[C++-sig] Questions about doing advanced things with pyplusplus

[Matthias Baas]

Roman Yakovenko wrote:
>> q1: How do I rename a method, class, or namespace?
> [...]
  > x.alias = "new_name"
> 
>> q2: How do I finalize a virtual class or method (so it doesn't get an
>> polymorphic wrapper)?
> 
> x.finalize()
> [...]
> 
>> q3: How do I set a call policy on a method?  (can you provide example
>> code to do this directly)
> 
> x.call_policies = code_creators.return_value_policy(
> code_creators.return_by_value )
> [...]
> 
>> q4: How do I add support for smart ptrs and holders?  (like pyste's
>> use_shared_ptr() and hold_with_shared_ptr())
> [...]
> x.held_type = smart_ptr #string that will be added/generated

Isn't there sort of a duality between the declaration tree and the code 
creator tree? From what I understand at the moment, at least every 
exposed node in the declaration tree will have a corresponding node in 
the code creator tree (is this correct?).
Then I think it would also be possible (and useful) to attach all of the 
above things to the declaration tree and pass them to the constructor of 
the corresponding code creator. This way I don't have to search for 
virtually the "same" node twice (once in the declaration tree and once 
in the code creator tree).

 >> q5: How do I add methods?  (can you provide example code to do this)
 >
 > Take a look on py_date_time example, _customize_classes method.
 > http://tinyurl.com/e44kh

This is extremely low level, isn't it? Aren't you virtually bypassing 
pyplusplus if you directly set things like

  def( "min", &%s::min, bp::default_call_policies() )

pyplusplus then has no chance of knowing the semantics of this text. 
What if this definition should be involved in the reordering process? Is 
it really guaranteed that the Boost.Python namespace is always called 
"bp"? I couldn't use the standard mechanism for assigning doc strings 
(once there is one).
I would prefer a solution where I can manipulate such a new method just 
as if it was part of the C++ class (hence the suggestion for an 
add_method() function).

- Matthias -