Static constructors, part four

We’ll finish up this series on static constructors by finally getting to the question which motivated me to write the series in the first place: should you use a static constructor like this?

public class Sensitive
{
  static Sensitive()
  {
    VerifyUserHasPermissionToUseThisClass();
  }  
  public static void Dangerous()
  {
    DoSomethingDangerous();
  }
  ...

The intention here is clear. The static constructor is guaranteed to run exactly once, and before any static or instance method. Therefore it will do the authorization check before any dangerous operation in any method. If the user does not have permission to use the class then the class will not even load. If they do, then the expense of the security check is only incurred once per execution, no matter how many methods are called.

If you’ve read the rest of this series, or anything I’ve written on security before, you know what I’m going to say: I strongly recommend that you do not use static constructors “off label” in this manner.

First off, as we’ve seen so far in this series, static constructors are a dangerous place to run fancy code. If they end up delegating any work to other threads then deadlocks can easily result. If they take a long time and are accessed from multiple threads, contention can result. If an exception is thrown out of the static constructor then you have very little ability to recover from the exception and the type will be forever useless in this appdomain.

Second, the security semantics here are deeply troubling. If this code is running on the user’s machine then this appears to be a case of the developer not trusting the user. But the .NET security system was designed with the principle that the user is the source of trust decisions. It is the user who must trust the developer, not the other way around! If the user is hostile towards the developer then nothing is stopping them from decompiling the code to IL, removing the static constructor, recompiling, and running the dangerous method without the security check. [1. The resulting program will of course not be strong-named or code-signed by the developer anymore, but who cares?]

Moreover, the pattern here assumes that security checks can be performed once and the result is then cached for the lifetime of the appdomain. What if the initial security check fails, but the program was going to impersonate a more trusted user? It might be difficult to ensure that the static constructor does not run until after the impersonation. What if different threads are associated with different users? Now we have a race to see which user’s context is used for the security check. What if a user’s permissions are revoked while the program is running? The check might be performed while permission is granted, and then the dangerous code runs after it has been revoked.

In short: Static constructors should be used to quickly initialize important static data, and that’s pretty much it. The only time that I would use the mechanisms of a static constructor to enforce an invariant would be for a very simple invariant like ensuring that a singleton is lazily initialized, as Jon describes. Complex policy mechanisms like security checks should probably use some other mechanism.


Next time on FAIC: I’m going to join the throngs of tech bloggers who have tried to explain what a monad is.

About these ads

Static constructors, part three

Earlier in this series I recommended that you brush up on how instance constructors work; if you did, then you’ll recall that instance field initializers are essentially moved into the beginning of an instance constructor at a point before the call to the base class constructor. You might think that static field initializers work the same: a static field initializer is silently inserted at the beginning of the static constructor. And that’s true. Mostly.

In the case where there is already a static constructor, even an empty one, that’s what happens: the field initializers become the prologue to the body of the static constructor, and the usual rules for static constructors then apply. (That is, the static constructor is invoked immediately before the first static member access or instance constructor access.) But suppose there is no user-supplied static constructor. Then what happens?

The C# compiler is not bound by the rules of static constructors in this case, and in fact, does not treat your program as though there was an empty static constructor that has static field initializers in it. Rather, it tells the runtime that the runtime may choose when to run static field initializers, entirely at its discretion, just so long as all the fields are initialized [1. The runtime is still responsible for ensuring that fields are initialized in the right order, because one static field might be initialized based on the contents of another. This is a bad idea, but it is legal, so the compiler has to honour that.] before they are used. In this scenario the runtime is permitted to run static field initializers as late as possible; it could wait until a static field is actually accessed, rather than waiting for any static member or instance constructor to be accessed. The runtime is also permitted to run static field initializers as early as possible; it could decide to run all the field initializers at once at the beginning of the program, even if the class in question was never used. It is up to the runtime implementation to decide.

Pre .NET 4 implementations of the runtime make an interesting choice; they run static field initializers of classes that have no static constructors when the first method that refers to the class is jitted. If we have:

class Alpha 
{
  static int x = 123;
  static void M() { }
}
class Bravo
{
  static int y = 456;
  static void N() { }
}
class Charlie
{
  static void Q(bool b)
  {
    if (b) Alpha.M(); else Bravo.N();
  }
  static void Main() 
  { 
    Q(true); 
  }
}

Then when Q is jitted, the field initializers for both x and y will be executed. If Bravo had a static constructor then it would not be initialized, because Bravo’s static constructor is only triggered by the actual execution of the call.

