[PyKDE] Sip problem with narrowed return types

[Phil Thompson]

This is now fixed in the current SIP 4 snapshot. I'm not sure it's fixed for 
all similar cases - let me know if you still have problems.

Phil

On Thursday 26 August 2004 4:46 pm, Lőrentey Károly wrote:
> Hi,
>
> (I'm sorry if this is a known issue, I could not find it mentioned in
> the list archives.)
>
> To try my hand on writing SIP specifications, I chose to write a
> simple Python interface to Scott Wheeler's TagLib.  Unfortunately I
> met an obstacle when I came across a situation similar to the
> following:
>
>         /* File test.h */
>
>  #ifndef TEST_H
>  #define TEST_H
>
>  class Test
>  {
>  public:
>    virtual Test *foo ();
>  };
>
>  class Child : public Test
>  {
>  public:
>    virtual Child *foo ();  /* Note the return value! */
>  };
>
>  #endif /* TEST_H */
>
> That is, Child::foo overrides Test::foo, but changes the type of the
> returned value to a subclass of the original type.  (See TagLib's
> TagLib::File::audioProperties vs. TagLib::MPEG::File::audioProperties.)
>
> I tried to write a SIP specification for this class hierarchy, but the
> narrowed return type of Child::foo seems to have confused sip.  Here
> is my (simple-minded) attempt:
>
>         /* File test.sip */
>  %Module test 0
>
>  class Test
>  {
>
>  %TypeHeaderCode
>  #include "test.h"
>  %End
>
>  public:
>    virtual Test *foo ();
>  };
>
>  class Child : Test
>  {
>
>  %TypeHeaderCode
>  #include "test.h"
>  %End
>
>  public:
>    virtual Child *foo ();
>  };
>
> Running `sip -c . test.sip' produces the following code in
> siptestChild.h:
>
>         ...
>  class sipChild : public Child
>  {
>  public:
>                 ...
>   /*
>    * There is a public method for every virtual method visible from
>    * this class.
>    */
>   Test * foo();               /* <---- Invalid */
>   Child * foo();
>                 ...
>  };
>         ...
>
> That is, sip fails to recognize that Child::foo overrides Test::foo,
> and tries to produce code for both.  Unfortunately, this is not valid
> C++ code.  The `Test * foo ()' declaration has the same signature as
> `Child * foo ()', therefore the C++ compiler signals an error.
>
> Is this a bug in sip, or is there something obvious missing from my
> specification?