>The issue is not that you can't implement the desired semantics, or
>even that you can't do it efficiently, but that the existing VM's are
>not amenable to implementing the desired semantics.
I'm becoming a bit bemused.
Earlier in this thread, the discussion seemed to be based on the idea that
the semantics as originally specified derived from a naive threading model
that was not practical in the real world, and that VMs had been implemented
that supported the practical reality. I am less convinced now as to the
validity of that position.
I would welcome a clear explanation of why VM developers should not be
expected to implement the documented behaviour. The mere fact that they got
it wrong in the past is not a reason.
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:46 EDT