ConnectionPool.lockConnection() hangs under certain conditions.

[japplegame]

When using ConnectionPool.lockConnection() in parallel in multiple tasks (single thread), under certain conditions it hangs.

On my CentOS 7 x86_64 + vibe.d 0.7.29 this code

/+ dub.json:
    {
        "name": "test",
        "dependencies": {
            "vibe-d:core": "0.7.29"
        }
    }
+/

import std.stdio;
import std.socket;
import std.datetime;
import std.functional;
import vibe.core.core;
import vibe.core.log;
import vibe.core.concurrency;
import vibe.core.connectionpool;

class Conn {}

void main() {
    runTask({
        // create pool with 2 max connections
        auto pool = new ConnectionPool!Conn({ return new Conn; }, 2);
        auto task = Task.getThis(); // main task
        void worker(int id) {
            {
                auto conn = pool.lockConnection(); // <-- worker(4) hangs here
                sleep(1.msecs); // <-- important, without sleep everything works fine
            }
            task.send(id); // send signal to the main task
        }
        // run 4 tasks (2 * pool max connections)
        runTask(&worker, 1);
        runTask(&worker, 2);
        runTask(&worker, 3);
        runTask(&worker, 4);

        // wait for first signal and run one more task
        writefln("recv  %s", receiveOnly!int);
        runTask(&worker, 5);

        // wait for other signals
        writefln("recv  %s", receiveOnly!int);
        writefln("recv  %s", receiveOnly!int);
        writefln("recv  %s", receiveOnly!int);
        writefln("recv  %s", receiveOnly!int);

        exitEventLoop();
    });
    runEventLoop();
}

outputs

recv  1
recv  2
recv  3
recv  5

and never exits the even loop.
Where is recv 4?

[s-ludwig]

Appears to be already fixed by #1563. I'll add this as a regression test.