Re: JavaMemoryModel: Isolation properties in the JMM

From: Miles Sabin (miles@milessabin.com)
Date: Sat Jan 31 2004 - 20:26:52 EST


Bill Pugh wrote,
> The following seems like a reasonable isolation property you would
> want in a memory model.
>
> Given a particular execution E of a program, create a partition of
> the threads and variables in the program so that if a thread accessed
> a variable in E, then the thread and variable are in the same
> partition. Note that we are lumping monitors, etc, in with variables.
>
> Given this partitioning, you should be able to explain the behavior
> in E of the threads in each partition, without having to examine the
> behavior or programs for the other threads.

I'm not necessarily disagreeing that this would be a desirable property,
but I need some convincing. It feels like you're trading on the obvious
reasonableness of the the much weaker property "For _every_ execution E
of a program ...".

With the weaker property, the partitions are independent of the actual
execution and my intutions are clear. But with the stronger property
the partitions are dependent on the actual execution, and I don't have
anything like as firm a grip on why I should accept the property as
reasonable.

Cheers,

Miles
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel



This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:57 EDT