Re: [announce] new tree: "fix all build warnings, on all configs"

[Ingo Molnar <mingo@elte.hu>]

* Jörn Engel <joern@logfs.org> wrote:

> On Mon, 20 October 2008 21:21:10 +0200, Ingo Molnar wrote:
> >  
> >  /* Alas, no aliases. Too much hassle with bringing module.h everywhere */
> >  #define fops_get(fops) \
> > -	(((fops) && try_module_get((fops)->owner) ? (fops) : NULL))
> > +	(((fops != NULL) && try_module_get((fops)->owner) ? (fops) : NULL))
> >  #define fops_put(fops) \
> > -	do { if (fops) module_put((fops)->owner); } while(0)
> > +	do { if (fops != NULL) module_put((fops)->owner); } while(0)
> 
> This, I would argue, makes the code worse.

Have a look at:

   $ git log -p --grep="NULL noise"

for example:

        for (i = 0; i < MAX_FEB_SIZE; i++)
-               if (tb->FEB[i] != 0)
+               if (tb->FEB[i] != NULL)
                        break;

so checking for != NULL is a valid way of testing a pointer's existence. 
The "if (tb->FEB[i])" is a valid shortcut for the same thing as well.

In this specific case the issue is that the 'fops' parameter can 
occasionally be a constant pointer (turning the test into always-true) 
so the compiler is at least minimally correct at asking the "are you 
sure you want this" question - which we answer in the affirmative via 
the explicit NULL check. But these are really nuances.

	Ingo