Hi Bill,
Isn't Test 4 the standard out-of-thin-air example that everyone agrees
must not be allowed (allows type-safety violations)? I'm ambivalent
about Test 5, and think Tests 1, 2 and 3 should be allowed. (Why
might the answers to Tests 2 and 3 be different?) But in any case, I
trust your judgment on these more than my own.
My opinion, however, is that if we consider incorrectly synchronized
code to be a mistake, then we should guarantee as little as possible
in those circumstances--that is, we should guarantee what we must to
ensure whatever safety properties (type-safety and security) are
deemed necessary, but we should *not* provide such guarantees for
debugging purposes. For that, we should provide better tools, which
directly address the issue, and will be much more effective than
providing stronger guarantees. There is a wealth of research on
data-race detection, for example, with several tools that should be
relatively easy to adapt for Java.
Victor
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:48 EDT