Brent Fulgham <[EMAIL PROTECTED]> writes:
> Are we prepared to live with the (possibly minimal) increased risk
> of getting the memory/disk images out of synch? Or would it be better
> to mmap the data, change it, then munmap it? Can we be sure that this
> would flush the memory back to disk? It certainly seems inefficient
> at any rate...
That would not be so inefficient, but it wouldn't work. munmap does
not guarantee any synchronization about when the data is actually
written, and in Mach, it will normally not get written until later.
