strange compiler error ( gcc 4.0.2)

[grundmann]

Hello,

i got a strange compiler error.
When compiling the following:

// forward declarations
typedef AvlTree<LineSeg ment,LineSegmen tComperator> LSTree;
void handleEventPoin t (const EventPoint& , LSTree& , double&,
std::list<Inter sectionPoint>& );
// the interesting function
std::list<Inter sectionPoint> findIntersectio ns ( const
std::list<LineS egment*>& segments)
{
// need two data structers
AvlTree<EventPo int> eventQueue;
double offset;
LineSegmentComp erator lsc(&offset);
LSTree sweepStatus ( lsc );

//output
std::list<Inter sectionPoint> output;

// ripped some lines here

while ( ! eventQueue.isEm pty () )
{
EventPoint p = eventQueue.min( );
eventQueue.dele teElem(p);
// pass sweepStatus by reference
50: handleEventPoin t (p, sweepStatus, offset, output );
}

return output;
}

everything works fine.
But if I replace the following lines:

LineSegmentComp erator lsc(&offset);
LSTree sweepStatus ( lsc );

with this shorter version

LSTree sweepStatus ( LineSegmentComp erator(&offset) );

I always got this strange error from gcc:
geometricalgos. cpp: In function 'std::list<Inte rsectionPoint,
std::allocator< IntersectionPoi nt> > findIntersectio ns(const
std::list<LineS egment*, std::allocator< LineSegment*> >&)':
geometricalgos. cpp:50: error: invalid initialization of non-const
reference of type 'LSTree&' from a temporary of type 'LSTree
(*)(LineSegment Comperator&)'
geometricalgos. cpp:9: error: in passing argument 2 of 'void
handleEventPoin t(const EventPoint&, LSTree&, double&,
std::list<Inter sectionPoint, std::allocator< IntersectionPoi nt> >&)'

I don't understand this. My Object sweepStatus is not temporary, there
is no type-conversion, nor any creation of a temporary instance of
LSTree. The only thing changed is, that I pass now a temporary object
for constructor of LSTree. But the constructor is defined as follows:

template <class T, class Compare>
AvlTree<T,Compa re>::AvlTree (const Compare& comp ) : root_ (0),
comp_(comp)
{

};

so the passed Compare-Object is copied to the internal member variable
comp_, which is of type Comp.

Does anybody know, where the problem is?

Best Regards,

Matthias Grundmann

[Ron Natalie]

grundmann wrote:

LSTree sweepStatus ( lsc );

What is LineSegmentComp arater defined to be?

Somehow it would appear that sweepStatus is being
interpretted as a function declaration. This is
possible if lsc somehow itself was a type name.

[John Harrison]

grundmann wrote:

Hello,

i got a strange compiler error.
When compiling the following:

// forward declarations
typedef AvlTree<LineSeg ment,LineSegmen tComperator> LSTree;
void handleEventPoin t (const EventPoint& , LSTree& , double&,
std::list<Inter sectionPoint>& );
// the interesting function
std::list<Inter sectionPoint> findIntersectio ns ( const
std::list<LineS egment*>& segments)
{
// need two data structers
AvlTree<EventPo int> eventQueue;
double offset;
LineSegmentComp erator lsc(&offset);
LSTree sweepStatus ( lsc );

//output
std::list<Inter sectionPoint> output;

// ripped some lines here

while ( ! eventQueue.isEm pty () )
{
EventPoint p = eventQueue.min( );
eventQueue.dele teElem(p);
// pass sweepStatus by reference
50: handleEventPoin t (p, sweepStatus, offset, output );
}

return output;
}

everything works fine.
But if I replace the following lines:

LineSegmentComp erator lsc(&offset);
LSTree sweepStatus ( lsc );

with this shorter version

LSTree sweepStatus ( LineSegmentComp erator(&offset) );

I always got this strange error from gcc:
geometricalgos. cpp: In function 'std::list<Inte rsectionPoint,
std::allocator< IntersectionPoi nt> > findIntersectio ns(const
std::list<LineS egment*, std::allocator< LineSegment*> >&)':
geometricalgos. cpp:50: error: invalid initialization of non-const
reference of type 'LSTree&' from a temporary of type 'LSTree
(*)(LineSegment Comperator&)'
geometricalgos. cpp:9: error: in passing argument 2 of 'void
handleEventPoin t(const EventPoint&, LSTree&, double&,
std::list<Inter sectionPoint, std::allocator< IntersectionPoi nt> >&)'

I don't understand this. My Object sweepStatus is not temporary, there
is no type-conversion, nor any creation of a temporary instance of
LSTree. The only thing changed is, that I pass now a temporary object
for constructor of LSTree. But the constructor is defined as follows:

Compiler (correctly) thinks that

LSTree sweepStatus ( LineSegmentComp erator(&offset) );

is a function prototype.

