Re: Removing -Wunsuffixed-float-constants, -Wdouble-promotion, -Wformat-zero-length

[Eric Blake]

On 11/29/2011 02:27 PM, Paul Eggert wrote:
> On 11/29/11 13:18, Jim Meyering wrote:
>> Have you found code that triggers a -Wformat-zero-length warning
>> yet that doesn't seem worth adjusting?
> 
> I haven't run into it one way or another, but all my instincts are
> against that diagnostic.  I suspect that the most-common way that it
> would happen would be something like this admittedly-contrived example:
> 
>    #if FEATURE_ENABLED
>     #define FEATURE_FORMAT "feature"
>    #else
>     #define FEATURE_FORMAT ""
>    #endif
>    ...
>    printf (buf, FEATURE_FORMAT);

I assume you meant sprintf (buf, FEATURE_FORMAT), or printf
(FEATURE_FORMAT)?

> 
> where the printf can be optimized away if FEATURE_ENABLED
> is zero, and the compiler is warning us about that.

Ah, but then I would use these more efficient idioms, which also avoids
the compiler warning:

puts (FEATURE_FORMAT) (replacing the printf case)
strcpy (buf, FEATURE_FORMAT) (replacing the sprintf case)

> This reminds me too much about bogus warnings that some compilers
> give for this:
> 
>    enum { N = FEATURE_ENABLED ? 1000 : 0 };
>    ...
>    for (i = 0; i < N; i++)
>      foo (i);
> 
> where the compiler proudly warns that it has optimized the loop
> away entirely, and did I really mean that?  (Yes I did! and I don't
> want to be warned about it! :-)

I agree that a noisy compiler telling you about dead code due to
optimizations is a pain, but until we actually encounter the noise, I
can't say whether the signal-to-noise ratio warrants removing the
warning by default.

-- 
Eric Blake   address@hidden    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature