Re: __STRICT_ANSI__ checks in headers

[Sam Ravnborg <sam@ravnborg.org>]

On Sun, Oct 01, 2006 at 10:34:56AM +0300, Ismail Donmez wrote:
> On Sunday 01 October 2006 08:20, Kyle Moffett wrote:
> > On Oct 01, 2006, at 00:53:43, Andrew Morton wrote:
> [...]
> > > Bisection shows that this patch causes these depmod warnings:
> > >
> > > WARNING: "snd_card_disconnect" [sound/usb/usx2y/snd-usb-usx2y.ko]
> > > has no CRC!
> > > [etc]
> > >
> > > I don't know why that would happen.
> > >
> > > From: Ismail Donmez <ismail@pardus.org.tr>
> > >
> > > __STRICT_ANSI__ usage in types.h header results in compile errors
> > > for some userspace packages[1] when used with gcc -ansi flag.  With
> > > the suggestion of Kyle Moffett I replace strict ansi checks with
> > > __extension__ to tell gcc not to error or warn on gcc extensions.
> > > Compile tested on x86 with 2.6.18.
> >
> > Best guess:  Depmod does some kind of funny type-based expansion and
> > hashing of the symbols which doesn't understand the "__extension__"
> > keyword.  Probably the simplest thing to do is to add "-
> > D__extension__=" to the depmod preprocessing flags.  Alternatively
> > you could teach depmod to completely ignore the __extension__ keyword
> > when it shows up in the sources, but the former seems like it would
> > be much simpler.
> >
> > Just thinking about it we probably also need to educate sparse about
> > __extension__ too.  Perhaps somebody could also add an sparse flag to
> > make it warn about nonportable constructs in exported header files.
> >
> > I'd submit a patch but my knowledge of kernel makefiles and depmod is
> > somewhere between zero and none, exclusive.
> 
> Thanks, I will have a look at it.
I assume you will same errors from the in-kernel modpost.
If you do not do so then there is some inconsistency between depmod
and modpost that ougth to be fixed.

I'm not up to the task atm - busy with my day job and travelling a lot

	Sam