[Openmcl-devel] Is setf atomic?

[Gary Byers]

On Wed, 7 Nov 2007, Pascal Costanza wrote:

> OK, I'll rephrase my question:
>
> Assume there is one thread executing the following endless loop on a shared 
> global variable:
>
> (defvar *shared-var* 42)
>
> ...
>  (loop (setf *shared-var* 84)
>        (setf *shared-var* 42))
> ...
>
> Further assume that this is the only thread that ever assigns to *shared-var* 
> - all other threads either don't touch it, or only read from it.
>
> Will the only values ever read from this variable be 42 and 84, or is it 
> possible that sometimes some threads read something else (that is, garbage) 
> from that variable?

Yes, the value will always be either 84 or 42 (barring hardware failure or
something like that); it will never have some of 42's bits and some of
84's.

I don't believe that any location that a lisp object could be stored in
could cross a page or cache-line boundary, but I'm fairly sure that this
would hold true in those cases as well.  We may have a need to write
across cache-line/page boundaries in the IA32 port, but the only cases
where I can imagine this happening are like a tree falling in the forest
(other threads are suspended by the GC or the object being written
to is being initialized and isn't visible to other threads yet.)


>
> To generalize, are there datatypes other than integers for which it could be 
> that other threads read garbage from such a variable?

Except for the case of trying to store 64-bit values into a foreign
data structure on a 32-bit machine, no.

>
> Thanks for your patience.

Likewise; sorry to have misunderstood.

>
>
> Pascal
>
> -- 
> Pascal Costanza, mailto:pc at p-cos.net, http://p-cos.net
> Vrije Universiteit Brussel, Programming Technology Lab
> Pleinlaan 2, B-1050 Brussel, Belgium
>
>
>
>