Reordering of function calls - compiler bug or allowed by standard?

[John Fine]

Jens Kilian wrote:
> /* model internal state of address register */
> int N;
>   
Most of your problem can be dealt with by making N volatile and making 
RD be or call a function (it could be inline so there is no extra overhead).

>   if (RD(42) == RD(23))
>   
But that still wouldn't control which of RD(42) or RD(23) happens 
first.  Did you expect to control that?

The compiler can't resequence volatile accesses across "sequence points".

With RD as a macro, you have:

if ( (writeit(42), readit()) == (writeit(23), readit()) )

I'm think the comma is a sequence point between each writeit and the 
corresponding readit, but I don't think you have any sequence point 
between either readit an the non corresponding writeit.  If RD were or 
called an inline function, there would be plenty of sequence points 
between either readit an the non corresponding writeit.