ansaurus

Question

Are private classes being sealed at compilation?

Answer 1

+1 A:

I didn't check, but I suppose that it will not do that. In any way, the JIT can determine whether a class can have descendants or not, so that I don't expect any difference at runtime.

Lucero 2010-03-25 19:48:13

The JIT can determine if a class has descendents but I don't believe that it does outside the presence of the sealed bit on a type. I don't have any references to back me up on this but I'm fairly certain on this point.

JaredPar 2010-03-25 19:50:07

@JaredPar, as far as I know many Java VMs have been doing this for a long time (which is really necessary due to the default-virtual-method - http://www.ibm.com/developerworks/library/j-jtp12214/ ), and I'm pretty sure that MS does do such optimizations if they have been seen as helping performance.

Lucero 2010-03-25 19:57:16

@Lucero the JIT definitely does for explititly sealed classes. But there is only a very small subset of cases where it can do so for non-sealed classes. I'm not sure if that micro-optimization would be worth it. But like I said, I can't find any docs one way or the other.

JaredPar 2010-03-25 19:58:39

@JaredPar, why do you think that this subset is so small? The CLR does know at any time whether a type has descendants in or not. If an assembly is loaded later on, it could throw away or fix the already compiled methods which involve those calls. Also, there are cases such as the one here where it is clear that there will be no inheritance. Again as far as In know, in Java they go as far as inlining virtual methods when those are seen as implicitly "final" and qualify for inlining like the same nonvirtual method would.

Lucero 2010-03-25 20:07:54

@Lucero, the CLR only knows the type currently doesn't have any descendants. At any point in time a new assembly could enter the `AppDomain` which is a descendant of the class. Only if the type is sealed or falls into another very small use case can the CLR know the type has no descendents.

JaredPar 2010-03-25 20:23:15

If it's a private nested class, then it cannot have any descendants in other assemblies - visibility requirements apply to correct IL, even if it's otherwise unverifiable. So, in practice, the JIT can make a valid assumption that a private nested class that doesn't have any classes derived from it in the same assembly is effectively sealed.

Pavel Minaev 2010-05-10 23:29:32

Answer 2

+6 A:

The compiler will not automatically mark this type as sealed.

True in this very specific scenario there is no real value in leaving the class as unsealed. However determining that you are in this scenario is not always so easy. You have to consider the following

The type is private
Must consider the presence of partial classes
Other private nested types could inherit.

These are not impossible to calculate but it's not trivial either. It's much cheaper to ask the user to just seal the type themselves

JaredPar 2010-03-25 19:49:04

ansaurus

tags:

views:

answers:

Are private classes being sealed at compilation?

related questions