[Bug c/86410] New: Incorrect init_array ordering

[iripoll at disca dot upv.es]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86410

            Bug ID: 86410
           Summary: Incorrect init_array ordering
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: iripoll at disca dot upv.es
  Target Milestone: ---

Created attachment 44353
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44353&action=edit
Samplig the init, fin, constructor and atexit functions.

Hello,

 While testing the glibc patch of Florian (from Red Hat) to remove some
unnecessary code from the executable image, we have found what seems to be a
gcc issue.

 The problem is related to the way the .init_array and .fini_array  sections
are created. When not using any optimizations the resulting  initialization
sequence is different (and seems to be  incorrect). When using any form of
optimization (i.e. -O1 -O2 -O3  -Os) the init sequence is correct.

 The issue occurs to dynamic and static executables.

 Attached is a PoC of the problem. The same code results in a different
execution when compiled with different flags:

---- When compile with -O2 ----
gcc testing_init_arrays.c -o testing_init_arrays-O2
./testing_init_arrays-O2
preinit2
preinit1
constructor1
constructor2
init2
init1
Main begins
my_atexit2
my_atexit
fini1
fini2
destructor


---- When compiled without optimization ----
gcc testing_init_arrays.c -o testing_init_arrays-none
./testing_init_arrays-none
preinit1
preinit2
constructor1
init1
init2
constructor2
Main begins
my_atexit2
my_atexit
destructor
fini2
fini1

Regards,

Ismael and Hector.