Re: [boost] Asynchronous library now in Boost Library Incubator

Subject: Re: [boost] Asynchronous library now in Boost Library Incubator
From: Christophe Henry (christophe.j.henry_at_[hidden])
Date: 2016-12-01 16:20:06

On 11/30/2016 10:27 PM, Thomas Heller wrote:
> On Mittwoch, 30. November 2016 21:47:34 CET Christophe Henry wrote:
>>> After having a deeper dive into the documentation and examples, one
>>> pattern I see popping up quite
>>> frequently is this:
>>> future<R> fut = my_promise.get_future();
>>> post_callback(some_work, [this](expected<R> r) {
>>> this->my_promise.set_value(r.get()); });
>>> return fut;
>>
>> This is simply to make examples easier to run (it finishes the example). It
>> demonstrates that the only one which blocks in an application built on
>> Asynchronous is main().
>> This might be more confusing than useful and I probably should get rid of
>> this.
>
> Understood. It's always a difficult task to present clear and good examples and
> best practices.
>
> Out of curiosity (I can't recall reading that in the documentation, correct me
> if I am wrong), how is a real and complex application supposed to signal
> shutdown/completion?
> The answer to this question is somehow important. You make it very clear, in
> the documentation and here in this thread, that having a chain of future.then
> (which conceptually really are just callbacks that are called whenever the
> future became ready), is not the way to go because eventually you need to
> block the thread of execution. At least as your very last call main, for
> example.
> Joining the threads of your active threadpools also blocks the thread of
> execution. So where is the difference? Which technique did I miss?

I know two ways:
- adding a SIGKILL handler (posix). In that case, the main simply has to 
destroy the proxys it keeps alive. This will cause the matching servants 
to be destroyed, which will trigger the proxies held by these servants 
to be destroyed, etc. When none of them holds any threadpool alive, they 
will be destroyed too. In any case, all outstanding tasks will be 
executed so that no memory or resource leak occurs due to unfinished tasks.

- as in the examples. The Servants held by main return a future, main 
waits for it. When the servants want to die (for example due to a button 
in a UI), they set the corresponding promise. Then happens same as above.

The difference is important for sizeable applications. In these 
applications we have not one chain of tasks but thousands. Possibly 
blocking on them is complicated to handle.

Using asynchronous, the only one blocking is main. This is quite a 
difference. Of course, if your whole application is one big functional 
programming model, it makes no difference. I however doubt most of us 
want this.

As I wrote here 
(https://htmlpreview.github.io/?https://github.com/henry-ch/asynchronous/blob/master/libs/asynchronous/doc/asynchronous.html#d0e6437) 
I disagree on the idea that a chain of futures is the same as callbacks.

Cheers,
Christophe