I should say why I favor the event loop-level inject for container, but not
for the connection engine scenarios.  In the former, we are providing the
event loop (in the form of a convenience class, Container).  In the latter,
we are requiring the user to bring their own event loop or loops.  As a
consequence, they are responsible for implementing any serialization
guarantees.

On Thu, Mar 24, 2016 at 7:16 AM, Justin Ross <[email protected]> wrote:

> Okay, then I think we're seeing the same picture more or less.  The
> difference is that I think the container inject is important.  The
> container API requires that the user give up their thread to run().  Having
> some way to thereafter control that thread is natural to want.
>
> Agreeing I think with what you have said, I don't see a need for a
> container.inject-like facility for the multithreaded (connection-engine
> based) scenarios.
>
> On Thu, Mar 24, 2016 at 7:07 AM, Andrew Stitcher <[email protected]>
> wrote:
>
> > On Thu, 2016-03-24 at 06:59 -0700, Justin Ross wrote:
> > > ...
> > > I think I don't understand what you're saying.  I don't need locking
> > > to modify objects (queues for instance) in the container
> > > serialization context.  That's the point: code I inject there is only
> > > ever executed in sequence by the container event loop.
> >
> > But that's only true in a single threaded environment, and in that case
> > it's also true you don't really need to inject either, because there
> > can be no actual concurrent execution going on.
> >
> > So what you say is true for container, but not for any more complex set
> > up. I thought we were trying to put together a more unified proposal
> > which would be usable in those too.
> >
> > BTW: An import point I failed to make clear before:
> >
> > Nothing in my proposal stops you from *also* having an inject that is
> > scoped to a container (or in fact any other object if you can figure
> > out its serialisation requirements) - I just don't think "container"
> > will be all that useful.
> >
> > Andrew
> >
> >
>

Reply via email to