views:

662

answers:

6

Occasionally I like to spend some time looking at the .NET code just to see how things are implemented behind the scenes. I stumbled upon this gem while looking at the String.Equals method via Reflector.

C#

[ReliabilityContract(Consistency.WillNotCorruptState, Cer.MayFail)]
public override bool Equals(object obj)
{
    string strB = obj as string;
    if ((strB == null) && (this != null))
    {
        return false;
    }
    return EqualsHelper(this, strB);
}

IL

.method public hidebysig virtual instance bool Equals(object obj) cil managed
{
    .custom instance void System.Runtime.ConstrainedExecution.ReliabilityContractAttribute::.ctor(valuetype System.Runtime.ConstrainedExecution.Consistency, valuetype System.Runtime.ConstrainedExecution.Cer) = { int32(3) int32(1) }
    .maxstack 2
    .locals init (
        [0] string str)
    L_0000: ldarg.1 
    L_0001: isinst string
    L_0006: stloc.0 
    L_0007: ldloc.0 
    L_0008: brtrue.s L_000f
    L_000a: ldarg.0 
    L_000b: brfalse.s L_000f
    L_000d: ldc.i4.0 
    L_000e: ret 
    L_000f: ldarg.0 
    L_0010: ldloc.0 
    L_0011: call bool System.String::EqualsHelper(string, string)
    L_0016: ret 
}

What is the reasoning for checking this against null? I have to assume there is purpose otherwise this probably would have been caught and removed by now.

+30  A: 

I assume you were looking at the .NET 3.5 implementation? I believe the .NET 4 implementation is slightly different.

However, I have a sneaking suspicion that this is because it's possible to call even virtual instance methods non-virtually on a null reference. Possible in IL, that is. I'll see if I can produce some IL which would call null.Equals(null).

EDIT: Okay, here's some interesting code:

.method private hidebysig static void  Main() cil managed
{
  .entrypoint
  // Code size       17 (0x11)
  .maxstack  2
  .locals init (string V_0)
  IL_0000:  nop
  IL_0001:  ldnull
  IL_0002:  stloc.0
  IL_0003:  ldloc.0
  IL_0004:  ldnull
  IL_0005:  call instance bool [mscorlib]System.String::Equals(string)
  IL_000a:  call void [mscorlib]System.Console::WriteLine(bool)
  IL_000f:  nop
  IL_0010:  ret
} // end of method Test::Main

I got this by compiling the following C# code:

using System;

class Test
{
    static void Main()
    {
        string x = null;
        Console.WriteLine(x.Equals(null));

    }
}

... and then disassembling with ildasm and editing. Note this line:

IL_0005:  call instance bool [mscorlib]System.String::Equals(string)

Originally, that was callvirt instead of call.

So, what happens when we reassemble it? Well, with .NET 4.0 we get this:

Unhandled Exception: System.NullReferenceException: Object
reference not set to an instance of an object.
    at Test.Main()

Hmm. What about with .NET 2.0?

Unhandled Exception: System.NullReferenceException: Object reference 
not set to an instance of an object.
   at System.String.EqualsHelper(String strA, String strB)
   at Test.Main()

Now that's more interesting... we've clearly managed to get into EqualsHelper, which we wouldn't have normally expected.

Enough of string... let's try to implement reference equality ourselves, and see whether we can get null.Equals(null) to return true:

using System;

class Test
{
    static void Main()
    {
        Test x = null;
        Console.WriteLine(x.Equals(null));
    }

    public override int GetHashCode()
    {
        return base.GetHashCode();
    }

    public override bool Equals(object other)
    {
        return other == this;
    }
}

Same procedure as before - disassemble, change callvirt to call, reassemble, and watch it print true...

Note that although another answers references this C++ question, we're being even more devious here... because we're calling a virtual method non-virtually. Normally even the C++/CLI compiler will use callvirt for a virtual method. In other words, I think in this particular case, the only way for this to be null is to write the IL by hand.


EDIT: I've just noticed something... I wasn't actually calling the right method in either of our little sample programs. Here's the call in the first case:

IL_0005:  call instance bool [mscorlib]System.String::Equals(string)

here's the call in the second:

IL_0005:  call instance bool [mscorlib]System.Object::Equals(object)

In the first case, I meant to call System.String::Equals(object), and in the second, I meant to call Test::Equals(object). From this we can see three things:

  • You need to be careful with overloading.
  • The C# compiler emits calls to the declarer of the virtual method - not the most specific override of the virtual method. IIRC, VB works the opposite way
  • object.Equals(object) is happy to compare a null "this" reference

If you add a bit of console output to the C# override, you can see the difference - it won't be called unless you change the IL to call it explicitly, like this:

IL_0005:  call   instance bool Test::Equals(object)

So, there we are. Fun and abuse of instance methods on null references.

