Re: multi-core software
Jon Harrop wrote:
I see no problem with mutable shared state.
In which case, Jon, you're in a small minority.
Patricia Shanahan wrote:
In my opinion, shared mutable state has a lot of problems. It is also
sometimes the best design for performance reasons.
As Dr. Jon pointed out upthread, one can write decent code with mutable shared
state. It is also true that mutable state presents a lot of problems -
potential problems, ones that can be solved, but not ones that can be solved
thoughtlessly. On the flip side, one can write a tremendous amount of
effective multi-threaded code involving shared mutable state with attention to
a few rules of thumb, like always synchronize access and don't use different
monitors to do so.
Unlike some environments (e.g., database management systems), Java's tools to
manage concurrency are explicit and low level. The programmer's job is to
make sure those tools are used correctly to avoid problems. As long as they
do that, then there is no special problem with shared mutable state.
There is, however, a cost. Certain things must happen slower when you share
mutable state, than when you share immutable state or don't share state.
Certain things must happen when you share mutable state, regardless of speed,
because without them your code doesn't work. For some reason, concurrent
programming is an area often not well understood by a significant percentage
of workaday programmers. When problems do arise, they tend to be
probabilistic in nature and vary widely with system characteristics like
attempted load.
So the meeting ground is, yes, concurrent mutable state can present problems
if not properly managed. Properly managing such is not necessarily a huge
burden, but it must be borne. When done properly, shared mutable state will
not present problems in production.
--
Lew