Re: [PATCH] checkpatch: prefer = {} initializations to = {0}

[Christophe JAILLET]

Le 14/08/2021 à 16:52, Al Viro a écrit :
> On Sat, Aug 14, 2021 at 03:59:22PM +0200, Christophe JAILLET wrote:
>
> > > +# prefer = {}; to = {0};
> > > +		if ($line =~ /= \{ *0 *\}/) {
> > > +			WARN("ZERO_INITIALIZER",
> > > +			     "= {} is preferred over = {0}\n" . $herecurr);
>
> Sigh...  "is preferred over" by whom?  Use the active voice, would you?
>
> > [1] and [2] state that {} and {0} don't have the same effect. So if correct,
> > this is not only a matter of style.
> >
> > When testing with gcc 10.3.0, I arrived at the conclusion that both {} and
> > {0} HAVE the same behavior (i.e the whole structure and included structures
> > are completely zeroed) and I don't have a C standard to check what the rules
> > are.
> > gcc online doc didn't help me either.
>
> http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1256.pdf, but empty
> initializer-list is gccism anyway.
>
> Section 6.7.8 is the one to look through there.
>
> > Can someone provide some rational or compiler output that confirms that {}
> > and {0} are not the same?
>
> Easily: compare
> 	int x[] = {0};
> and
> 	int x[] = {};
>
> For more obscure example,
> 	int x = {0};
> is valid, if pointless, but
> 	int x = {};
> will be rejected even by gcc.
>
> Incidentally, do *NOT* assume that initializer will do anything with padding
> in a structure, no matter how you spell it.  Neither {} nor {0} nor explicit
> initializer for each member of struct do anything to the padding.  memset()
> does, but anything short of that leaves the padding contents unspecified.

Thanks for the explanations and exemples.

IIUC, code like [1] may leak some data (1 char) because of the layout of 
'struct atyclk' and we have the erroneous feeling that it is fully 
initialized, because of the "{ 0 }".

Correct?

CJ

[1]: 
https://elixir.bootlin.com/linux/v5.14-rc5/source/drivers/video/fbdev/aty/atyfb_base.c#L1859