If you've made it this far, you might also like to look at my blog post about how value types can declare parameterless constructors... in IL.

Jon Skeet
/me grabs some popcorn and waits for the show.
Greg
I don't have .NET 4 on this machine. It would be interesting to see what the implementation looks like in that version. Interesting hypothesis. Let us know what you find out.
Brian Gideon
Fascinating. So a hardcore IL programmer could use an API I develop incorrectly and the only way I could guard against it is to check `this != null` on every instance method? Just think about the implications if said library were to be security related!
Brian Gideon
Freakin' Jon Skeet. I knew you were crazy and good at everything, but this is beyond ridiculously amazing. I'd take my hat off to you, but I'm not wearing one. Here's an upvote instead, you cheeky no-goodnik you.
Platinum Azure
@Brian, My guess is that the jitter adds a check for non-null arg0 in .NET4 for call, whereas .NET2 only performed this check for callvirt (where it was definitely necessarily, since it needed to get at the virtual method table.)
Dan Bryant
@Dan: No, I believe the check for non-nullity is performed in IL in the .NET 4 code, and the only reason we're not seeing it is inlining.
Jon Skeet
@Brian: I believe you'll get a `NullReferenceException` as soon as you try to access a field - or indeed call another instance method from the one which has been infiltrated (as that would have been compiled with the C# compiler).
Jon Skeet
@Jon: True, I guess as the developer of said library you'd have to make sure you allow that `NullReferenceException` to propagate. Catching it could, theoretically anyway, lead to granting access to resource that was suppose to be denied if the class had some seriously faulty architecture or code. It would be an interesting exercise to see if you could wreak some serious havoc on a poorly designed class using this exploit.
Brian Gideon
+1  A: 

Let's see... this is the first string you're comparing. obj is the second object. So it looks like it's an optimization of sorts. It's first casting obj to a string type. And if that fails, then strB is null. And if strB is null while this isn't, then they're definitely not equal and the EqualsHelper function can be skipped.

That will save a function call. Beyond that, perhaps a better understanding of the EqualsHelper function might shed some light on why this optimization is needed.

EDIT:

Ah, so the EqualsHelper function is accepting a (string, string) as parameters. If strB is null, then that essentially means that it was either a null object to begin with, or it couldn't successfully be cast into a string. If the reason for strB being null is that the object was a different type that couldn't be converted to a string then you wouldn't want to call EqualsHelper with essentially two null values (that'll return true). The Equals function should return false in this case. So this if statement is more than an optimization, it actually ensures proper functionality as well.

Steve Wortham
+7  A: 

The reason why is that it is indeed possible for this to be null. There are 2 IL op codes which can be used to invoke a function: call and callvirt. The callvirt function causes the CLR to perform a null check when invoking the method. The call instruction does not and hence allows for a method to be entered with this being null.

Sound scary? Indeed it is a bit. However most compilers ensure this doesn't ever happen. The .call instruction is only ever outputted when null is not a possibility (I'm pretty sure that C# always uses callvirt).

This isn't true for all languages though and for reasons I don't exactly know the BCL team chose to further harden the System.String class in this instance.

Another case where this can popup is in reverse pinvoke calls.

JaredPar
+7  A: 

The short answer is that languages like C# force you to create an instance of this class before calling the method, but the Framework itself does not. There are two¹ different ways in CIL to call a function: call and callvirt.... Generally speaking, C# will always emit callvirt, which requires this to not be null. But other languages (C++/CLI comes to mind) could emit call, which doesn't have that expectation.

(¹okay, it's more like five if you count calli, newobj etc, but let's keep it simple)

Warren
Nope, C++ would emit `callvirt` here as well. It's a virtual method, after all. See my answer.
Jon Skeet
A: 

If the argument (obj) does not cast to a string then strB will be null and the result should be false. Example:

    int[] list = {1,2,3};
    Console.WriteLine("a string".Equals(list));

writes false.

Remember that string.Equals() method is called for any argument type, not only for other strings.

jalexiou
Thats definitely true. Thats actually boiler plate code that is typical of all `Equals` implementations. The real meat of my question was why the test `this != null` is made. The naive assertion is that it is superfluous, but with an incredibly deep knowledge of the CLR and the C# compiler you can understand and truly appreciate the implementation. See the accepted answer for more information!
Brian Gideon
By the way, welcome to Stackoverflow :)
Brian Gideon