Try this

LSTree sweepStatus = LSTree(LineSegm entComperator(& offset));

john

[grundmann]

Why should the compiler think, that
LSTree sweepStatus ( LineSegmentComp erator(&offset) );
is a function prototype. Isn't it a normal declaration of a object with
calling the constructor passing the needed parameters?

I also tried your syntax before, confusing about that gcc claimed that
the copy-constructor is private. I have done this in the class AvlTree,
because copying should not be allowed. But the "Type obj = Type (
....);" syntax should not invoke the copy-constructor. The compiler
should recognise, that the rhs Object is temporary and should call the
constructor instead. I am pretty sure, that MS VC-Compiler is working
this way, but I am not sure about gcc.

@Ron:
The class LineSegmentComp erator is implemented as follows:

class LineSegmentComp erator
{
public:
explicit LineSegmentComp erator ( double* offset =0) : offset_ (offset)
{}
bool operator()(cons t LineSegment& lhs, const LineSegment& rhs);
protected:
double* offset_;
};

I am not yet understanding the error.

Best regards,

Matthias Grundmann

John Harrison schrieb:

Compiler (correctly) thinks that

LSTree sweepStatus ( LineSegmentComp erator(&offset) );

is a function prototype.

Try this

LSTree sweepStatus = LSTree(LineSegm entComperator(& offset));

john

[Ben Pope]

grundmann wrote:

Why should the compiler think, that
LSTree sweepStatus ( LineSegmentComp erator(&offset) );
is a function prototype. Isn't it a normal declaration of a object with
calling the constructor passing the needed parameters?

I also tried your syntax before, confusing about that gcc claimed that
the copy-constructor is private. I have done this in the class AvlTree,
because copying should not be allowed. But the "Type obj = Type (
....);" syntax should not invoke the copy-constructor. The compiler
should recognise, that the rhs Object is temporary and should call the
constructor instead. I am pretty sure, that MS VC-Compiler is working
this way, but I am not sure about gcc.

The compiler is allowed to optimise away the copy, but the fact remains that the syntax:

A a = A(...);

Is akin to doing:
A a(A(...));

And so, the compiler must act as if the copy constructor is required, but it is private.

Ben
--
I'm not just a number. To many, I'm known as a String...

[grundmann]

I tried this and you are right, allowing the Type AvlTree to be
copy-constructable make the compiler able to accept without any errors.
But the initial expression

LSTree sweepStatus (LineSegmentCom perator (&offset));

further leads to the error, stated above. Why?

Matthias Grundmann

[John Harrison]

grundmann wrote:

I tried this and you are right, allowing the Type AvlTree to be
copy-constructable make the compiler able to accept without any errors.
But the initial expression

LSTree sweepStatus (LineSegmentCom perator (&offset));

further leads to the error, stated above. Why?

Matthias Grundmann

Because, as I stated, the above is a function prototype. Nothing to do
with copy constructors.

john

[John Harrison]

John Harrison wrote:

grundmann wrote:

I tried this and you are right, allowing the Type AvlTree to be
copy-constructable make the compiler able to accept without any errors.
But the initial expression

LSTree sweepStatus (LineSegmentCom perator (&offset));

further leads to the error, stated above. Why?

Matthias Grundmann

Because, as I stated, the above is a function prototype. Nothing to do
with copy constructors.

Presumably you have no trouble accepting

LSTree sweepStatus(Lin eSegmentCompera tor &offset);

as a function prototype. Presumably you also think that the extra pair
of parens makes a difference but it does not.

john

[grundmann]

Thank you for your help. I tried this once again with intel c++ 9.0 and
got the same error. I have always thought that LineSegmentComp erator
(&offset) would be interpreted in any situations as an constructor call
returning an object and not as an declaration of a reference of type
LineSegmentComp erator. So I was wrong.