Re: C++ to Python and returning members of a C++ object

Hi Jake,

Ah, just worked it out. You must be using the swig "-builtin" option. I 
have modified the code to use "self" in that case.
Next item: I need to try and workout how not to call 
SwigContainerReference() when returning base types such as int for 
efficiency.

Terry
On 23/09/2020 07:34, Terry Barnaby wrote:
> Hi Jake,
>
> Ok, I will look at doing a pull once it is in a state to do so.
>
> I was intending to only use the reference counting system for member 
> gets for now and only when the containerReference feature is enabled 
> so it won't affect any other code (apart from the slight increase in 
> memory usage).
> However, I can look at using if for the reference counting in your 
> container code if people think it is worth going further with it ?
>
> I'm not sure on your self vs args change, I am missing something here, 
> are you using some particular swig "features" when generating the 
> wrapper? A typical *_get function on my wrapper looks like it needs 
> the args, the code looks like:
>> SWIGINTERN PyObject *_wrap_ClassB_b_get(PyObject 
>> *SWIGUNUSEDPARM(self), PyObject *args) {
>>   PyObject *resultobj = 0;
>>   ClassB *arg1 = (ClassB *) 0 ;
>>   void *argp1 = 0 ;
>>   int res1 = 0 ;
>>   PyObject *swig_obj[1] ;
>>   ClassA *result = 0 ;
>>
>>   if (!args) SWIG_fail;
>>   swig_obj[0] = args;
>>   res1 = SWIG_ConvertPtr(swig_obj[0], &argp1,SWIGTYPE_p_ClassB, 0 |  0 );
>>   if (!SWIG_IsOK(res1)) {
>>     SWIG_exception_fail(SWIG_ArgError(res1), "in method '" 
>> "ClassB_b_get" "', argument " "1"" of type '" "ClassB *""'");
>>   }
>>   arg1 = reinterpret_cast< ClassB * >(argp1);
>>   result = (ClassA *)& ((arg1)->b);
>>   resultobj = SWIG_NewPointerObj(SWIG_as_voidptr(result), 
>> SWIGTYPE_p_ClassA, 0 |  0 );
>>   SwigContainerReference(args, resultobj); // Increment reference 
>> count of the container object and add a pointer to the container to 
>> this in the returned object
>>   return resultobj;
>> fail:
>>   return NULL;
>> }
> My current patch and example is attached.
>
> Terry
> On 22/09/2020 21:49, Jake wrote:
>> Terry,
>>
>> You should open a pull request on Github with your proposed change.  
>> William or another maintainer will have to accept it to get in.
>>
>> I'm not a maintainer, but I would recommend the following:
>>
>> Replace my implementation of the container owner function by your 
>> own, but keep the template specialization so the container code works 
>> as before.
>> In your insertion points, use swig::pointer_category as the template 
>> argument so the ref-counting specialization is called.
>>
>> Yes, I replaced 'args' with 'self' where you showed. 'self' is the 
>> wrapper for the parent instance, 'args' is a tuple of function 
>> arguments to be unpacked.
>>
>> When you have in Python:
>>
>> m = Mixed()
>> o = m.obj
>>
>> And Mixed.obj is some wrapped type (e.g. not a string or other value 
>> that gets copied), look at the generated code:
>>
>> 6903 SWIGINTERN PyObject *_wrap_Mixed_obj_get(PyObject *self, 
>> PyObject *args) {
>>
>> 6904 PyObject *resultobj = 0;
>>
>> 6905 Mixed *arg1 = (Mixed *) 0;
>>
>> 6906 void*argp1 = 0;
>>
>> 6907 intres1 = 0;
>>
>> 6908 PyObject *swig_obj[1] ;
>>
>> 6909 MyObject *result = 0;
>>
>> 6910
>>
>> 6911 if(!SWIG_Python_UnpackTuple(args, "Mixed_obj_get", 0, 0, 0)) 
>> SWIG_fail;
>>
>> 6912 res1 = SWIG_ConvertPtr(self, &argp1,SWIGTYPE_p_Mixed, 0|0);
>>
>> 6913 if(!SWIG_IsOK(res1)) {
>>
>> 6914 SWIG_exception_fail(SWIG_ArgError(res1), "in method 
>> '""Mixed_obj_get""', argument ""1"" of type '""Mixed *""'");
>>
>> 6915 }
>>
>> 6916 arg1 = reinterpret_cast< Mixed * >(argp1);
>>
>> 6917 result = (MyObject *)& ((arg1)->obj);
>>
>> 6918 resultobj = SWIG_NewPointerObj(SWIG_as_voidptr(result), 
>> SWIGTYPE_p_MyObject, 0|0);
>>
>> 6919 SwigContainerReference(args, resultobj); // Increment reference 
>> count of the container object and add a pointer to the container to 
>> this in the returned object
>>
>>
>> Without that change I was getting crashes.
>>
>> -Jake
>>
>>
>>
>> On Tue, Sep 22, 2020 at 10:28 AM Terry Barnaby <[email protected] 
>> <mailto:[email protected]>> wrote:
>>
>>     Hi Jake,
>>
>>     I have been away from this for a while with other work.
>>
>>     You stated:
>>     On 19/08/2020 09:36, Terry Barnaby wrote:
>>     > It segfaults on pointer-category objects because you should be
>>     passing
>>     > "self" as the first argument not "args", I corrected that for the
>>     > benchmarks above.
>>
>>     Not sure what you mean here, the only place I use "args " is in:
>>
>>         Printf(f->code, "SwigContainerReference(args, resultobj); //
>>     Increment reference count of the container object and add a
>>     pointer to
>>     the container to this in the returned object\n");
>>
>>     And this seems correct, certainly I can't use "self" in place of
>>     "args"
>>     here as the code won't compile. May I ask what change you made here ?
>>
>>     I'm not sure where the SWIG group want to go on this. For the
>>     projects
>>     I'm involved in I need this fix to eliminate the memory faults when
>>     returning C++ members in Python so I will be creating my own local
>>     version of SWIG, with my patch applied, for now. But I am keen to
>>     try
>>     and get something like this included in the main tree if possible.
>>
>>     Terry
>>
>

_______________________________________________
Swig-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/swig-user