On Aug 19, 2011, at 05:25, Kris Buelens wrote:

> Yes, that would seem better.  But, I no longer have the email(s?) I
> exchanged with you when I discovered this problem many years ago.  I think
> to remember that your reply was: "the STATE stage should not issue a Commit
> and commit everything".  And, that's right indeed.  Maybe my question was
> unclear/misleading.
> 
> So indeed, if you'd change STATE to use its own workunit, fewer people would
> get the surprises of the current behaviour.
> 
A sticky question.  There are strong arguments to be made on
both sides:

o The Heisenberg Principle shouldn't apply here.  It should
  be possible to make an observation of the state of a system
  without affecting the quantity being observed.  John Hartmann
  appears to favor this position.

Or:

o If a programmer queries the state of a file, it's reasonable
  that the result of that query will be used subsequently.  No
  process in another workunit, even from another VM, should be
  allowed to alter that state in the interim.

-- gil

Reply via email to