The runtime makes this rather odd-seeming choice as an optimization; this way it doesn’t have to generate code on every static member access that checks to see if the field initializers have run yet! It knows that if you get as far as accessing a static member, then the method that does so must have been jitted, and therefore the field initializers have been run already.

I originally believed this to be the case in .NET 4 as well, but Jon informs me that current versions of the runtime are even lazier than that. See Jon’s comment below for details. (Thanks Jon!)

I find this to be one of the strangest C#/CLR features and for many years I did not understand it at all well — and, since Jon has corrected my understanding of it, apparently I still don’t! Every time I encountered a question about it, I just referred the questioner to Jon’s excellent page on the subject. For more information and discussion on this rather odd feature, check it out.


Next time on FAIC: We’ll finish up this series by looking at an abuse of static constructors.

Static constructors, part two

Previously on FAIC I gave a quick overview of the basic operation of a static constructor. Today, three unusual corner cases.

The first odd case I want to talk about involves static methods. Take a look at the sample program from last time. Now suppose we edited the Main method to say:

static void Main() 
{
  D.M();
}

First off, is that even legal? Sure! Inheritance means that all inheritable members of B are also members of D. M is an inheritable member of B, so it is a member of D, right?

Unfortunately, this corner case is the one that exposes the leaky abstraction. The compiler generates code as though you had said B.M();, and therefore D’s static constructor is not called even though “a member of D” has been invoked. This actually makes a fair amount of sense. The method B.M is going to be called, and there’s no reason to go to all the work of running D’s static constructor when B.M probably does not depend on any work done by D’s constructor. And it would seem strange if calling the same method by two different syntaxes would result in different static constructor invocations.

Now let’s consider a second case involving static method invocation. Suppose now we edited Main to say:

static void Main() 
{
  D.N();
}

Clearly D’s static constructor must be invoked. What about B? Is its static constructor invoked? No! A static constructor is triggered by a usage of a static member, or by the creation of an instance. Invoking D.N does not use any static member of B and it does not create an instance of B, so B’s static constructor is not invoked. People sometimes expect that static constructors of base classes will always be invoked before static constructors of derived classes, but that’s not the case.

Our third odd case is: what happens when a static constructor throws an exception?

Absolutely nothing good! First off, of course if the exception goes unhandled then all bets are off. The runtime is permitted to do anything it likes if there is an unhandled exception, including such options as starting up a debugger, terminating the appdomain immediately, terminating the application after running finally blocks, and so on. And an exception in a static constructor can easily go unhandled; trying to wrap every possible first usage of a type with a try-catch block is onerous.

And even if by some miracle the exception gets handled the first time, odds are very good that your program is now in such a damaged state that it is going to go down in flames soon. Remember, I said that a static constructor runs once, and by that I meant once; if it throws, you don’t get a second chance. Instead, when a static constructor terminates abnormally, the runtime marks the type as unusable, and every attempt by your program to use that type results in another exception.

An interesting fact about static constructors that throw exceptions is that when the runtime detects that a static constructor has terminated abnormally, it wraps the exception in its own exception and throws that instead. Check out this StackOverflow answer, where Jon demonstrates this in action.


Next time on FAIC: I’ll defer to Jon again when I discuss how the runtime is permitted to optimize some static constructors.

Static constructors, part one

Previously on FAIC we saw how easy it was to deadlock a program by trying to do something interesting in a static constructor.[1. Static constructors are also called “class constructors”. Since the actual method generated has the name .cctor they are often also called “cctors”. Since “static constructor” is the jargon used in the C# specification, that’s what I’ll stick to.] Static constructors and destructors[2. Astonishingly, I’ve never blogged about how difficult it is to write a correct destructor, though it has come up on StackOverflow. That’s a good topic for a future fabulous adventure.] are the two really weird kinds of methods, and you should do as little as possible in them.

Before I expound further on that topic though, a look at how static constructors work is in order. And before I do that, it’s probably a good idea that you get a refresher on how instance constructors work. My article “Why do initializers run in the opposite order of constructors?” provides a detailed look at constructor semantics, so maybe check that out if you have a few minutes. Part one is here and part two is here.

