Re: gl_MANYWARN_ALL_GCC() leads to many spurious warnings

[Samuel Bronson]

On Jul 9, 2012, at 11:23 AM, Samuel Bronson wrote:
> On Jul 8, 2012, at 10:15 PM, Paul Eggert wrote:
>
> > On 07/08/2012 07:10 AM, Samuel Bronson wrote:
> >
> > > I don't really know how to go about investigating that, short of  
> > > building GCC manually
> >
> > I wasn't thinking of anything that drastic -- just
> > compile with gcc -E instead of gcc -c and look
> > to see what happened to the pragmas, something like that.
> >
> > I just now tried with stock Ubuntu 12.04, i.e.,
> > gcc (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3, with
> > its bundled GNUstep libraries, and did not observe
> > either the regex.c -Wunused-macros warnings or the
> > -Wformat-security warnings.  So this is not merely
> > a GCC 4.7.1 issue, for whatever that's worth.  I
> > did see a passle of other warnings.  Perhaps the
> > warnings that you're observing come up only after
> > you fix the ones I saw?
>
> Hmm.  I'm beginning to suspect that the regex.c warnings are somehow  
> related to my use of ccache...

Well, I was right; it was related to my use of ccache.  But it's still  
a GCC bug; gcc -E isn't honoring #pragma diagnostic ignored "-Wunused- 
macros", and since ccache defaults to running the preprocessor first,  
then compiling its output (so it can skip the compilation if it  
already has results cached for that preprocessor output), this causes  
ccache to output the bogus warnings.  I reported this as <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53920 
>.

Presumably, the reason that you don't get the format warnings with  
GNUstep because, since only Apple's compilers have full support for  
__attribute__((format(NSString,...)), GNUstep probably doesn't use  
this attribute in its prototype for NSLog.  I submitted a bug report  
about them at <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53905>.