Reflection and Nullable
You might have already read
this: VS2005 made the last-minute DCR related to boxed Nullable<T>.
Runtime now treats Nullable<T> differently from other generic value types
when boxing:
Int32? x = null;
object y = x;
// y is simply null, not a truly boxed "nullable<int> struct" (which then
tells it has no value)
And
Int32 ? x = 100;
object y = x;
// y is boxed Int32 100 (same as "object y = 100"), no longer
a boxed "nullable<int> struct" with a field of value 100.
Reflection late-bind invocation always deals with "object"; value type instances
are always getting boxed first before invocation. One of such APIs: PropertyInfo.SetValue(Object
obj, Object value, Object[]
index). There was
one question in Microsoft Technical Forums: Exception is thrown when setting
DateTime value to a property with type Nullable<DateTime>. With this DCR change,
such late-bind calls can be written in a more natural way.
class C
{
Int32? m_field;
public Int32?
Property
{
get { return
m_field; }
set { m_field =
value; }
}
}
PropertyInfo pi = typeof(C).GetProperty("Property");
C c = new C();
pi.SetValue(c, 100, null);
// works
pi.SetValue(c, (Int32?)100, null);
// no longer have to write like this
After boxing, 100 and (Int32?)100 are the same thing: boxed Int32 object. There
is no more boxed object with type "Int32?"; late-invocation has to accept it and
set to Int32? property. Or think it this way, runtime can not tell where a boxed
Int32 object was originally from: either just Int32 instance or Int32? instance;
it is reasonable to allow setting a boxed Int32 object to field/property of either
Int32 or Int32? type (same for argument pass-in when doing method invocation)
pi.SetValue(c, null, null);
// works
This actually worked before the DCR. And ,
Object ret = pi.GetValue(c, null);
ret will be either boxed Int32 object or simply null. There is no problem in casting
the result to Int32? type.
Here are some reflection behavior changes as a result of this DCR:
Probably the most important one is "typeof(T?).IsAssignableFrom(typeof(T)) == true".
PropertyInfo.SetValue/GetValue use this first to ensure the pass-in value is assignable
to Nullable<T> property.Object.GetType() will never get back any type of Nullable<T>. Joe had an interesting
post on this.
Activator.CreateInstance could never be expected to return null before; with this
DCR, it will return null when creating instance of type Nullable<T> but not
providing non-null T value. For example, Activator.CreateInstance(typeof(Int32?))
returns null.
Comments
- Anonymous
August 23, 2005
Haibo Luo has a nice description of the changes made by Microsoft to nullable types and boxing for .NET 2.0 RTM: http://blogs.msdn.com/haibo_luo/archive/2005/08/23/455241.aspx This is good news for Sooda, because it would require very little work to properly support nullable types... - Anonymous
August 23, 2005
The comment has been removed - Anonymous
August 28, 2005
Hi barrkel - I talked to Peli (http://blog.dotnetwiki.org/), who tested this for JIT. He will give you an answer to your question. - Anonymous
August 29, 2005
The following snippet should answer naturally both questions
static void Main()
{
int? i = 1;
object o = i; // A
int? j = (int?)o; // B
int k = (int)o; // C
}
.method private hidebysig static void Main() cil managed
{
.entrypoint
// Code size 32 (0x20)
.maxstack 2
.locals init (valuetype [mscorlib]System.Nullable1<int32> V_0,<br> object V_1,<br> valuetype [mscorlib]System.Nullable
1<int32> V_2,
int32 V_3)
IL_0000: nop
IL_0001: ldloca.s V_0
IL_0003: ldc.i4.1
IL_0004: call instance void valuetype [mscorlib]System.Nullable1<int32>::.ctor(!0)<br> IL_0009: nop<br> IL_000a: ldloc.0<br> IL_000b: box valuetype [mscorlib]System.Nullable
1<int32> // A
IL_0010: stloc.1
IL_0011: ldloc.1
IL_0012: unbox.any valuetype [mscorlib]System.Nullable`1<int32> // B
IL_0017: stloc.2
IL_0018: ldloc.1
IL_0019: unbox.any [mscorlib]System.Int32 // C
IL_001e: stloc.3
IL_001f: ret
} // end of method Program::Main