OK, now that you know how instance constructors work, let’s dig into static constructors. The idea is pretty simple: a static constructor is triggered to run immediately before the first static method on its class is called, or immediately before the first instance of its class is created. As we saw previously, the runtime tracks when a static constructor is “in flight” and uses that mechanism to ensure that each static constructor is invoked no more than once.

Now that you know all of that, you can predict the output of this simple program:

using System;
class B
{
  static B() { Console.WriteLine("B cctor"); }
  public B() { Console.WriteLine("B ctor"); }
  public static void M() { Console.WriteLine("B.M"); }
}
class D : B
{
  static D() { Console.WriteLine("D cctor"); }
  public D() { Console.WriteLine("D ctor"); }
  public static void N() { Console.WriteLine("D.N"); }
}
class P 
{
  static void Main()
  {
    System.Console.WriteLine("Main");
    new D();
  }  
}

We know that B’s instance constructor must be invoked before D’s instance constructor, and we know that D’s static constructor must be invoked before D’s instance constructor. The only interesting question here is “when will B’s static constructor be invoked?” An instance of D is also an instance of B, so B’s static constructor has to be invoked at some point.

As you know from reading my article on instance constructors, what actually happens is that the compiler generates D’s instance constructor so that the first thing it does is call B’s instance constructor; that’s how we get the appearance that B’s instance constructor runs first. Thus, the actual order of events here can be best conceptualized like this:

  • Main starts. It prints out its message and then tries to invoke D’s instance constructor on a new instance of D.
  • The runtime detects that D’s instance constructor is about to be invoked, so it invokes D’s static constructor.
  • D’s instance constructor invokes B’s instance constructor. The runtime detects that, so it invokes B’s static constructor.
  • B’s instance constructor runs and returns control to D’s instance constructor, which finishes normally.

Pretty straightforward. Let’s mix it up a little.


Next time on FAIC: A brief digression for fun on a Friday. Then next week we’ll resume this series and take a look at a few less straightforward cases.

The no-lock deadlock

People sometimes ask me if there is a cheap-and-easy way to guarantee thread safety. For example, “if my method only reads and writes local variables and parameters, can I guarantee that my method is threadsafe?” Questions like that are dangerous because they are predicated on an incorrect assumption: that if every method of a program is “threadsafe”, whatever that means, then the entire program is “threadsafe”. I might not be entirely clear on what “threadsafe” means, but I do know one thing about it: thread safety is a property of entire programs, not of individual methods.

To illustrate why these sorts of questions are non-starters, today I present to you the world’s simplest deadlocking C# program:[1. Thanks to my erstwhile colleague Neal Gafter for this example.]

class C
{
  static C() 
  {
    // Let's run the initialization on another thread!
    var thread = new System.Threading.Thread(Initialize);
    thread.Start();
    thread.Join();
  }
  static void Initialize() { }
  static void Main() { }
}

At first glance clearly ever method of this incredibly simple program is “threadsafe”. There is only a single variable anywhere in the program; it is local, is written once, is written before it is read, is read from the same thread it was written on, and is guaranteed to be atomic. There are apparently no locks anywhere in the program, and so there are no lock ordering inversions. Two of the three methods are empty. And yet this program deadlocks with 100% certainty; the program “globally” is clearly not threadsafe, despite all those nice “local” properties. You can build a hollow house out of solid bricks; so too you can build a deadlocking program out of threadsafe methods.

The reason why this deadlocks is a consequence of the rules for static constructors in C#; the important rule is that a static constructor runs exactly zero or one times, and runs before a static method call or instance creation in its type. Therefore the static constructor of C must run to completion before Main starts. The CLR notes that C‘s static constructor is “in flight” on the main thread and calls it. The static constructor then starts up a new thread. When that thread starts, the CLR sees that a static method is about to be called on a type whose static constructor is “in flight” another thread. It immediately blocks the new thread so that the Initialize method will not start until the main thread finishes running the class constructor. The main thread blocks itself waiting for the new thread to complete, and now we have two threads each waiting for the other to complete.


Next time on FAIC: We’re opening up the new Coverity office in Seattle! After which, we’ll take a closer look at the uses and abuses of the static constructor.