Re: multi-threading

[Steve Richter <stephenrichter@xxxxxxxxx>]

On Fri, Sep 4, 2009 at 6:17 PM, Jon Paris<jon.paris@xxxxxxxxxxxxxx> wrote:
> I've designed and built at least two systems like that and neither
> used or needed multi-threading.

In windows programming the default is to use a single thread.  99% of
the time when a window is unresponsive the culprit is user code in the
foreground thread which does not return to the message pumping loop.
Consider if we did not have an "input inhibited" state in the
interactive subsystem you would have to use threads ( just as they are
necessary in windows ).

> I agree with the earlier comment (I think from Scott) that you
> invariably end up with so much infrastructure to avoid screwing up
> that any potential advantages are lost.  Maybe in real-time control
> systems or indeed operating systems it would be worth the effort - but
> not in applications programming when there are simple alternatives.

what I like about .NET is that Microsoft software engineers are not
giving up on the problem of parallel programming. With BeginInvoke and
InvokeRequired the programmer can make sure that code that needs to
run on the foreground thread will run on that thread. The parallel
extensions library allows the programmer to run every iteration of a
do while loop in parallel.
http://blogs.msdn.com/pfxteam/archive/2009/08/12/9867246.aspx  .  The
latest is an implementation of  the concept of "software transactional
memory".  What I think STM is all about is you apply your transaction
to shared memory optimistically, without the traditional locking and
unlocking of mutexes. The STM framework keeps track of all the updates
the transaction is making to shared memory and runs a commit operation
at the end to apply the update for all threads to see.
http://msdn.microsoft.com/en-us/devlabs/ee334183.aspx