identity_current_thread is bad "default" scheduler for operators

[victimsnino]

Hi everyone and hi @kirkshoop!

Most of the operators in rxcpp which requires schedulers has such an fallback scheduler:

If scheduler is omitted, identity_current_thread is used.

For example, merge operator also has it as default. BUT it doesn't provide ANY synchronization/serialization in case of multithreaded application.

Example:

    rxcpp::observable<>::just(1, rxcpp::observe_on_new_thread())
        .repeat()
        .merge(rxcpp::observable<>::just(2, rxcpp::observe_on_new_thread())
                    .repeat())
    .take(10).as_blocking().subscribe([](int v){
        std::cout << "==================\n" << std::this_thread::get_id() << " START " << std::endl;
        std::this_thread::sleep_for(std::chrono::seconds{1});
        std::cout << v << std::endl;
        std::cout << std::this_thread::get_id() << " END " << std::endl << "==================\n\n";

    });

Possible output:

==================
0x70000ee59000 START 
==================
0x70000eedc000 START 
1
0x70000ee59000 END 
==================

2
0x70000eedc000 END 
==================

==================
0x70000ef5f000 START 
==================
0x70000efe2000 START 
1
0x70000efe2000 END 
==================

2
0x70000ef5f000 END 
==================

==================
0x70000ee59000 START 
==================
0x70000eedc000 START 
1
0x70000ee59000 END 
==================

2
0x70000eedc000 END 
==================

==================
0x70000ef5f000 START 
==================
0x70000efe2000 START 
1
0x70000ef5f000 END 
==================

2
0x70000efe2000 END 
==================

==================
0x70000ee59000 START 
==================
0x70000eedc000 START 
1
0x70000ee59000 END 
==================

2
0x70000eedc000 END 

As you can see, it is mixed, but ReactiveX requires that any observable should be serialized.

In case of using any valid scheduler in merge like observe_on_new_thread output is valid:

==================
0x700001ce1000 START 
1
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
2
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
1
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
2
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
1
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
2
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
1
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
2
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
1
0x700001ce1000 END 
==================

==================
0x700001ce1000 START 
2
0x700001ce1000 END 
==================

Expected: at least default behavior of any operator should be thread-safe...

In my understanding, best default scheduler for such an "multhithreaded" opertators can be "serialize_immediate" (not exist, but actually just emit emissions under mutex to provide exclusive access to subscriber and guarantee that only one observable pushes item at the same time). Tested locally: also provides valid output

BTW: it is how i've implemented merge in ReactivePlusPlus: each callback to subscriber of merge just called under mutex. As a result there is no way to obtain "mixed" log. @kirkshoop, what do you think ?