[Gdal-dev] NG Python bindings, and SWIG 1.3.29

[Mateusz Loskot]

Frank Warmerdam wrote:
> I'm following up on failures in the autotest suite with the NG Python
> bindings, especially when built with SWIG 1.3.29 (the most recent public
> release of SWIG).  It seems that invoking Destroy() or Release() on an
> ogr.DataSource is *not* causing it to be closed immediately and this is
> causing serious problems for tests that require this.
> 
> A few questions:
>  o The committed python wrappers seem to work substantially better than the
>    ones I generate myself.  Are the committed bindings generated with an
>    older or a newer SWIG than 1.3.29?

Frank,

I'd like to add myself to this question :-)

>  o Are there any suggestions on how to force immediate deletion in the
>    ogr.DataSource destructor?  Currently this just seems to decrement the
>    Python reference count leaving it up to the Python VM to invoke the
>    close function when it pleases.

Do you mean that explicit call of __del__ in Destroy() does not cause
object deletion?
There may be a few reasons of this issue:
- reference to DataSource still present, so then GC won't delete DataSource
- cyclic references between DataSource and o

if a is a DataSource then you may try to call:

del a

but this will *only* decrement reference counter and call __del__ dctor
only if refcounter is 0. So, in following case, __del__ won't be called:

a = MyClass()
b = a
del a # <----

Another option is to cause deletion by assigning None to DataSource
instance:

a = None

Calling __del__ explicitly won't promise that the object is deleted. It
still depends if there are any references existing.

Summarizing, I'd try to assing None at the first place:

def Destroy(self):
    self = None

Cheers
-- 
Mateusz Łoskot
http://mateusz.loskot.net