How is LINQ compiled into the CIL?

Question

For example:

var query = from c in db.Cars select c;
foreach(Car aCar in query)
{
     Console.WriteLine(aCar.Name);
}

How would this translate once it is compiled? What happens behind the scenes?

Answer 1

It is compiled in the following way:

1. First, the LINQ query expression is transformed into method calls:

   public static void Main()
   {
       var query = db.Cars.Select<Car, Car>(c => c);
       foreach (Car aCar in query)
       {
            Console.WriteLine(aCar.Name);
       }
   }
   

2. If db.Cars is of type IEnumerable<Car> (which it is for LINQ-to-Objects), then the lambda expression is turned into a separate method:

   private Car lambda0(Car c)
   {
       return c;
   }
   private Func<Car, Car> CachedAnonymousMethodDelegate1;
   public static void Main()
   {
       if (CachedAnonymousMethodDelegate1 == null)
           CachedAnonymousMethodDelegate1 = new Func<Car, Car>(lambda0);
       var query = db.Cars.Select<Car, Car>(CachedAnonymousMethodDelegate1);
       foreach // ...
   }
   

   In reality the method is not called lambda0 but something like <Main>b__0 (where Main is the name of the containing method). Similarly, the cached delegate is actually called CS$<>9__CachedAnonymousMethodDelegate1.

   If you are using LINQ-to-SQL, then db.Cars will be of type IQueryable<Car> and this step is very different. It would instead turn the lambda expression into an expression tree:

   public static void Main()
   {
       var parameter = Expression.Parameter(typeof(Car), "c");
       var lambda = Expression.Lambda<Func<Car, Car>>(parameter, new ParameterExpression[] { parameter }));
       var query = db.Cars.Select<Car, Car>(lambda);
       foreach // ...
   }
   

3. The foreach loop is transformed into a try/finally block (this is the same for both):

   IEnumerator<Car> enumerator = null;
   try
   {
       enumerator = query.GetEnumerator();
       Car aCar;
       while (enumerator.MoveNext())
       {
           aCar = enumerator.Current;
           Console.WriteLine(aCar.Name);
       }
   }
   finally
   {
       if (enumerator != null)
           ((IDisposable)enumerator).Dispose();
   }
   

4. Finally, this is compiled into IL the expected way. The following is for IEnumerable<Car>:

   // Put db.Cars on the stack
   L_0016: ldloc.0 
   L_0017: callvirt instance !0 DatabaseContext::get_Cars()
   
   
   // “if” starts here
   L_001c: ldsfld Func<Car, Car> Program::CachedAnonymousMethodDelegate1
   L_0021: brtrue.s L_0034
   L_0023: ldnull 
   L_0024: ldftn Car Program::lambda0(Car)
   L_002a: newobj instance void Func<Car, Car>::.ctor(object, native int)
   L_002f: stsfld Func<Car, Car> Program::CachedAnonymousMethodDelegate1
   
   
   // Put the delegate for “c => c” on the stack
   L_0034: ldsfld Func<Car, Car> Program::CachedAnonymousMethodDelegate1
   
   
   // Call to Enumerable.Select()
   L_0039: call IEnumerable<!!1> Enumerable::Select<Car, Car>(IEnumerable<!!0>, Func<!!0, !!1>)
   L_003e: stloc.1
   
   
   // “try” block starts here
   L_003f: ldloc.1 
   L_0040: callvirt instance IEnumerator<!0> IEnumerable<Car>::GetEnumerator()
   L_0045: stloc.3
   
   
   // “while” inside try block starts here
   L_0046: br.s L_005a
   L_0048: ldloc.3   // body of while starts here
   L_0049: callvirt instance !0 IEnumerator<Car>::get_Current()
   L_004e: stloc.2 
   L_004f: ldloc.2 
   L_0050: ldfld string Car::Name
   L_0055: call void Console::WriteLine(string)
   L_005a: ldloc.3   // while condition starts here
   L_005b: callvirt instance bool IEnumerator::MoveNext()
   L_0060: brtrue.s L_0048  // end of while
   L_0062: leave.s L_006e   // end of try
   
   
   // “finally” block starts here
   L_0064: ldloc.3 
   L_0065: brfalse.s L_006d
   L_0067: ldloc.3 
   L_0068: callvirt instance void IDisposable::Dispose()
   L_006d: endfinally 
   

   The compiled code for the IQueryable<Car> version is also as expected. Here is the important part that is different from the above (the local variables will have different offsets and names now, but let’s disregard that):

   // typeof(Car)
   L_0021: ldtoken Car
   L_0026: call Type Type::GetTypeFromHandle(RuntimeTypeHandle)
   
   
   // Expression.Parameter(typeof(Car), "c")
   L_002b: ldstr "c"
   L_0030: call ParameterExpression Expression::Parameter(Type, string)
   L_0035: stloc.3 
   
   
   // Expression.Lambda(...)
   L_0036: ldloc.3 
   L_0037: ldc.i4.1           // var paramArray = new ParameterExpression[1]
   L_0038: newarr ParameterExpression
   L_003d: stloc.s paramArray
   L_003f: ldloc.s paramArray
   L_0041: ldc.i4.0                    // paramArray[0] = parameter;
   L_0042: ldloc.3 
   L_0043: stelem.ref 
   L_0044: ldloc.s paramArray
   L_0046: call Expression<!!0> Expression::Lambda<Func<Car, Car>>(Expression, ParameterExpression[])
   
   
   // var query = Queryable.Select(...);
   L_004b: call IQueryable<!!1> Queryable::Select<Car, Car>(IQueryable<!!0>, Expression<Func<!!0, !!1>>)
   L_0050: stloc.1