Re: asio and custom socket implementation

From: Mathias Gaunard (mathias.gaunard_at_[hidden])
Date: 2020-04-03 18:04:39

On Fri, 3 Apr 2020 at 07:14, Gavin Lambert via Boost-users
<boost-users_at_[hidden]> wrote:
>
> On 2/04/2020 11:57, Richard Hodges wrote:
> > On Thu, 2 Apr 2020 at 00:48, Gavin Lambert wrote:
> >     But custom epoll/select/poll is an entirely different kettle of fish.
> >     If your custom library must use those (and cannot provide a standard
> >     platform file descriptor that works with the standard platform
> >     epoll/select/poll), then you're largely out of luck.  For that, you
> >     would have to write a custom reactor instead; and as far as I am aware
> >     without deep modifications to Asio there is no provision for doing so
> >     (and it's a massive amount of work that is hard to get correct,
> >     although you can use the existing implementations as a guide).
> >
> > I respectfully disagree. You'd have to write a new executor and
> > associated context. It would look very similar to the existing one.
>
> (Granted I was mostly thinking of pre-executor Asio when writing the
> above, but...)
>
> Having recently had an adventure myself in implementing a custom
> executor, I stand by my statement.
>
> There are many required operations it does not seem possible to
> implement without either depending on or reimplementing 'detail' code
> from Asio, and there is very little guidance on doing so.  Unless you
> know of some example...?

Most people who design systems with low-latency networking end up
re-implementing their own executor framework.
I suppose it could be useful if it could be done entirely with
asio-compliant concepts.