On 8/31/06, Andrew Borley <[EMAIL PROTECTED]> wrote:

Hi all,

I'm inches away from getting an extension working that allows Python
scripts
to be used as components in SCA C++. I should be putting that up some time
tomorrow - I'll start a thread about a few minor issues with the extension
framework that I've come across, but it hasn't been too complex to get
this
going.

At the moment the extension provides an implementation.python element to
the
.composite of the form:

<implementation.python module="PythonModuleName"
path="relative/path/to/module" class="PythonClassName"/>

This allows other C++ components or C++ clients to call the service
implemented by code inside the Python module. The service interface is
currently described by a C++ header file, and uses the
interface.cppinterface binding. Currently, references from the Python
components are not
enabled (see below).

There are two further areas that are needed for full integration of Python
into SCA (or SCA into Python):

1) A programming model for SCA in Python to allow Python components or
Python clients to call SCA services. This will enable components with
Python
implementations to have references.
We could either create a Python extension that will emulate the service
lookup that we do in C++, e.g. a Python equivalent of the following C++
snippet:

    // First, get the current ComponentContext
    osoa::sca::ComponentContext myContext =
osoa::sca::ComponentContext::getCurrent();

    // Find the required service, as referenced in
CalculatorImpl.componentType
    DivideService* divideService =
(DivideService*)myContext.getService("divideService");

    // Finally, invoke the service
    result = divideService->divide(arg1, arg2);

Alternatively, we could look at doing some kind of annotations within the
Python component module - I'm not sure whether Python has annotation
support
(I don't believe so), or whether there is a way to 'fake' it to make
comments/documentation act as annotations.

2) An interface.python extension so that python components do not have to
have their interfaces converted to a C++ or WSDL description. Due to the
loose typing that Python allows this isn't the easiest thing to do - most
functions have signatures like: def myMethod(arg1, arg2) - and (as far as
I
can currently tell) there is no way to determine the types of the
arguments
or return values aside from at invocation time.
At the moment the SCAGEN tool generates the C++ component proxy and
wrapper
classes based on the C++ header specified in the interface.cpp element in
the .componentType file. If the .componentType file has an
interface.pythonelement instead, and specifies the Python module,
generating the correct C++
code neccessary to call the python service is hard. Alternatively, the
interface.python could specify the function signatures itself, but that
starts to stray into WSDL territory, e.g:

<interface.python>
  <function name="myMethod">
     <argument name="arg1" type="string"/>
     <argument name="arg2" type="int"/>
     <return type="string"/>
   </function>
</interface.python>

There may well be a way to specify things a bit tighter in Python that i
don't know about - I'll keep looking into this.

What do people think about these various issues? Has anyone got any
suggestions?

Thanks!

Andy

Hi Andy

Sounds like great progress.

1) I would favour whatever hides SCA from the Python programmer the most so
I like annotations. However if annotations are hard then the first approach
would be a stake in the ground until you have a better feel for how the
extension framework operates.

2) PHP is loosly typed also but you can get at the comments in the source so
you can describe function signatures there and look at them at run time (see
http://www.osoa.org/display/Main/PHP+and+SCA+White+Paper). Can you do the
same in Python?

If you do use support in Python to get at types its likely to be a run time
operation so we would need to rely on the generic proxies/wrappers in C++
and then call up into Python doing the apporpriate transformation as
dictated by the runtime generated interface description. As this is likely
to be a runtime thing I would go with WSDL to describe the interface. You
then have the mechanism for describing the python script as a remote service
ready to go.  It would be good to have a component in C++ that is able to
generate WSDL on the fly (this would be handy for PHP also) but we would
have to agree how to describe the interface to the WSDL generator!

Maybe you could say a little bit more about what you do now to help us
envisage the development process and flow of control.
  - you write a python class/operation
  - you write your .componentType file including implementation.python
  - you hand craft a .h file to match the python class ?
  - you run SCA gen to generate the wrapper
  - you have some code that converts the wrapper call into a call into the
Python engine targetted at the appropriate operation

Is that correct?

Regards

Simon

Reply via email to