Re: [Swig-devel] To be in

[jiangzhiyang <ja...@gm...>]

William S Fulton wrote:
> jiangzhiyang wrote:
>> Hi all
>> I had written a application for this year's GSOC , my proposal is about
>> "Better Multiple Inheritance Support",unfortunately i am not
>> selected,and William said, he will help me even though i am not a formal
>> student of GSOC, i think it is a good opportunity to be involved in the
>> open source community,also i would like to do something for the SWIG for
>> i am a user of SWIG. what i want to do is as follow:
>> Now there is not a complete support for multiple inheritance when
>> wrapping c++ code for java and c# in SWIG, because there is no direct
>> multiple inheritance mechanism in java and c#, it can be implemented
>> through the interfaces in java and c#, some sample c++ codes like this:
>> //cpp
>> struct CBase1 {
>> virtual int foo1(){
>> return 1;
>> }
>> virtual ~CBase1(){
>> }
>> };
>> struct CBase2{
>> virtual int foo2(){
>> return 2;
>> }
>> virtual ~CBase2(){
>> }
>> };
>> struct ABase1{
>> virtual int bar1()=0;
>> virtual ~ABase1(){
>> }
>> };
>>
>> struct Derived1 : CBase1,CBase2{
>> virtual int foo1(){
>> return 3;
>> }
>> virtual int foo2(){
>> return 4;
>> }
>> };
>> struct Derived2:CBase1,ABase1{
>> virtual int bar1(){
>> return 5;
>> }
>> virtual int foo1(){
>> return 6;
>> }
>>
>> };
>> struct Derived3:CBase1,CBase2,ABase1{
>> virtual int foo1(){
>> return 7;
>> }
>> virtual int foo2(){
>> return 8;
>> }
>> virtual int bar1(){
>> return 9;
>> }
>> virtual ABase1* clone(){
>> return new Derived3(*this);
>> }
>> };
>> For supporting multiple inheritance completely ,firstly a directive like
>> "%feature("interface")" is needed, it is used to show which class in c++
>> can be implemented as interface in java or c#.For the aboved code :
>>
>> // a interface file for multiple interitace
>> %feature("interface") CBase1,CBase2,ABase1
>>
> Actually the %feature syntax requires you to write this in 3 lines:
> %feature("interface") CBase1;
> %feature("interface") CBase2;
> %feature("interface") ABase1;
>
I am sorry about my poor swig syntax.
>> it shows that "CBase1,CBase2,ABase1" can be interfaces in java or c#.
>> Some rules about the %feature("interface") like this:
>> 1.About What does the definition of interface look like?
>> I) If a derived class has two or more base classes, every base class may
>> be implements as a interface of java and a concrete class of java if it
>> is a concrete class in c++ according to the declaration of
>> "%feature("interface")" in interface file.
>> II)If a deprived class has two or more base classes, at least a base
>> class will be implements as a interface according to the declaration of
>> "%feature("interface")" ,others will be interfaces and concrete classes.
>> III)If the base class is a concrete class, the corresponding interfaces
>> are named with a prefix like "SWIGTYPE_interface_"
>> IV)If the classes are abstract classes with definition of method in
>> them, they will be the same as the concrete classes, or they will be
>> just implemented as an interface of java .
>> 2. the rule about which is interface and which is concrete class in 
>> java?
>> Depending the order of base classes.
>> If there is an abstract class in base classes, it has the highest 
>> priority,
>> If there are two more abstract classes as base classes, depending the
>> order of base classes.
>> For example:
>> CBase1,CBase2,ABase1 are all base classes, they can be implemented as
>> interfaces through
>> %feature("interface") CBase1,CBase2,ABase1
>> For Derived1, the java codes may like this :
>> public Derived1 extends CBase1 implements SWIGTYPE_interface_CBase2{
>>
>> . . .
>> }
>> because CBase1 is before CBase2,so a interface
>> "SWIGTYPE_interface_CBase2" is needed,the java codes may like this
>> interface SWIGTYPE_interface_CBase2{
>> int foo2();
>> }
>> also there still is a concrete java class of CBase2 likes:
>> public class CBase2 {
>> . . .
>> }
>>
>> For Derived2, the java codes may like this:
>> public Derived2 extends CBase1 implements ABase1{
>> . . .
>> }
>> interface ABase1 {
>> int bar1();
>> }
>> ABase1 is an abstract class in c++ , it doesn't need to add new
>> interface in java, it is direct map.
>>
>>
>> I have read some source code of swig, and made a java test case in
>> testsuite under William's help .
>> Any suggestions will be appreciated,thanks all in advance.
>
> Hi Jiang
>
> I'm just considering your test cases. I'm not clear on some things...
>
> You mention that there are some rules for what an interface will look 
> like and the priority of class being the base class or an interface. 
> Is this something that a user must know and decide before specifying 
> %feature("interface") or is this something that SWIG will detect? I'd 
> always assumed that a user will specify what is going to be an 
> interface, with maybe SWIG doing some checking that it is a "pure 
> abstract base class". So there is no need for some rules... if a class 
> has multiple base classes the 2nd and subsequent classes get ignored 
> (like they currently do). Unless some of the classes are specified as 
> %feature("interface") in which case the first class in the list that 
> does not have %feature("interface") becomes the base class.
>
My original idea is that SWIG will decide which is interface and which 
is base class according the order of c++ source code,As you said it is 
better specify by a user.
> So using your examples above, if you have:
>
> %feature("interface") ABase1;
>
> and then for c++:
>   struct Derived2:CBase1,ABase1 {...};
> you'd have in java:
>   public class Derived2 extends CBase1 implements ABase1 {...}
>
> and for this c++:
>   struct Derived3:CBase1,CBase2,ABase1 {...};
> the java code would be:
>   public class Derived3 extends CBase1 implements ABase1 {...}
> and swig would warn about CBase2 being ignored.
>
> I'm not sure I like III) where you introduce some SWIGTYPE_interface 
> types. I don't see how it adds any value other than adding complexity. 
> Let's do one thing at a time and map pure abstract base classes into 
> Java interfaces first. Secondly, lets also deal with non pure abstract 
> base classes with a rule that if one is marked with 
> %feature("interface") all the non virtual methods are ignored so that 
> an interface can be generated out of it. For example your CBase2 will 
> turn into:
>
> interface CBase2 {
>  int foo2();
> }
>
> and if you had something like:
>
> %feature("interface") Concrete;
> struct Concrete {
>   int a;
>   virtual void virtualMethod {...}
>   void nonVirtualMethod() {...}
> };
>
> it would map to:
>
> interface Concrete {
>   void virtualMethod;
> }
>
> ie the non virtual members would be ignored. Not a perfect solution 
> for multiple inheritance, but better than ignoring the whole class as 
> is currently done in multiple inheritance.
>
When use " %feature("interface") Concrete; " to map a concrete class  
into a  java interface,if just generate a interface like this, 
interface Concrete {
  void virtualMethod;
}
a user may not instantiate a concrete class directly like this:
Concrete c=new Concrete();
c.nonVirtualMethod();
And the "SWIGTYPE_interface" may be a redundant type the user doesn't 
map a concrete class into a java interface, but may be useful in aboved 
example.
> There is a patch knocking about which detects pure abstract base 
> classes and marks them as such (if I recall right). Have you looked at 
> this patch yet? This could be used to help detect classes that can 
> successfully be used as Java interfaces and warn if it is going to fail.
>
> William
>
Is it in test-suite?

Jiang