Re: [IA64] Add mapping table between irq and vector

Re: [IA64] Add mapping table between irq and vector

[Andrew Morton]

On Fri, 20 Jul 2007 19:59:17 GMT
Linux Kernel Mailing List <linux-kernel@vger.kernel.org> wrote:

> +	BUG_ON(bind_irq_vector(irq, vector));

It's not good practice to do assert(expression-with-side-effects).  Because
if someone wants to create a build which has all the assertions disabled,
the resulting binary will not work.

In the present implementation our BUG_ON(expression) will evaluate
`expression' even if CONFIG_BUG=n.  But that's totally lame and we are just
leaving optimisation opportunities on the floor.

Our objective _should_ be to make BUG_ON(expr) generate no code at all if
CONFIG_BUG=n.

So please, prefer to do

	if (bind_irq_vector(irq, vector))
		BUG();

Re: [IA64] Add mapping table between irq and vector

[Andrew Morton]

On Fri, 20 Jul 2007 13:17:28 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:

> On Fri, 20 Jul 2007 19:59:17 GMT
> Linux Kernel Mailing List <linux-kernel@vger.kernel.org> wrote:
> 
> > +	BUG_ON(bind_irq_vector(irq, vector));
> 
> It's not good practice to do assert(expression-with-side-effects).  Because
> if someone wants to create a build which has all the assertions disabled,
> the resulting binary will not work.
> 
> In the present implementation our BUG_ON(expression) will evaluate
> `expression' even if CONFIG_BUG=n.  But that's totally lame and we are just
> leaving optimisation opportunities on the floor.
> 
> Our objective _should_ be to make BUG_ON(expr) generate no code at all if
> CONFIG_BUG=n.
> 
> So please, prefer to do
> 
> 	if (bind_irq_vector(irq, vector))
> 		BUG();
> 

hm, now I think about it, our present implementation seems OK.  If you
have CONFIG_BUG=n then this:

	BUG_ON(foo < bar);

will generate no code and this:

	BUG_ON(some_function());

will still call some_function().

So I guess we're OK.