Re: [newbie] Question on embedding/extending with boost::python

"Dirk Gerrits" <dirk@xxxxxxxxxxxxxxxxxx> wrote:
> Brett Calcott wrote:
> >
> > I had previously embedded python in my c++ app so that I could
> > script my agents, but the result was very slow as this effectively
> > did this (vastly simplified)
> >
> > while (engine_running)
> > {
> >     for (agents_container:iterator a = agents.begin(); a !=
> >         agents.end(); ++a)
> >     {
> >         // here we call out to python for each agent
> >         (*a).update_yourself_via_python()
> >         update_visual_stuff()
> >     }
> > }
> >
> > The calls to python are in the inner loop and this is slow.
>
> Well I suppose you could expose a visual_stuff_updating_iterator to
> Python, but extending is of course a much easier and cleaner.

Not sure I follow what you mean here..


>
> > What I am doing now is different - I am constructng an engine in
> > C++, making all the agents and world parameters accessible to
> > python, BUT the update main loop remains entirely in C++:
> >
> > import engine # boost python module
> >
> > while engine_running:
> >     engine.cycle()
>
> I'm not sure I follow. Is engine.cycle() still doing a
> a->update_yourself_via_python() for every agent? Then where's the
> speedup compared to the embedded approach?
>
>

No, it's not. The engine.cycle() function never calls back to python.

Let me try and clarify.

The big difference in what I am doing now is how I am using python -- it
is serving a completely different role in my software development.

Before, I wrote everything in C++ - and python was playing the role of
allowing me to script the actions of my agents. I was using python
because it allowed me to change the behaviour of my agents without
recompiling. The degree to which I benefitted from python was the ease
with which I could quickly change the behaviour of my agents -- the
problem was that it slowed the simulation down.

Now, I am using python to construct the entire framework, and I am
writing the agents, including their "update" function, in C++. The
advantage I get from this is the speed of development of all the support
framework, including inspection/pickling/scripted setups, and the power
and ease of the entire python library. However, the core engine that has
to be quick will be in C++. Now, I don't get the advantage of being able
to script my agents behaviour, that is still in C++, but I get a
headstart on everything else.

Of course, it is not as bad as that. I can still parameterise the agents
behaviour and set those parameters interactively in python and, as
before, I can still choose to use an agent that is extensible in python
and really slow, but just for the purpose of testing -- then I can push
the code back to C++. This is particularly easy with the boost_python
stuff.

To summarise -- that big change is in how I look at python. Before, I
saw C++ as a serious development framework, and python as a scripting
language for doing the little bits of my app that I wanted to change
quickly. Now, I see python as an incredibly fast development framework,
and C++ as something to make the computationally intensive bits of my
app go faster.

So I things have sort of been stood on their head.  I still love C++,
but I can make stuff happen so much more quickly in python.

Cheers,
Brett