See my article on this very issue.
Basically, beforefieldinit
means “the type can be initialized at any point before any static fields are referenced.” In theory that means it can be very lazily initialized – if you call a static method which doesn’t touch any fields, the JIT doesn’t need to initialize the type.
In practice it means that the class is initialized earlier than it would be otherwise – it’s okay for it to be initialized at the start of the first method which might use it. Compare this with types which don’t have beforefieldinit
applied to them, where the type initialization has to occur immediately before the first actual use.
So, suppose we have:
public static void DoSomething(bool which)
{
if (which)
{
FirstType.Foo();
}
else
{
SecondType.Bar();
}
}
If both types have beforefieldinit
applied to them (which in C# they do by default unless the type has a static constructor) then they’ll both be initialized at the start of the DoSomething
method (usually – it’s not guaranteed). If they don’t have beforefieldinit
then only one of them will be initialized, based on the flag.
This is why it’s common to use a static constructor (even an empty one!) when implementing the singleton pattern.