I have several "procedural" questions:
- What do we actually need to have by Friday?
- How much can we tweak whatever comes out on Friday, and how does
that change as we approach September 8? (I believe Sept 8 is the
deadline for beginning public review, or something like that.)
- Similarly, what do we need to have by September 8, and how much
can that be tweaked, and until when?
I hope that we don't have to have a full final draft of a model that
we expect to remain largely unchanged before being voted on (or acted
on in whatever way these things are handled). If we do, I think there
is little hope to have that.
Assuming we have a bit (but not much) more flexibility, it seems like
there are three sets of things to do (not counting the interrupt
exception handling issues being discussed in another thread of this
mailing list):
1. Fix up the Manson-Pugh specification by giving a full formalization
of the model (or at least one that someone other than Bill or Jeremy
feels confident of being able to fill in all the details, if it were
necessary), and an informal description that we can agree captures
our intuitive understanding of the model (regardless of whether we
like what the model does or does not guarantee).
2. Augment the Adve model to handle final fields, wait sets and
notification, etc. as in the Manson-Pugh chapter. I have no idea
how hard this will be, or even what is involved. (I still have a
few questions about Sarita's model, esp. the version with CnC-like
guarantees, but I'm fairly confident that, if necessary, I could
resolve these questions satisfactorily.)
3. Decide on what guarantees we want to provide, strong CnC-like
guarantees or weaker ones, and on what basis.
Obviously, only one of 1 and 2 need to be done, and both could be
replaced by yet another proposed formalism. But it seems for now
there is not agreement to abandon either direction.
Is there something else that needs to be done that I missed? In any
case, how will a decision be made about what finally gets proposed
on Friday (or whenever)?
Victor
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:49 EDT