Re: [Boost-users] Using Coroutines in the Visitor Pattern (use-case: BGL event visitor)

Subject: Re: [Boost-users] Using Coroutines in the Visitor Pattern (use-case: BGL event visitor)
From: alex (alexhighviz_at_[hidden])
Date: 2016-01-19 06:02:16

>-----Original Message-----
>> It was somehow disappointing that I couldn't simply do:
>>
>>     while (dijkstra_object() )
>>    { }
>
>You could instead, as in Daniel's example program a couple posts back,
>use a range 'for' over the dijkstra_object.

Yes, that would do perfectly.

>
>> I can see in the documentation that the constructor enters the coroutine-
>function, but it is not clear to me why. Would it not have been neater if
this was
>avoided?
>
>With a pull_type coroutine, the expectation is that every time it
>suspends (until it exits), it has produced a value for its consumer.
>If the coroutine constructor didn't enter the coroutine-function, the
>first invocation would have to be a special case.

To me it seems that the first invocation may either exit or produce a value,
just as each subsequent invocation. I don't see the why the first invocation
is special.

I was also somehow expecting that the construction of the coroutines would
be separate from there invocation, because I would assume that makes it
easier to manage their appropriate execution order.

> Even if that were
>desirable, how could you distinguish the case in which the coroutine
>produces zero values?

That case would just never enter the while loop. 

>That said, if you want finer-grained control over exactly when the
>coroutine is entered, you might consider using execution_context
>instead [0].
>

No, I don't want that. As a potential user I just thought it would be
helpful to report where I found the interface (slightly) counterintuitive.