Use synchronization domains. Avoid manual synchronization, because that often leads to deadlocks and race conditions.
Never call outside your synchronization domain.
Manage asynchronous call completion on a callback method. Do not wait, poll, or block for completion.
Always name your threads:
Thread currentThread = Thread.CurrentThread;
string threadName = "Main UI Thread";
currentThread.Name = threadName;
The name is traced in the debugger Threads window, making debug sessions more productive.
Do not call Suspend( ) or Resume( ) on a thread.
Do not call Thread.Sleep( ), except in the following conditions:
Thread.Sleep(0) is an acceptable optimization technique to force a context switch.
Thread.Sleep( ) is acceptable in testing or simulation code.
Do not call THRead.SpinWait( ).
Do not call Thread.Abort( ) to terminate threads. Use a synchronization object instead to signal the thread to terminate.
Avoid explicitly setting the thread priority to control execution. You can set the thread priority based on task semantics (such as ThreadPriority.BelowNormal for a screensaver).
Do not read the value of the ThreadState property. Use Thread.IsAlive( ) to determine whether the thread is dead or alive.
Do not rely on setting the thread type to background thread for application shutdown. Use a watchdog or other monitoring entity to deterministically kill threads.
Do not use the thread local storage unless thread affinity is guaranteed.
Do not call Thread.MemoryBarrier( ).
Never call Thread.Join( ) without checking that you are not joining your own thread:
void WaitForThreadToDie(Thread thread)
{
Debug.Assert(Thread.CurrentThread.ManagedThreadId != thread.ManagedThreadId);
thread.Join( );
}
Always use the lock( ) statement rather than explicit Monitor manipulation.
Always encapsulate the lock( ) statement inside the object it protects:
public class MyClass
{
public void DoSomething( )
{
lock(this)
{...}
}
}
You can use synchronized methods instead of writing the lock( ) statement yourself.
Avoid fragmented locking.
Avoid using a Monitor to wait or pulse objects. Use manual or auto-reset events instead.
Do not use volatile variables. Lock your object or fields instead to guarantee deterministic and thread-safe access. Do not use THRead.VolatileRead( ), Thread.VolatileWrite( ), or the volatile modifier.
Avoid increasing the maximum number of threads in the thread pool.
Never stack lock( ) statements, because that does not provide atomic locking:
MyClass obj1 = new MyClass( );
MyClass obj2 = new MyClass( );
MyClass obj3 = new MyClass( );
//Do not stack lock statements
lock(obj1)
lock(obj2)
lock(obj3)
{
obj1.DoSomething( );
obj2.DoSomething( );
obj3.DoSomething( );
}
Use WaitHandle.WaitAll( ) instead.
Prefer the binary formatter.
Mark serialization event-handling methods as private.
Use the generic IGenericFormatter interface.
Always mark non-sealed classes as serializable.
When implementing IDeserializationCallback on a non-sealed class, make sure to do so in a way that allows subclasses to call the base class implementation of OnDeserialization( ).
Always mark unserializable member variables as non-serializable.
Always mark delegates on a serialized class as non-serializable fields:
[Serializable]
public class MyClass
{
[field:NonSerialized]
public event EventHandler MyEvent;
}
Prefer administrative configuration to programmatic configuration.
Always implement IDisposable on single-call objects.
Always prefer a TCP channel and a binary format when using remoting, unless a firewall is present.
Always provide a null lease for a singleton object:
public class MySingleton : MarshalByRefObject
{
public override object InitializeLifetimeService( )
{
return null;
}
}
Always provide a sponsor for a client-activated object. The sponsor should return the initial lease time.
Always unregister the sponsor on client application shutdown.
Always put remote objects in class libraries.
Avoid using SoapSuds.exe.
Avoid hosting in IIS.
Avoid using uni-directional channels.
Always load a remoting configuration file in Main( ), even if the file is empty and the application does not use remoting:
static void Main( )
{
RemotingConfigurationEx.Configure( );
/* Rest of Main( ) */
}
Avoid using Activator.GetObject( ) and Activator.CreateInstance( ) for remote object activation. Use new instead.
Always register port 0 on the client side, to allow callbacks.
Always elevate type filtering to Full on both client and host, to allow callbacks.
Always demand your own strong name on assemblies and components that are private to the application, but are public (so that only you can use them):
public class PublicKeys
{
public const string MyCompany = "1234567894800000940000000602000000240000"+
"52534131000400000100010007D1FA57C4AED9F0"+
"A32E84AA0FAEFD0DE9E8FD6AEC8F87FB03766C83"+
"4C99921EB23BE79AD9D5DCC1DD9AD23613210290"+
"0B723CF980957FC4E177108FC607774F29E8320E"+
"92EA05ECE4E821C0A5EFE8F1645C4C0C93C1AB99"+
"285D622CAA652C1DFAD63D745D6F2DE5F17E5EAF"+
"0FC4963D261C8A12436518206DC093344D5AD293";
}
[StrongNameIdentityPermission(SecurityAction.LinkDemand,
PublicKey = PublicKeys.MyCompany)]
public class MyClass
{...}
Apply encryption and security protection on application configuration files.
When importing an interop method, assert unmanaged code permission and demand appropriate permission instead:
[DllImport("user32",EntryPoint="MessageBoxA")]
private static extern int Show(IntPtr handle,string text,string caption,
int msgType);
[SecurityPermission(SecurityAction.Assert,UnmanagedCode = true)]
[UIPermission(SecurityAction.Demand,
Window = UIPermissionWindow.SafeTopLevelWindows)]
public static void Show(string text,string caption)
{
Show(IntPtr.Zero,text,caption,0);
}
Do not suppress unmanaged code access via the SuppressUnmanagedCodeSecurity attribute.
Do not use the /unsafe switch of TlbImp.exe. Wrap the RCW in managed code so that you can assert and demand permissions declaratively on the wrapper.
On server machines, deploy a code access security policy that grants only Microsoft, ECMA, and self (identified by a strong name) full trust. Code originating from anywhere else is implicitly granted nothing.
On client machines, deploy a security policy that grants client application only the permissions to execute, to call back the server, and to potentially display user interface. When not using ClickOnce, client application should be identified by a strong name in the code groups.
To counter a luring attack, always refuse at the assembly level all permissions not required to perform the task at hand:
[assembly:UIPermission(SecurityAction.RequestRefuse,
Window=UIPermissionWindow.AllWindows)]
Always set the principal policy in every Main( ) method to Windows:
public class MyClass
{
static void Main( )
{
AppDomain currentDomain = AppDomain.CurrentDomain;
currentDomain.SetPrincipalPolicy(PrincipalPolicy.WindowsPrincipal);
}
//other methods
}
Never assert a permission without demanding a different permission in its place.