Re: [pygccxml-development] Two patches proposal

[Roman Y. <rom...@gm...>]

On Wed, Dec 23, 2009 at 1:07 PM, Berserker <ber...@ho...> wrote:
> Hi, I'd like to discuss about two issues

Hello

> 1) properties generation
>
> I have some classes like this one:
>
> class Foo
> {
>     Bar getBar() const;
>     void setBar(const Bar &bar);
> };
>
> getBar returns a temporary object because the return value is created by a
> conversion
> from an internal stored string (setBar saves the value in the same way) .
> In this case name_based_recognizer_t creates a "readonly" property "Bar"
> instead of a read/write one,
> so I'm overriding the method in this way:
>
> def check_type_compatibility(self, fget, fset):
>     if decl_wrappers.name_based_recognizer_t.check_type_compatibility(self,
> fget, fset):
>         return True
>
>     t1 = fget.return_type
>     t2 = fset.arguments[0].type
>
>     if declarations.is_reference(t2):
>         t2 = declarations.remove_cv( declarations.remove_reference( t2 ) )
>         if declarations.is_same( t1, t2 ):
>             return True
>
>     return False
>
> What about this change? Is it reasonable to be "merged" in the svn?

The short answer is no :-(. The main reason for it: such code does not
meat user expectations. Let me explain:

foo = Foo()
foo.Bar.x = 2
assert foo.Bar.x == 2

The following is explanation how I thought "add_properties"
functionality should be used in such case.

class custom_name_based_recognizer_t( name_based_recognizer_t ):
    def __init__( ... ): ...

    def check_type_compatibility(self, fget, fset):
         <your code >

cls = mb.class_( ... )
cls.add_properties( recognizer=custom_name_based_recognizer_t() )

Could it be an acceptable solution?

> 2) Implicit conversions:
>
> class Foo
> {
> public:
>     Foo(int v, bool flag = true);   // convertible from int
>     Foo(double v1, double v2);    // not convertible from double because v2
> hasn't default value
> };
>
> I looked at constructor_t.does_define_implicit_conversion implementation and
> I think that
> the following two lines are "incomplete":
>
> if 1 != len( self.arguments ):
>     return False
>
> Isn't supposed to be:
>
> if 1 != len( self.arguments ) and (arguments > 1 haven't default value):
>     return False
>
> Using the above patch Foo should be implicit convertible from int right?

I am not sure. According to the MSDN (
http://msdn2.microsoft.com/en-us/library/h1y7x448.aspx ) "...C++
constructors that have just one parameter automatically perform
implicit type conversion. ...". In your case the constructor has 2
arguments. I tested the following code with gcc 4.4.1

struct Foo{
    Foo( int i, bool b=false ){}
};

void dosmth( Foo ){
}

int main( int argc, const char* argv[] ){
    dosmth( 11 );
};

and it compiles fine.

I don't know C++ at the level of standard. Do you know what the
standard says? Does the code compile fines with MSVC compiler?

> How can I write it?

It is not a problem to write it:

if 1 != len( self.required_args ):
    return False

> Any hope to apply the code to the svn?

The first issue - I believe you have a solution, the second one - I
prefer to know the answers. If they are positive - I will apply the
patch almost immediately.

-- 
Roman Yakovenko
C++ Python language binding
http://www.language-binding.net/