Re: [RFC] Default ld -r should not combine .init_array sections

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Thu, Feb 25, 2010 at 8:05 AM, Jie Zhang <jie@codesourcery.com> wrote:
> We found this issue on arm-none-eabi target. The default ld -r combines
> .init_array.* sections. Such that priority information recorded in section
> name will be lost. When the object file from the output of "ld -r" is linked
> into the final executable, the static constructors will do initialization in
> wrong order.
>
> For example:
>
> $ cat a.cpp
> #include <stdio.h>
>
> class A
> {
> public:
> ?A() {
> ? ?printf ("A\n");
> ?}
> };
>
> A a __attribute__ ((init_priority (1000)));
>
> $ cat b.cpp
> #include <stdio.h>
>
> class B
> {
> public:
> ?B() {
> ? ?printf ("B\n");
> ?}
> };
>
> B b __attribute__ ((init_priority (2000)));
>
> int main ()
> {
> ?return 0;
> }
>
> #include <stdio.h>
>
> class C
> {
> public:
> ?C() {
> ? ?printf ("C\n");
> ?}
> };
>
> C c __attribute__ ((init_priority (3000)));
>
> $ arm-none-eabi-g++ -o m a.cpp b.cpp c.cpp
> $ arm-none-eabi-run m
> A
> B
> C
>
> But
>
> $ arm-none-eabi-g++ -c a.cpp b.cpp c.cpp
> $ arm-none-eabi-ld -r -o ac.o a.o c.o
> $ arm-none-eabi-g++ -o m ac.o b.o
> $ arm-none-eabi-run m
> B
> A
> C
>
> With the attach patch, we can get the expected result. This patch also do
> the same thing for .fini_array sections. Any comments?
>
>

Please include a generic testcase.

Thanks.

-- 
H.J.