[gimple] assignments to volatile

[Michael Matz]

Hi,

On Wed, 23 Jun 2010, Mike Stump wrote:

> >  (I think we should define it to be no access, unlike 'x = vobj;' even if x is unused later.).
> 
> Again, it is wrong to abuse the latitude in the standard like this.  If 
> the standard says there is a read access on the abstract machine,

The point of this whole thread is that the standard doesn't say there's a 
read access for "vobj;" (full expression).  I tried to argue in this 
thread that there should be a read and a write access of vobj in
 "x = vobj = y;" .  From that position some surprising consequences would 
follow, which I tried to either argue irrelevant or wrongly implied.

I'm not sure what exactly your position on this whole matter is, except 
for "let's do exactly what the standard says".  I'm not at all disagreeing 
with the position, but this very thread shows quite clearly that there's 
disagreement about _what_ exactly the standard says.  So you should rather 
try to explain what you think the standard implies (I myself don't know if 
you're for or against a re-read).


Ciao,
Michael.