Using InvokeMethod to circumvent compile-time type checking

17 Apr 2006 7:54 PM

Jeff Stewart

I've got several classes that are all concrete implementations of an
abstract class. I have a separate comparer class full of AreEqual()
methods, each with a signature that accepts a pair of concrete classes:

AreEqual(ConcreteA x, ConcreteA y)
AreEqual(ConcreteB x, ConcreteB y)
etc.

I've got another method designed to compare arrays of concrete types.
I'd like to have a single method,

AreEqual(Abstract[] x, Abstract[] y)

that calls the appropriate AreEqual() function above for all
permutations of the elements in the arrays.

Of course, the compiler wouldn't let me do this at compile-time, 'cause
even though the arrays are full of concrete instances, it can't perform
the (widening?) conversion from Abstract to ConcreteA or ConcreteB or
whatever the type may be.

So I started investigating reflection, and came up with this solution:
I created an additional AreEqual() method,

public bool AreEqual(Abstract x, Abstract y) {
Type t = this.GetType();
return (bool)t.InvokeMember("AreEqual", BindingFlags.InvokeMethod,
null, this, new Object[] {x, y});
}

And it works fine! I use that method in my array comparison and it
seems to work like a charm. But I'm worried that this was too easy,
and that I'm missing something. (The BindingFlags don't all seem to be
very clear.) Should it really be so trivial to bypass compile-time
type checking? Am I "cheating" doing this?

--
Jeff S.

18 Apr 2006 3:22 AM

Cowboy (Gregory A. Beamer) - MVP

Abstract classes can serve you well, but accepting an instance of an abstract
class merely to get around a contract is not a wise move (if this is your
motivation). This is the biggest danger I can see.

Another potential issue here is having to call methods that are outside of
the abstract class (derived class specific methods). In these cases, you end
up booting polymorphism and end up with a tightly coupled implementation that
has the false appearance of being loosely coupled. Rather than deleting
dependencies, you are obfuscating them

Those are two potential gotchas I can see.

--
Gregory A. Beamer
MVP; MCP: +I, SE, SD, DBA

***************************
Think Outside the Box!
***************************

Show quote

"Jeff Stewart" wrote:

> I've got several classes that are all concrete implementations of an
> abstract class. I have a separate comparer class full of AreEqual()
> methods, each with a signature that accepts a pair of concrete classes:
>
>   AreEqual(ConcreteA x, ConcreteA y)
>   AreEqual(ConcreteB x, ConcreteB y)
>   etc.
>
> I've got another method designed to compare arrays of concrete types.
> I'd like to have a single method,
>
>   AreEqual(Abstract[] x, Abstract[] y)
>
> that calls the appropriate AreEqual() function above for all
> permutations of the elements in the arrays.
>
> Of course, the compiler wouldn't let me do this at compile-time, 'cause
> even though the arrays are full of concrete instances, it can't perform
> the (widening?) conversion from Abstract to ConcreteA or ConcreteB or
> whatever the type may be.
>
> So I started investigating reflection, and came up with this solution:
> I created an additional AreEqual() method,
>
>   public bool AreEqual(Abstract x, Abstract y) {
>     Type t = this.GetType();
>     return (bool)t.InvokeMember("AreEqual", BindingFlags.InvokeMethod,
> null, this, new Object[] {x, y});
>   }
>
> And it works fine! I use that method in my array comparison and it
> seems to work like a charm. But I'm worried that this was too easy,
> and that I'm missing something. (The BindingFlags don't all seem to be
> very clear.) Should it really be so trivial to bypass compile-time
> type checking? Am I "cheating" doing this?
>
> --
> Jeff S.
>
>

18 Apr 2006 1:21 PM

Jeff Stewart

My motivation is two-fold: 1) don't implement a separate AreEqual()
overload for every possible combination of -arrays- of elements,
because the array comparison algorithm is common to all scenarios, and
2) facilitate extensibility by allowing inherited comparer classes to
define their own AreEqual() methods for any concrete classes that may
be invented in the future. Again, the array comparison algorithm has
only one foreseeable implementation, so it seems natural to want to
reuse it as much as possible.

I'm not clear on your second point about obfuscating dependencies.
Though I don't foresee using this technique on derived-class-specific
methods in this project, if I decided to in another, how does this
tight coupling you're talking about manifest itself? To me, calling
this InvokeMember mechanism seems very flexible. (Indeed, it seemed so
flexible that I worried it might be too flimsy.)

--
Jeff S.