Re: [Swig-user] passing Python buffer/string to C++ function as void *

[Angus M. <am...@gm...>]

2009/8/26 Josh Cherry <jc...@nc...>:
> On Wed, 26 Aug 2009, Angus McMorland wrote:
>
>> This really seems like a poor-man's solution to this problem, which
>> should be solvable without constructing cases for each structure type.
>
> I wish there were a better way--perhaps there is, and perhaps one could be
> added--but I've had success using this approach.
>
>> The magic underlying the ".this" Python attribute of swig-wrapped
>> structures provides a valid void * pointer that SetData can use. The
>> only limitation of this approach is that I can't work out how to find
>> out the C-size of the structure in Python. Alternatively, using ctypes
>> I can get the size, but not the correct pointer to the data. Is there
>> a way to get the equivalent of swig's ".this" attribute for a
>> ctypes-wrapped structure? This could then be passed to a function like
>> SetData, swig-wrapped function, and expecting a "void *" pointer.
>
> I think so.  My recollection is that you can get an integer representation
> of a ctypes poiner, in which case this is possible.

I'm not sure that we're looking for an integer representation: in
Python, SWIG's ".this" attribute is a <Swig Object of type
'MDF_TEST_DATA *' at ...>, which seems like some special thing that
swig uses to keep track of its objects. I looked inside the swig
wrapper code, to see how ".this" was generated, but it comes from the
binary .so generated during the swig build, so I follow that lead any
further.

> You could, for example,
> add a function to your interface that takes an integral type and returns it
> as a void *.  But why is this necessary?  If you can get the size from
> ctypes, can't you use that along with a SWIG representation of a pointer?

Yes, in principle I have a workable solution now, creating, in Python,
both a swig-wrapped structure populated with the data which I can
pass, and a second ctypes-version of the structure from which I can
work out the size. This, however, seems incredibly clunky, and I was
thinking there should be a more elegant solution. Creating only one,
ctypes-version, structure, and passing a pointer to _its_ data and
getting _its_ size, seemed like one such cleaner solution. I'm
surprised this issue doesn't come up more often.

Thanks for your thoughts.

Angus.
-- 
AJC McMorland
Post-doctoral research fellow
Neurobiology, University of Pittsburgh