm sorry that i have wasted your time. In the future i need to dig deeper in my code, dang

[madmongo1]

@reapler fantastic news. You didn't waste my time at all. It gave me a reason to familiarise myself with the websocket code. Thanks for letting us know. If you need any more help don't hesitate. We're on slack in the #beast channel. https://cppalliance.org/slack/

[vinniefalco]

I'm sorry that i have wasted your time.

Actually, this helped us identify some small improvements, so it wasn't a waste of time.