Re: [PATCH][2.6.6-rc3] gcc-3.4.0 fixes

[Paul Wagland]

On Apr 29, 2004, at 11:30, Ihar 'Philips' Filipau wrote:

> Mikael Pettersson wrote:
> > This patch fixes three warnings from gcc-3.4.0 in 2.6.6-rc3:
> > - drivers/char/ftape/: use of cast-as-lvalue
> >  		if (get_unaligned((__u32*)ptr)) {
> > -			++(__u32*)ptr;
> > +			ptr += sizeof(__u32);
> >  		} else {
>
>   Can anyone explain what is the problem with this?
>   To me it seems pretty ligitimate code - why it was outlawed in gcc 
> 3.4?

http://es-sun2.fernuni-hagen.de/cgi-bin/info2html?(gcc)Subscripting

The ability to manipulate a cast was always a gcc extension, whether by 
design or feature I am not sure. The problem is that it breaks C++ 
templates in a bad way. Here is where the extension is documented for 
GCC 3.3.3

http://gcc.gnu.org/onlinedocs/gcc-3.3.3/gcc/Lvalues.html

This extension is now obsoleted.

>   Previous code was agnostic to type of ptr, but you code presume ptr 
> being char pointer (to effectively increment by 4 bytes).

Yes, and the previous code assumed that that ptr was not a pointer to a 
64 bit value, since dereferencing it on most platforms would then 
explode. I bet that both assumptions are just as safe ;-)  However if 
you really wanted to be safe then you could do

ptr += sizeof (__32) / sizeof (*ptr);

Again, assuming that we are not using a greater then 4 byte ptr*, since 
otherwise it would not increment... But personally I think that this is 
not required.

Cheers,
Paul

Attachment: PGP.sig
Description: This is a digitally signed message part