Re: new warning with 2.19

[Joel Sherrill <joel dot sherrill at oarcorp dot com>]

Tristan Gingold wrote:
> On Dec 4, 2008, at 9:32 AM, Alan Modra wrote:
>
>   
> > On Thu, Dec 04, 2008 at 08:39:30AM +0100, Ralf Corsepius wrote:
> >     
> > > * Is the call to ld triggering the warning correct?
> > >       
> > Well it is rather unusual to specify -e with -r.  It really only makes
> > sense if your system runs relocatable executables.
> >     
>
> Unless you also want to use --gc-sections.  In this case, -e SYM  
> specifies the root symbol.
>   
We are extending the gcc specs with this rule:

*link:
%{!qrtems: %(old_link)} %{qrtems: -Qy -dp -Bstatic -N -e _start}

And apparently gcc has always passed that through for -r as
well but ld did not complain.

It is benign.  It is a behaviour that changed.  And there is
probably room for everyone involved to fix something:

+ gcc probably shouldn't pass the same *spec for -r
   and final links.
+ ld shouldn't complain when the entry argument
   doesn't matter. 
+ rtems bsp's can fix their specs. (example patch
   attached).

I can fix the RTEMS bsp_specs fairly quickly so 1 of the
three can be killed.  Ralf.. are you OK with the bsp_specs
change?  You shouldn't be specifying the start address unless
you are using the start file.

--joel
Index: bsp_specs
===================================================================
RCS file: /usr1/CVS/rtems/c/src/lib/libbsp/arm/csb336/bsp_specs,v
retrieving revision 1.6
diff -u -r1.6 bsp_specs
--- bsp_specs	12 Apr 2007 03:03:48 -0000	1.6
+++ bsp_specs	4 Dec 2008 14:31:56 -0000
@@ -4,10 +4,10 @@
 
 *startfile:
 %{!qrtems: %(old_startfile)} \
-%{!nostdlib: %{qrtems: start.o%s crti.o%s crtbegin.o%s}}
+%{!nostdlib: %{qrtems: start.o%s crti.o%s crtbegin.o%s -e _start}}
 
 *link:
-%{!qrtems: %(old_link)} %{qrtems: -Qy -dp -Bstatic -N -e _start}
+%{!qrtems: %(old_link)} %{qrtems: -Qy -dp -Bstatic -N}
 
 *endfile:
 %{!qrtems: *(old_endfiles)} %{qrtems: crtend.o%s crtn.o%s }