Why do managed event handlers still use '&" ?

[Peter Oliphant]

I've upgraded my code to VS C++.NET 2005 (Express) using /clr pure. My
question is, why is there an exception to the rule in how pointer syntax is
done in the event handlers?

For example, why is this proper syntax:

button->Click += gcnew EventHandler( this, &MyClass::MouseHandler ) ;

The '&' seems like a hold over from the old syntax, except this is the NEW
syntax (i.e., it never wanted the '&' before 2005 at all). Why was the '&'
added (it's not like anything BUT a pointer would make sense), and why was
the old syntax 'address of' symbol used? That is, before this was done like
this:

button->Click += new EventHandler( this, MyClass::MouseHandler ) ;

That is, no '&'. Thus, it seems the proper transition would be to this
syntax:

button->Click += gcew EventHandler( this, MyClass::MouseHandler ) ;

This means the only reason to add the '&' is if the line immediately above
could somehow be ambiguous, which I don't think it can be. So adding the
superfulous '&' seems odd.

Further, almost humerously, the compile upon seeing the non-'&' form
generates an error that basically says 'you need to add the '&'". Well, if
it knows this, and there is no ambiguity involved, why not accept BOTH
forms? Or dang it, why not provide the ability to click on such an error
message and have an option for IT make the correction? And possibly allow a
MAKE ALL CORRECTIONS button, letting the compiler do some of the work! Its
really a bit irritating when it says 'this is what you did wrong, this is
how its done' without the ability to fix it automatically (and such a method
would not be prone to typos!)...

[==P==]

 

[Doug Harrison [MVP]]

I've upgraded my code to VS C++.NET 2005 (Express) using /clr pure. My
question is, why is there an exception to the rule in how pointer syntax is
done in the event handlers?

For example, why is this proper syntax:

button->Click += gcnew EventHandler( this, &MyClass::MouseHandler ) ;

The '&' seems like a hold over from the old syntax, except this is the NEW
syntax (i.e., it never wanted the '&' before 2005 at all). Why was the '&'
added (it's not like anything BUT a pointer would make sense), and why was
the old syntax 'address of' symbol used? That is, before this was done like
this:

button->Click += new EventHandler( this, MyClass::MouseHandler ) ;

That is, no '&'. Thus, it seems the proper transition would be to this
syntax:

button->Click += gcew EventHandler( this, MyClass::MouseHandler ) ;

This means the only reason to add the '&' is if the line immediately above
could somehow be ambiguous, which I don't think it can be. So adding the
superfulous '&' seems odd.

Further, almost humerously, the compile upon seeing the non-'&' form
generates an error that basically says 'you need to add the '&'". Well, if
it knows this, and there is no ambiguity involved, why not accept BOTH
forms? Or dang it, why not provide the ability to click on such an error
message and have an option for IT make the correction? And possibly allow a
MAKE ALL CORRECTIONS button, letting the compiler do some of the work! Its
really a bit irritating when it says 'this is what you did wrong, this is
how its done' without the ability to fix it automatically (and such a method
would not be prone to typos!)...

Earlier versions of VC++ accepted just about any syntax to mean
pointer-to-member, which allowed people to make many subtle mistakes. For
example, how is the compiler to know in general that A::F isn't simply the
result of leaving off the parens on a function call? What about a->f? The
C++ Standard says the only way to form a pointer-to-member is with the
&A::F syntax, and that's a good thing.

They've actually fixed a real problem with delegates; it used to be
possible to create a C++ delegate from an unrelated object and
pointer-to-member, but I just translated my old VC7.1 example to C++/CLI,
and the compiler complains appropriately:

TDelegate^ d = gcnew TDelegate(gcnew X, &Y::f);

a.cpp(40) : error C3754: delegate constructor: member function 'Y::f'
cannot be called on an instance of type 'X ^'

In this program, X and Y are unrelated classes, but VC7.1 would allow the
equivalent of the above to pass, causing errors at runtime instead of
compile-time.

 

[Peter Oliphant]

Hi Doug,

That brings up an interesting question. If it is illegal to create a handler
with unrelated objects (which totally makes sense), why is it required to
completely qualify the handler's name? That is, why do we have to do:

Click += gcnew EventHandler( this, &MyClass::Handler ) ;

why isn't this just fine:

Click += gcnew EventHandler( this, &Handler ) ;

[==P==]

 

[Carl Daniel [VC++ MVP]]

Hi Doug,

That brings up an interesting question. If it is illegal to create a
handler with unrelated objects (which totally makes sense), why is it
required to completely qualify the handler's name? That is, why do we have
to do:

Click += gcnew EventHandler( this, &MyClass::Handler ) ;

why isn't this just fine:

Click += gcnew EventHandler( this, &Handler ) ;

Because the simple name 'Handler' is not in scope under normal C++ scoping
rules.

-cd

 

[Doug Harrison [MVP]]

Hi Doug,

That brings up an interesting question. If it is illegal to create a handler
with unrelated objects (which totally makes sense), why is it required to
completely qualify the handler's name? That is, why do we have to do:

Click += gcnew EventHandler( this, &MyClass::Handler ) ;

why isn't this just fine:

Click += gcnew EventHandler( this, &Handler ) ;

It could do that, but it would make for a gratuitous incompatibility with
Standard C++. Remember, in Standard C++, there's exactly one way to form a
pointer to member. Also, qualifying the name with MyClass:: tells the
compiler exactly where in the class hierarchy to start looking for the name
"Handler". Normally, this would be the static type of *this, but it could
also be a base class. You might argue that since there's an obvious choice
of default starting point, the qualifier should be optional, but then we're
back to the "gratuitous incompatibility". I do get the feeling you'd like
the C# syntax better, which is, IIRC, something like
EventHandler(this.Handler).