Should a return statement be inside or outside a lock?

Question

I just realized that in some place in my code I have the return statement inside the lock and sometime outside. Which one is the best?

1)

void exampl         


        

Answer 1

Essentially, which-ever makes the code simpler. Single point of exit is a nice ideal, but I wouldn't bend the code out of shape just to achieve it... And if the alternative is declaring a local variable (outside the lock), initializing it (inside the lock) and then returning it (outside the lock), then I'd say that a simple "return foo" inside the lock is a lot simpler.

To show the difference in IL, lets code:

static class Program
{
    static void Main() { }

    static readonly object sync = new object();

    static int GetValue() { return 5; }

    static int ReturnInside()
    {
        lock (sync)
        {
            return GetValue();
        }
    }

    static int ReturnOutside()
    {
        int val;
        lock (sync)
        {
            val = GetValue();
        }
        return val;
    }
}

(note that I'd happily argue that ReturnInside is a simpler/cleaner bit of C#)

And look at the IL (release mode etc):

.method private hidebysig static int32 ReturnInside() cil managed
{
    .maxstack 2
    .locals init (
        [0] int32 CS$1$0000,
        [1] object CS$2$0001)
    L_0000: ldsfld object Program::sync
    L_0005: dup 
    L_0006: stloc.1 
    L_0007: call void [mscorlib]System.Threading.Monitor::Enter(object)
    L_000c: call int32 Program::GetValue()
    L_0011: stloc.0 
    L_0012: leave.s L_001b
    L_0014: ldloc.1 
    L_0015: call void [mscorlib]System.Threading.Monitor::Exit(object)
    L_001a: endfinally 
    L_001b: ldloc.0 
    L_001c: ret 
    .try L_000c to L_0014 finally handler L_0014 to L_001b
} 

method private hidebysig static int32 ReturnOutside() cil managed
{
    .maxstack 2
    .locals init (
        [0] int32 val,
        [1] object CS$2$0000)
    L_0000: ldsfld object Program::sync
    L_0005: dup 
    L_0006: stloc.1 
    L_0007: call void [mscorlib]System.Threading.Monitor::Enter(object)
    L_000c: call int32 Program::GetValue()
    L_0011: stloc.0 
    L_0012: leave.s L_001b
    L_0014: ldloc.1 
    L_0015: call void [mscorlib]System.Threading.Monitor::Exit(object)
    L_001a: endfinally 
    L_001b: ldloc.0 
    L_001c: ret 
    .try L_000c to L_0014 finally handler L_0014 to L_001b
}

So at the IL level they are [give or take some names] identical (I learnt something ;-p). As such, the only sensible comparison is the (highly subjective) law of local coding style... I prefer ReturnInside for simplicity, but I wouldn't get excited about either.

Answer 2

I would definitely put the return inside the lock. Otherwise you risk another thread entering the lock and modifying your variable before the return statement, therefore making the original caller receive a different value than expected.

Answer 3

To make it easier for fellow developers to read the code I would suggest the first alternative.

Answer 4

If think the lock outside looks better, but be careful if you end up changing the code to:

return f(...)

If f() needs to be called with the lock held then it obviously needs to be inside the lock, as such keeping returns inside the lock for consistency makes sense.

Answer 5

lock() return <expression> statements always:

1) enter lock

2) makes local (thread-safe) store for the value of the specified type,

3) fills the store with the value returned by <expression>,

4) exit lock

5) return the store.

It means that value, returned from lock statement, always "cooked" before return.

Don't worry about lock() return, do not listen to anyone here ))

Answer 6

It depends,

I am going to go against the grain here. I generally would return inside of the lock.

Usually the variable mydata is a local variable. I am fond of declaring local variables while I initialize them. I rarely have the data to initialize my return value outside of my lock.

So your comparison is actually flawed. While ideally the difference between the two options would be as you had written, which seems to give the nod to case 1, in practice its a little uglier.

void example() { 
    int myData;
    lock (foo) { 
        myData = ...;
    }
    return myData
}

vs.

void example() { 
    lock (foo) {
        return ...;
    }
}

I find case 2 to be considerably easier to read and harder to screw up, especially for short snippets.