Re: [avr-gcc-list] OT Generic C question

[Richard Urwin]

On Tuesday 20 Sep 2005 15:04, Dave Hansen wrote:
> From: "David Brown" <address@hidden>
>
> >----- Original Message -----
> >From: "Trampas" <address@hidden>
> >
> > > I was helping a friend debug some code, he is new to C,  using
> > > the Keil version of GCC for ARM. Anyway I found the following:
> > >
> > > int i;
> > >
> > > i=0;
> > > i=i++;
> > > //i was still zero that
>
> [...]
>
> >I'd agree with you that i should be 1 after "i = i++", despite the
> >sillyness
>
> Actually, I think the naive interpretation would be that i should be
> zero after the above is executed:
>
> The statement implies three operations:
>
>    1) read the variable on the RHS (i) and save the value
>    2) increment the variable on the RHS (i).
>    3) store the value preserved in step 1 (0) in the variable on the
> LHS (i)
>
> The "logical" order of the operations is given above.  In reality,
> the operations can occur in just about any order.  The only
> guarantees the C standard makes about the order of execution involves
> a concept called "sequence points."  Sequence point occur in several
> places, but the most common is at the end of each full expression (as
> a first estimation, you can think of it as being at each semicolon).
>
> The standard says that the value of a variable should only be
> modified once between sequence points.  The expression "i = i++;"
> attempts to modify i twice: once with the ++ operator and once with
> the = operator.  Modifying a variable twice between sequence points
> is undefined behavior -- there is no "correct" answer.  After the
> expression is executed, i could be 0, 1, 42, or the system might
> format your hard drive.  All of the above and many other results are
> "legal" as far as the standard is concerned.

No, you're wrong.
There is a sequence point immediately before the assignment.
So, far from being a compiler bug, this is absolutely correct behaviour. 
Anything else would be a bug.

1) The value of i is taken (0)
2) i is incremented (to 1)
---- sequence point---
3) i is set to the value taken (0)

-- 
Richard Urwin