Why does this string extension method not throw an exception?
You are using yield return
. When doing so, the compiler will rewrite your method into a function that returns a generated class that implements a state machine.
Broadly speaking, it rewrites locals to fields of that class and each part of your algorithm between the yield return
instructions becomes a state. You can check with a decompiler what this method becomes after compilation (make sure to turn off smart decompilation which would produce yield return
).
But the bottom line is: the code of your method won't be executed until you start iterating.
The usual way to check for preconditions is to split your method in two:
public static IEnumerable<int> AllIndexesOf(this string str, string searchText)
{
if (str == null)
throw new ArgumentNullException("str");
if (searchText == null)
throw new ArgumentNullException("searchText");
return AllIndexesOfCore(str, searchText);
}
private static IEnumerable<int> AllIndexesOfCore(string str, string searchText)
{
for (int index = 0; ; index += searchText.Length)
{
index = str.IndexOf(searchText, index);
if (index == -1)
break;
yield return index;
}
}
This works because the first method will behave just like you expect (immediate execution), and will return the state machine implemented by the second method.
Note that you should also check the str
parameter for null
, because extensions methods can be called on null
values, as they're just syntactic sugar.
If you're curious about what the compiler does to your code, here's your method, decompiled with dotPeek using the Show Compiler-generated Code option.
public static IEnumerable<int> AllIndexesOf(this string str, string searchText)
{
Test.<AllIndexesOf>d__0 allIndexesOfD0 = new Test.<AllIndexesOf>d__0(-2);
allIndexesOfD0.<>3__str = str;
allIndexesOfD0.<>3__searchText = searchText;
return (IEnumerable<int>) allIndexesOfD0;
}
[CompilerGenerated]
private sealed class <AllIndexesOf>d__0 : IEnumerable<int>, IEnumerable, IEnumerator<int>, IEnumerator, IDisposable
{
private int <>2__current;
private int <>1__state;
private int <>l__initialThreadId;
public string str;
public string <>3__str;
public string searchText;
public string <>3__searchText;
public int <index>5__1;
int IEnumerator<int>.Current
{
[DebuggerHidden] get
{
return this.<>2__current;
}
}
object IEnumerator.Current
{
[DebuggerHidden] get
{
return (object) this.<>2__current;
}
}
[DebuggerHidden]
public <AllIndexesOf>d__0(int <>1__state)
{
base..ctor();
this.<>1__state = param0;
this.<>l__initialThreadId = Environment.CurrentManagedThreadId;
}
[DebuggerHidden]
IEnumerator<int> IEnumerable<int>.GetEnumerator()
{
Test.<AllIndexesOf>d__0 allIndexesOfD0;
if (Environment.CurrentManagedThreadId == this.<>l__initialThreadId && this.<>1__state == -2)
{
this.<>1__state = 0;
allIndexesOfD0 = this;
}
else
allIndexesOfD0 = new Test.<AllIndexesOf>d__0(0);
allIndexesOfD0.str = this.<>3__str;
allIndexesOfD0.searchText = this.<>3__searchText;
return (IEnumerator<int>) allIndexesOfD0;
}
[DebuggerHidden]
IEnumerator IEnumerable.GetEnumerator()
{
return (IEnumerator) this.System.Collections.Generic.IEnumerable<System.Int32>.GetEnumerator();
}
bool IEnumerator.MoveNext()
{
switch (this.<>1__state)
{
case 0:
this.<>1__state = -1;
if (this.searchText == null)
throw new ArgumentNullException("searchText");
this.<index>5__1 = 0;
break;
case 1:
this.<>1__state = -1;
this.<index>5__1 += this.searchText.Length;
break;
default:
return false;
}
this.<index>5__1 = this.str.IndexOf(this.searchText, this.<index>5__1);
if (this.<index>5__1 != -1)
{
this.<>2__current = this.<index>5__1;
this.<>1__state = 1;
return true;
}
goto default;
}
[DebuggerHidden]
void IEnumerator.Reset()
{
throw new NotSupportedException();
}
void IDisposable.Dispose()
{
}
}
This is invalid C# code, because the compiler is allowed to do things the language doesn't allow, but which are legal in IL - for instance naming the variables in a way you couldn't to avoid name collisions.
But as you can see, the AllIndexesOf
only constructs and returns an object, whose constructor only initializes some state. GetEnumerator
only copies the object. The real work is done when you start enumerating (by calling the MoveNext
method).
You have an iterator block. None of the code in that method is ever run outside of calls to MoveNext
on the returned iterator. Calling the method does noting but create the state machine, and that won't ever fail (outside of extremes such as out of memory errors, stack overflows, or thread abort exceptions).
When you actually attempt to iterate the sequence you'll get the exceptions.
This is why the LINQ methods actually need two methods to have the error handling semantics they desire. They have a private method that is an iterator block, and then a non-iterator block method that does nothing but do the argument validation (so that it can be done eagerly, rather than it being deferred) while still deferring all other functionality.
So this is the general pattern:
public static IEnumerable<T> Foo<T>(
this IEnumerable<T> souce, Func<T, bool> anotherArgument)
{
//note, not an iterator block
if(anotherArgument == null)
{
//TODO make a fuss
}
return FooImpl(source, anotherArgument);
}
private static IEnumerable<T> FooImpl<T>(
IEnumerable<T> souce, Func<T, bool> anotherArgument)
{
//TODO actual implementation as an iterator block
yield break;
}