[concurrency-interest] The Atomic*FieldUpdater situation
forax at univ-mlv.fr
Sat Jul 14 17:22:52 EDT 2012
On 07/14/2012 10:01 PM, Stanimir Simeonoff wrote:
> On Sat, Jul 14, 2012 at 9:18 PM, Doug Lea <dl at cs.oswego.edu
> <mailto:dl at cs.oswego.edu>> wrote:
> On 07/14/12 14:01, David M. Lloyd wrote:
> What is the purpose of the access-time access check in the
> atomic field updater
> It is because there is no way to check that you haven't handed
> your Updater to some untrusted party, so the caller context must
> be checked on each use. I agree it is very annoying and slow.
> But how that's a problem - you create, you hand it over, exactly as
> adding an extra method public. I can't even see how that can be an
> argument. The checks should be during creation only, not during usage
> time - the calling class is trivial to infer. I wanted to voice that
> for very long time and used to always used to forget.
> Alternatively the SecurityManager should be checked and if not present
> a non-checking sub-class shall be returned. W/o SecurityManager it's
> possible to create modifiable java.lang.String not even touching
> Field/Method.setAccessible or Unsafe.
> That option doesn't require API changes.
I agree with you that creating several subclasses is better than the
current strategy which use several final field
that aren't considered as constant by the VM. The other solution is to
mark this field as trusted i.e will not be changed
by reflection, which is in my opinion a better solution.
The next question is how to guarantee that the update value as the right
type, otherwise you can
corrupt the memory by putting a String in an Integer field (and you
can't rely on generics for that).
> Since there would be a single subclass only instantiated in the JVM
> there would be no performance loss from multi-site invocations.
You don't care if there are several subclasses because you've put your
AtomicReferenceFieldUpdater in a static final field.
> Exactly that reason [pointless checks] has forced quite a few
> developers (and libs) going straight to sun.misc.Unsafe... or extend
> the AtomicXXX instead java.lang.Object (that at least is portable).
yes, far from ideal.
> Kind Regards
More information about the Concurrency-interest