Re: Remove warnings from RTEMS crt0.c

[Ralf Corsepius <ralf dot corsepius at rtems dot org>]

On 08/03/2011 05:11 PM, Joel Sherrill wrote:
> On 08/03/2011 09:59 AM, Ralf Corsepius wrote:
> > On 08/03/2011 03:10 PM, Joel Sherrill wrote:
> > > On 08/03/2011 12:38 AM, Ralf Corsepius wrote:
> > > > On 08/02/2011 08:13 PM, Joel Sherrill wrote:
> > > > > Hi,
> > > > >
> > > > > The RTEMS crt0.c is just a stub so autoconf
> > > > > will find methods. It uses a macro to
> > > > > generate stub bodies for routines we want
> > > > > autoconf to find that are not in libc.
> > > > > The macro that generates the stubs does
> > > > > not generate a return statement with properly
> > > > > typed argument. So when you do a -Wall, you
> > > > > get lots of warnings. This just eliminates
> > > > > the warnings.
> > > > >
> > > > > OK to commit?
> > > > Not for now. I'd firstly want to reproduce the warnings you are
> > > > mentioning and check why I don't see them.
> > > I sent a private email to you on 19 July with this and an
> > > explanation.
> > I saw it, but was busy otherwise and did not have any time to look into
> > them.
> >
> > > I was trying to compile RTEMS with clang.
> > As you know, I consider these tries to be broken and invalid, because
> > you are using an different toolchain for a different target.
> I get the same warning from gcc and clang.
I repeat ... your clang tries are broken and invalid.

xgcc warnings are worth to be looked into and to be checked for it they 
are serious.

> > > FWIW You will also see a number of other warnings.
> > FYI: I see ZERO warnings when bootstrapping GCC+newlib.
>
> I repeat... newlib for most targets including *-rtems* is NOT
> NOT NOT built with warnings enabled.  That is why most people
> compiling newlib are NOT seeing warnings.
You see all warning an xgcc emits by default - This probably had been 
the case ever since newlib exists.

> When clang flagged
> this, I double-checked that gcc would warn for the same thing
> if presented with -Wall.  You have to add this to the
> *-rtems* stanza in configure.host.
>
>         newlib_cflags="${newlib_cflags} -Wall"
Blindly adding -Wall contradicts portability. It won't harm rtems, 
because RTEMS hardly will ever be compilable by any other compiler but 
GCC - nevertheless it's a hack.

Apart of this, IMO, the right place to specify -Wall would be gcc's 
toplevel configure, such that _all_ target compiled libraries will 
receive such flags, not only newlib (which is just one amongst many 
target-libs).

Ralf