[cfe-commits] [PATCH] [libcxx] cleanup a few compile warnings emitted by GCC

[Joerg Sonnenberger]

On Thu, Dec 27, 2012 at 04:05:37PM -0800, Chandler Carruth wrote:
> On Thu, Dec 27, 2012 at 4:01 PM, Joerg Sonnenberger <joerg at britannica.bec.de
> > wrote:
> 
> > On Thu, Dec 27, 2012 at 01:49:06PM -0500, Howard Hinnant wrote:
> > > > @@ -4583,7 +4584,7 @@
> > > >  string
> > > >  __time_get_storage<char>::__analyze(char fmt, const ctype<char>& ct)
> > > >  {
> > > > -    tm t = {0};
> > > > +    tm t = {0,0,0,0,0,0,0,0,0,0,0};
> > > >      t.tm_sec = 59;
> > > >      t.tm_min = 55;
> > > >      t.tm_hour = 23;
> > > > @@ -4729,7 +4730,7 @@
> > > >  wstring
> > > >  __time_get_storage<wchar_t>::__analyze(char fmt, const
> > ctype<wchar_t>& ct)
> > > >  {
> > > > -    tm t = {0};
> > > > +    tm t = {0,0,0,0,0,0,0,0,0,0,0};
> > > >      t.tm_sec = 59;
> > > >      t.tm_min = 55;
> > > >      t.tm_hour = 23;
> > >
> > > Rejected.  tm contains *at least* 9 int data members.  The portable and
> > > concise way to zero initialize this struct is with the single {0}.
> >
> > What about using memset as alternative?
> 
> 
> Why? The compiler warning from GCC is simply wrong. The other members are
> in fact initialized. I see no reason to use a more verbose construct which
> requires a declaration of an external function merely to avoid a patently
> incorrect warning.

The compiler warning from GCC is for the (common) case that a field is
added later and so the initializer depends only on the default value of
the field. That warning is helpful for plain C, but not so much for C++
data structures, where constructures serve the purpose of preserving API
compatibility.

Joerg