icelava
icelava

Reputation: 9857

Can object constructor return a null?

We have taken over some .NET 1.1 Windows Service code that spawns threads to read messages off a queue (SeeBeyond eGate JMS queue, but that is not important) and in turn spawns threads to process the message in the target application service. We are continually encountering logic and design decisions that is puzzling us to no end. Here is one example, where the message (lsMessage) has been retrieved from the queue and ready for processing

if(lsMessage != null)
{
    // Initialize a new thread class instance, pass in message
    WorkerThread worker = new WorkerThread(lsMessage);

Process:
    // Start a new thread to process the message
    Thread targetWorker = new Thread(new ThreadStart(worker.ProcessMessage));
    if(targetWorker != null)
    {
        targetWorker.Priority = ThreadPriority.Highest;
        targetWorker.Name = "Worker " + queueKey.ToString();
        targetWorker.Start();

        // wait for worker thread to join back in specified period
        bool isFinished = targetWorker.Join(SYNC_THREAD_TIMEOUT);

        string message = worker.replyMsg;

        if ( !isFinished )  // BF is timeout
        {
            targetWorker.Abort();

            // [obscure developer name] 25/10/2004: calling Join() to wait for thread to terminate.
            // for EAI listener threads problem, ensure no new thread is started 
            // before the old one ends
            targetWorker.Join();

            // prepare reply message
            string errorMsg = string.Format("EAIMsg {0}: BF is timeout. Send sync message back to caller.", worker.messageKey);
            log.Debug(errorMsg);

            message = worker.GenErrorCode(message, errorMsg);
        }

        // Commit message
        MQ.ReceiverCommit(queueKey, worker.messageKey, false);

        // Send back the response to the caller
        MQ.RespondSend(queueKey, message); 
    } 
    else 
    {
        log.Debug(string.Format("Fail to start worker thread to process sync message. Thread returned is null. Sleep for {0} milliseconds.", LIMIT_RESOURCE_SLEEP));
        Thread.Sleep(LIMIT_RESOURCE_SLEEP);
        goto Process;
    }
}

Please ignore the use of label and goto for the moment; that is not the question. Our bewilderment is the check whether the Thread object is null right after instantiation. The else statement below seems to suggest the previous developers have encountered situations like this before. Of course, the original developers are long gone. So we would like to know, can the CLR really instantiate an object after the call to the constructor and return a null? We have no knowledge of such a possibility.

Upvotes: 40

Views: 36176

Answers (5)

Stevens Miller
Stevens Miller

Reputation: 1480

As core mentions, operator overloading can make it appear that a constructor returned null, when that's not what really happened. The authors of the article core found say they haven't seen it used, but it actually is used in a very popular product: Unity.

This will compile and log a message at run time:

using UnityEngine;

public class Test:MonoBehaviour
{
    void Start()
    {
        AudioSource as = new AudioSource();

        if (as == null)
        {
            Debug.Log("Looks null to me.");
        }
    }
}

Now, the fault here is mine, because one should not call the AudioSource constructor directly. But one should know that the == operator is overloaded at the root of the inheritance tree for the objects Unity can reference. Here's what the Unity manual says about UnityEngine.Object's == operator:

Be careful when comparing with null.

e.g.

GameObject go = new GameObject();
Debug.Log (go == null); // false

Object obj = new Object();
Debug.Log (obj == null); // true

Instatiating a GameObject adds it to the scene so it's completely initialized (!destroyed). Instantiating a simple UnityEngine.Object has no such semantics, so the(sic) it stays in the 'destroyed' state which compares true to null.

While instantiating a GameObject initializes it, instantiating an AudioSource object doesn't, so the comparison with null returns true.

This unusual idiom is made even more stealthy by virtue of the fact that attempts to reference properties of the uninitialized AudioSource object will throw null-reference exceptions, which I initially misinterpreted as meaning the object reference was null, not the property.

Others have answered the OP's question, but I wanted to add this answer because the OP's code might actually make sense if the Thread class therein isn't the one we would expect it to be, just as Object (and its descendants) isn't quite what you might expect it to be in a Unity script (that is, it is actually UnityEngine.Object, rather than System.Object, which gets you the overloaded == operator that so confused me).

Upvotes: 3

Gorpik
Gorpik

Reputation: 11028

In my opinion, what the else statement suggests is that the previous developers didn't know their C#. A constructor always returns a constructed object or throws an exception.

In the very old times, C++ constructors could return null, so maybe the problem comes from that. This is no longer true in C++ either, at least for the default new operator.

Upvotes: 45

Marc Gravell
Marc Gravell

Reputation: 1062510

Edit: for clarification there is an insane edge case where you can get null from a class constructor, but frankly I don't think any real code should ever expect to deal with this level of crazy: What's the strangest corner case you've seen in C# or .NET? . To all normal intents : it won't happen.


No, you can't get null from a class constructor (Thread is a class). The only case I know of where a constructor can (seem to) return null is Nullable<T> - i.e.

object foo = new int?(); // this is null

This is a slightly bigger problem with generics:

static void Oops<T>() where T : new() {
    T t = new T();
    if (t == null) throw new InvalidOperationException();
}

static void Main() {
    Oops<int?>();
}

(of course, there are ways of checking/handling that scenario, such as : class)

Other than that, a constructor will always either return an object (or initialize a struct), or throw an exception.

Upvotes: 32

core
core

Reputation: 33059

You can make it appear like an object ctor returns null:

http://seattlesoftware.wordpress.com/2008/03/05/returning-null-from-a-class-constructor/

Search for "Another pattern which I haven’t seen used allows an invalid object to emulate a null reference" and read from there.

Upvotes: 2

Sesh
Sesh

Reputation: 6182

NO! that null check is redundant. Lot of C++ devs who moved to C# have this habit of a null check and I guess it is the same here.

The only thing is you should check the documentation to see if the constructor can throw any exception. In your case refer to http://msdn.microsoft.com/en-us/library/xx3ezzs2.aspx and as mentioned the constructor will always return a valid obj.

Upvotes: 2

Related Questions